[go: up one dir, main page]

WO2024158259A1 - Method and system for predicting thyroid dysfunction of subject - Google Patents

Method and system for predicting thyroid dysfunction of subject Download PDF

Info

Publication number
WO2024158259A1
WO2024158259A1 PCT/KR2024/001308 KR2024001308W WO2024158259A1 WO 2024158259 A1 WO2024158259 A1 WO 2024158259A1 KR 2024001308 W KR2024001308 W KR 2024001308W WO 2024158259 A1 WO2024158259 A1 WO 2024158259A1
Authority
WO
WIPO (PCT)
Prior art keywords
heart rate
day
section
concentration
date
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/KR2024/001308
Other languages
French (fr)
Korean (ko)
Inventor
신규보
김종찬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Thyroscope Inc
Original Assignee
Thyroscope Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Thyroscope Inc filed Critical Thyroscope Inc
Publication of WO2024158259A1 publication Critical patent/WO2024158259A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/02Detecting, measuring or recording for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
    • A61B5/024Measuring pulse rate or heart rate
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders

Definitions

  • This application relates to a method of analyzing a patient's heart rate and predicting thyroid dysfunction based on the analysis results, and a device and system for performing the method.
  • the thyroid gland is an endocrine organ located in the front of the neck. Its job is to produce thyroid hormones by receiving signals from thyroid-stimulating hormone secreted by the pituitary gland in the brain.
  • Thyroid hormones produced by the thyroid gland play a role in regulating our body's metabolic rate.
  • these thyroid hormones are secreted at a higher than normal level, the body's metabolism progresses abnormally quickly, causing symptoms such as feeling hotter than usual or feeling heart palpitations.
  • symptoms such as feeling extremely cold, fatigue and helplessness, and a slow pulse appear.
  • thyroid gland When the function of the thyroid gland is normal, the secretion of thyroid hormones occurs without a problem, but when there is an abnormality in the function of the thyroid gland, the secretion of thyroid hormones decreases or increases.
  • a series of diseases caused by decreased secretion of thyroid hormones are caused by the thyroid gland. It is called hypothyroidism, and a series of diseases caused by increased secretion of thyroid hormones are called hyperthyroidism. Additionally, these two diseases are called thyroid dysfunction.
  • the only diagnostic method to confirm abnormalities in thyroid function is to check the concentration of thyroid hormones in the patient's blood and determine whether the concentration of thyroid hormones is within the normal range, high level, or low level using clinical pathology. The only way is to check.
  • hyperthyroidism and hypothyroidism there are various methods to treat hyperthyroidism and hypothyroidism, but the most widely used treatment method is drug treatment. Most patients suffering from hyperthyroidism and patients suffering from hypothyroidism regain normal levels of thyroid hormone production and secretion function by taking medication for a certain period of time. However, hyperthyroidism and hypothyroidism are diseases that do not have a high cure rate and have a very high recurrence rate, so there is a need to continuously monitor whether hyperthyroidism or hypothyroidism recurs.
  • the only diagnostic method to check the function of the thyroid gland is to check the concentration of hormones through blood sampling, so the patient must visit the hospital periodically to monitor whether the thyroid function is abnormal. There is.
  • the problem to be solved by the contents disclosed by this application is to predict whether thyroid function is abnormal based on heart rate information obtained using a personal electronic device that can be used by the general public rather than a professional medical diagnostic device. It provides a predictive model that can be predicted.
  • Another problem to be solved by the contents disclosed by the present application is to provide a method of learning a prediction model that can predict whether thyroid function is abnormal based on the above-described heart rate information.
  • Another problem to be solved by the contents disclosed by this application is to use a model to predict whether thyroid function is abnormal based on the above-mentioned heart rate information, so that the general public can continuously monitor the thyroid gland without the help of a doctor and without visiting the hospital directly.
  • the purpose is to provide a method and system for monitoring malfunctions.
  • a method for predicting dysthyroidism for a subject performed by one or more processors includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; i) the relationship between the first heart rate information determined for the target date and the second heart rate information determined for the reference date, and ii) the thyroid hormone concentration corresponding to the target date by inputting the reference day hormone concentration corresponding to the reference date into the first prediction model Obtaining a tentative prediction result for dysfunction - the first heart rate information and the second heart rate information are heart rate information about the subject; i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the tentative prediction result is input into a second prediction model to produce a result for the predicted hormone concentration corresponding to the target date.
  • the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iv) the results of the predicted hormone concentration corresponding to the target date are input into a third prediction model to determine thyroid dysfunction. It may include obtaining a prediction result for.
  • a result for the tentative predicted hormone concentration corresponding to the target date is further input, and the tentative predicted hormone concentration is: i) the It can be obtained by inputting the relationship between the first heart rate information and the second heart rate information and ii) the hormone concentration on the reference day into the fourth prediction model.
  • the result of the predicted hormone concentration may be a predicted thyroid hormone concentration value corresponding to the target date.
  • the result for the predicted hormone concentration may correspond to the relationship between the thyroid hormone concentration on the target day and the hormone concentration on the reference day.
  • the relationship is: 1) the amount of change in the average of section heart rates of the target day with respect to the reference date, 2) the rate of change of the average of section heart rates of the target day with respect to the reference date , 3) the amount of change in the standard deviation of the target date with respect to the reference date, 4) the amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) the above It may be one or a combination thereof selected from the group consisting of the amount of change in kurtosis of the target date with respect to the reference date and 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date.
  • the predicted hormone concentration is 1) the concentration of the subject's thyroid stimulating hormone (TSH), 2) the subject's tetraiodothyronine (T4) ) concentration, 3) the concentration of free T4 (free T4) in the serum of the subject, 4) the concentration of triiodothyronine (T3) in the subject, 5) the concentration of free T3 (free T4) in the serum of the subject It may correspond to one or a combination thereof selected from the group consisting of the concentration of T3) and 6) the concentration of Thyrotropin-Releasing Hormone (TRH) of the subject.
  • TSH thyroid stimulating hormone
  • T4 tetraiodothyronine
  • the first prediction model includes 1) the change in the average of the heart rates of the target day on the target day with respect to the reference date, 2) the heart rates of the section on the target day with respect to the reference date.
  • the rate of change in the average 3) the amount of change in the standard deviation of the target date with respect to the reference date, 4) the amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) The amount of change in kurtosis of the target date with respect to the reference date, 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date, 8) Thyroid stimulation of the subject corresponding to the reference date Concentration of thyroid stimulating hormone (TSH), 9) Concentration of tetraiodothyronine (T4) in the subject corresponding to the reference date, 10) Free T4 (free) in the serum of the subject corresponding to the
  • the combination value may not be input to the second prediction model, and the combination value may be further input to the third prediction model.
  • the standard deviation, skewness, kurtosis, and difference between the target date and the reference date of the section heart rates corresponding to the target date (day gap) ) or a combination thereof may be further entered.
  • the section heart rates corresponding to the target date are all resting heart rates corresponding to a predetermined period based on the target date
  • the section corresponding to the reference date The heart rates may be all resting heart rates corresponding to the predetermined period based on the reference date.
  • the first prediction model is a model that classifies hyperthyroidism or not hyperthyroidism
  • the third prediction model is a model that classifies hyperthyroidism or not hyperthyroidism. It could be a model.
  • the second prediction model may be a model learned using the heart rate information and hormone levels of patients confirmed to have hyperthyroidism as a learning data set.
  • the first prediction model is a model that classifies hyperthyroidism, normal function, or hypothyroidism
  • the third prediction model is a model that classifies hyperthyroidism, normal function, or hypothyroidism. It may be a classification model.
  • the first prediction model is a model learned using as a learning data set the heart rate information of patients confirmed to have hyperthyroidism and whether there is an abnormality in thyroid function at that time
  • the first prediction model is a model learned with hypothyroidism
  • the fifth prediction model learned using the heart rate information of the patients identified as a learning data set and the presence or absence of thyroid function abnormalities at that time is i) the relationship between the first heart rate information and the second heart rate information, and ii) corresponding to the reference date.
  • the hormone concentration on the reference day is input to obtain a tentative prediction result for thyroid dysfunction
  • the second prediction model can further receive the tentative prediction result output from the fifth prediction model to obtain a result for the predicted hormone concentration. there is.
  • a method for predicting dysthyroidism for a subject performed by one or more processors includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; i) the relationship between the first heart rate information regarding the target date heart rate determined for the target date and the second heart rate information regarding the reference date heart rate determined for the reference date, and ii) the reference day hormone concentration corresponding to the reference date to the first prediction model
  • Input to obtain a result of the predicted hormone concentration corresponding to the target date - the first heart rate information and the second heart rate information are heart rate information related to the target; i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the results of the predicted hormone concentration corresponding to the target date are input into the second prediction model to determine thyroid dysfunction.
  • Obtaining a prediction result may include.
  • a prediction model may be provided.
  • a method of learning a prediction model that can predict whether the thyroid function is abnormal based on the above-described heart rate information is provided.
  • the general public can continuously detect abnormal thyroid function without the help of a doctor and without visiting the hospital directly, using a model that predicts abnormality of thyroid function based on the above-mentioned heart rate information.
  • a method and system for monitoring is provided.
  • FIG. 1 is a diagram illustrating a system for predicting thyroid dysfunction according to an embodiment disclosed by the present application.
  • FIG. 2 is a block diagram of a heart rate measurement device disclosed by the present application.
  • Figure 3 is a block diagram of a user terminal disclosed by the present application.
  • FIG. 4 is a block diagram of the server disclosed by this application.
  • Figure 5 is a flow chart to explain the method for predicting thyroid dysfunction disclosed by the present application.
  • Figure 6 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.
  • Figure 7 is a block diagram for explaining the thyroid dysfunction prediction module.
  • FIGS 8 and 9 are block diagrams for illustrating a thyroid function abnormality prediction module including a thyroid function classification model and/or thyroid hormone concentration prediction model specialized for hyperactivity, normal and/or hypothyroidism by selecting clinical data during learning. am.
  • Figures 10 and 11 are diagrams for explaining a method for predicting thyroid dysfunction.
  • Figure 12 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.
  • a method for predicting thyroid dysfunction for a subject includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; Obtaining zone heart rates corresponding to the determined target date; For the object, a first preprocessing result is obtained for the zone heart rates corresponding to the acquired target date - At this time, the first preprocessing result is the average and distribution of the zone heart rates corresponding to the target day.
  • the second preprocessing result is at least one information about the average and distribution of the section heart rates corresponding to the reference date.
  • the prediction result for thyroid dysfunction may be a result of processing input values including the difference and at least one hormone concentration corresponding to the reference date by a thyroid dysfunction prediction model.
  • At least one indicator regarding the distribution of zone heart rates corresponding to the target date includes standard deviation, skewness, and kurtosis of zone heart rates corresponding to the target date
  • At least one indicator regarding the distribution of section heart rates corresponding to the reference date may include standard deviation, skewness, and kurtosis of section heart rates corresponding to the reference date.
  • the difference between the first preprocessing result and the second preprocessing result is 1) the amount of change in the average of the heart rates of the intervals of the target day with respect to the reference date, 2) the reference date , 3) the change in the standard deviation of the target day with respect to the reference date, 4) the change in the relative standard deviation of the target date with respect to the reference date, 5) with respect to the reference date.
  • the concentration of at least one hormone corresponding to the reference date is 1) the concentration of thyroid stimulating hormone (TSH), 2) tetraiodothyronine (T4) concentration, 3) concentration of free T4 (free T4) in serum, 4) concentration of triiodothyronine (T3), 5) concentration of free T3 (free T3) in serum, and 6) thyroid-stimulating hormone.
  • TSH thyroid stimulating hormone
  • T4 tetraiodothyronine
  • T4 free T4
  • T3 triiodothyronine
  • T3 free T3
  • T3 free T3
  • the method for predicting thyroid dysfunction for the subject includes: 1) the amount of change in the average of the heart rates of the section of the subject day with respect to the reference date, 2) the subject with respect to the reference date The rate of change in the average of the daily heart rates, 3) the amount of change in the standard deviation of the target day with respect to the reference date, 4) the amount of change in the relative standard deviation of the target day with respect to the reference date, 5) the amount of change in the target day with respect to the reference date Amount of change in skewness, 6) Amount of change in kurtosis of the target date with respect to the reference date, 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date, 8) Corresponding to the reference date concentration of thyroid stimulating hormone (TSH) in the subject, 9) concentration of tetraiodothyronine (T4) in the subject corresponding to the reference date, 10) serum of the subject corresponding to the reference date Concentr
  • the method for predicting thyroid dysfunction for the subject may further include obtaining a difference (day gap) between the subject date and the reference date.
  • Input values including hormone concentration, the difference, and the difference between the target date and the reference date may be the result of being processed by a thyroid dysfunction prediction model.
  • the section heart rates corresponding to the target date are all resting heart rates corresponding to a predetermined period based on the target date
  • the section corresponding to the reference date The heart rates may be all resting heart rates corresponding to the predetermined period based on the reference date.
  • the predetermined period may be any one selected from 8, 9, 10, 11, 12, 13, 14, 15, 16, and 17 days.
  • the dysthyroidism prediction model may include a hyperthyroidism prediction model and a hypothyroidism prediction model.
  • the prediction result for dysthyroidism is the first prediction result processed by the hyperthyroidism prediction model and the input values including the difference and at least one hormone concentration corresponding to the reference date and the reference date.
  • Input values including at least one hormone concentration and the difference may be determined by considering the second prediction result processed by the hypothyroidism prediction model.
  • heart rate refers to the number of heart beats per unit time. For example, heart rate may be measured in beats per minute (bpm). Meanwhile, in this specification, unless specifically mentioned or the term is used for the purpose of general scientific explanation, “heart rate” is considered to correspond to or be linked to the estimation time point at which the heart rate was measured.
  • rest period refers to a period in which a person's physical movement is below a certain level.
  • rest period refers to a period in which the subject is awake and does not move, but in this specification, the term “rest period” also refers to a period in which the subject is sleeping.
  • resting heart rates refers to heart rates that are within a person's resting period when the heart rate is measured.
  • the resting period generally includes a person's sleeping time, and ii) considering people's sleeping habits, there are many people who start sleeping before 12:00:00 a.m.
  • “resting heart rates corresponding to day Corresponding resting heart rates may be included.
  • “resting heart rates corresponding to day” the resting heart rates corresponding to the time from the time you start falling asleep on Day X to 11:59:59 PM on Day It may be included in “resting heart rates.”
  • period heart rates means all resting heart rates corresponding to a period of two or more days.
  • period resting heart rates may be used interchangeably with “period resting heart rates.”
  • the term “period average heart rate” refers to the average value of resting heart rates (i.e., zone heart rates) measured over a period of two or more days. For example, the average heart rate over a 5-day period is the resting heart rates corresponding to the first day, the resting heart rates corresponding to the second day, and the resting heart rates corresponding to the third day. It means the average value of heart rates, resting heart rates corresponding to the fourth day, and resting heart rates corresponding to the fifth day. In this specification, the term “average of period heart rates” may be used with the same meaning as “average period heart rate.”
  • Section harmonic average heart rate ( period harmonic average heart rate)
  • the term “period harmonic average heart rate” refers to the harmonic average value of resting heart rates (i.e., interval heart rates) measured over a period of 2 days or more.
  • the 5-day interval harmonic average heart rate is the resting heart rates corresponding to the first day, the resting heart rates corresponding to the second day, and the resting heart rates corresponding to the third day. It means the harmonic average value of resting heart rates, resting heart rates corresponding to the fourth day, and resting heart rates corresponding to the fifth day.
  • the term “harmonic average of period heart rates” may be used with the same meaning as “period harmonic average heart rate.”
  • thyroid hormone refers to thyroxine (T4) or tetraiodothyronine, which are hormones secreted by the thyroid gland, and serum free T4 (T4 in the form of T4 free in serum).
  • free T4 as well as triiodothyronine (T3), which is converted to remove iodine from T4, serum free T3 (free T3), which is a form of T3 that is free in the serum, and thyroid-stimulating hormone (Thyroid) secreted by the pituitary gland. It also includes Stimulating Hormone (TSH) and Thyrotropin-Releasing Hormone (TRH).
  • TSH Stimulating Hormone
  • TRH Thyrotropin-Releasing Hormone
  • thyroid hormone may be used with the same meaning as the term “thyroid hormone.”
  • hypothyroidism refers to a condition in which the concentration of thyroid hormones in the blood is excessively high due to excessive production or secretion of hormones secreted by the thyroid gland or other factors, as well as all clinicopathological symptoms resulting therefrom. It refers to
  • hypothyroidism refers to a condition in which hormones secreted by the thyroid gland are produced or secreted excessively or the concentration of thyroid hormones in the blood is excessively low due to other factors, and the clinical-pathological symptoms that occur therefrom. It refers to everyone.
  • thyroid dysfunction includes both the aforementioned “hypothyroidism” and “hyperthyroidism.”
  • deep sleep rate refers to the ratio of deep sleep time to the subject's total sleep time.
  • deep sleep ratio corresponding to day This may mean the proportion it occupies.
  • deep sleep ratio corresponding to It can refer to the ratio of deep sleep time to the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X.
  • deep sleep ratio corresponding to day This may mean the proportion it occupies.
  • period deep sleep rate refers to the ratio of total deep sleep time to a subject's total sleep time measured over a period of two or more days.
  • deep sleep ratio for Y days corresponding to X days includes the period from 12:00:00 AM on Day It may refer to the proportion of deep sleep time among the total sleep time.
  • the deep sleep ratio for a 5-day period is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day.
  • Time, the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day are added to calculate the total sleep time, the deep sleep time corresponding to the first day, Deep sleep time corresponding to the second day, deep sleep time corresponding to the third day, deep sleep time corresponding to the fourth day, and deep sleep time corresponding to the fifth day.
  • the total deep sleep time can be calculated by adding the corresponding deep sleep time, and then calculated as the ratio of the total deep sleep time to the total sleep time.
  • proportion of deep sleep for Y days corresponding to X days may mean the ratio of deep sleep time to the total sleep time corresponding to the time from the time you started falling asleep the day before on day X-Y+1 to waking up on day X.
  • "interval deep sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.
  • light sleep rate refers to the proportion of light sleep time among the subject's total sleep time.
  • light sleep ratio corresponding to day This may mean the proportion it occupies.
  • light sleep ratio corresponding to day It can refer to the proportion of light sleep time among the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X.
  • light sleep ratio corresponding to day This may mean the proportion it occupies.
  • period light sleep rate refers to the proportion of the total light sleep time of a subject's total sleep time measured over a period of two or more days.
  • light sleep ratio for Y days corresponding to X days includes the period from 12:00:00 AM on Day It may refer to the proportion of light sleep time among the total sleep time.
  • the light sleep ratio for a 5-day interval is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day.
  • the total sleep time is calculated by adding the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day, the light sleep time corresponding to the first day, Light sleep time corresponding to the second day, light sleep time corresponding to the third day, light sleep time corresponding to the fourth day, and light sleep time corresponding to the fifth day.
  • the total light sleep time can be calculated by adding the corresponding light sleep time, and then calculated as the ratio of the total light sleep time to the total sleep time.
  • percentage of light sleep for Y days corresponding to X days may mean the ratio of light sleep time to the total sleep time corresponding to the time from the time you started falling asleep the day before on day X-Y+1 to waking up on day X.
  • percentage of light sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.
  • REM sleep rate refers to the ratio of REM sleep time to the subject's total sleep time.
  • REM sleep ratio corresponding to day This may mean the proportion it occupies.
  • REM sleep ratio corresponding to day It can refer to the proportion of REM sleep time among the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X.
  • REM sleep ratio corresponding to day This may mean the proportion it occupies.
  • period REM sleep rate refers to the ratio of the total REM sleep time to the subject's total sleep time measured over a period of 2 days or more.
  • REM sleep ratio for Y days corresponding to It may refer to the proportion of REM sleep time among the total sleep time.
  • the REM sleep ratio for a 5-day period is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day.
  • Time, the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day are added to calculate the total sleep time, the REM sleep time corresponding to the first day, REM sleep time corresponding to the second day, REM sleep time corresponding to the third day, REM sleep time corresponding to the fourth day, and REM sleep time corresponding to the fifth day.
  • the total REM sleep time can be calculated by adding the corresponding REM sleep time, and then calculated as the ratio of the total REM sleep time to the total sleep time.
  • REM sleep ratio for Y days corresponding to X days may mean the ratio of REM sleep time to the total sleep time corresponding to the time from the time you started falling asleep on the day before X-Y+1 to waking up on day X.
  • "REM sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.
  • the applicant of this application determined that the patient's thyroid function was "normal” based on the correlation between heart rate and hyperthyroidism in order to solve the problems of the prior art and develop technology that meets the needs of the market.
  • the heart rate corresponding to the point in time and the current heart rate are used as the main input variables to determine whether the patient's current state is a state in which thyroid function is normal or a state in which there is an abnormality in thyroid function (i.e., hyperthyroidism or thyroid function).
  • each person's heart rate (average heart rate for a predetermined period) when thyroid function is normal is different for each person, the heart rate in normal state and the current heart rate are simply used as factors to determine whether thyroid function is abnormal. There was difficulty in deriving a function that outputs .
  • Personal information about selected patients is obtained.
  • personal information of selected patients is obtained, including name, gender, age, weight, height, etc.
  • Each selected patient is provided with a wearable device that can be worn on the wrist and is equipped with a function that can measure the patient's heart rate.
  • a doctor's diagnosis result for the patient is secured. That is, as a result of the thyroid hormone test, it is confirmed whether the patient has hyperthyroidism, hypothyroidism, or normal condition.
  • the value obtained by subtracting the section average heart rate corresponding to the second test day from the section average heart rate corresponding to the first test day and the hormone concentration of the second test day It can be used as an input value, and the diagnosis result corresponding to the first inspection date can be used as a labeling value.
  • classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.
  • the learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.
  • a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.
  • clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the second examination day.
  • Learning data sets in the following format can be created: corresponding thyroid hormone concentration, diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date.
  • the diagnosis result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnosis result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnostic result for the third test was 'normal'
  • the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:
  • the diagnostic result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'hyperactivity'.
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test.
  • not all clinical data obtained when learning a hypothyroidism prediction model can be used, and only clinical data of patients who have been diagnosed with hypothyroidism at least once can be used as learning data.
  • hypothyroidism patient a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’
  • hypothyroidism prediction model in order to learn a hypothyroidism prediction model, clinical data of patients with hypothyroidism were used to calculate [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the second examination day.
  • Learning data sets in the following format can be created: corresponding thyroid hormone concentration, diagnostic result (low or not low) corresponding to the first test date.
  • the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'decreased'
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.
  • the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.
  • the section average heart rate corresponding to the first test day is subtracted from the section average heart rate corresponding to the second test day, the hormone concentration on the first test day, and
  • the section average heart rate corresponding to the second test date can be used as an input value, and the diagnosis result corresponding to the first test date can be used as a labeling value.
  • classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.
  • the learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.
  • the second embodiment when learning a hyperthyroidism prediction model, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once in the diagnosis results can be used as learning data. .
  • a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.
  • a hyperactivity disorder prediction model in order to learn a hyperactivity disorder prediction model, clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the first examination day.
  • Learning data sets in a format such as the corresponding section average heart rate, thyroid hormone concentration corresponding to the second test date, and diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day can be created.
  • the diagnosis result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnosis result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnostic result for the third test was 'normal'
  • the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the first test day second concentration, hyperactivity
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day
  • section average heart rate corresponding to the first test day third concentration, hyperactivity
  • section average heart rate corresponding to the second test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the second test day first concentration, no hyperactivity
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the third test day first concentration, no hyperactivity
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the third test day second concentration, no hyperactivity
  • the diagnostic result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'hyperactivity'.
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test.
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the first test day second concentration, hyperactivity
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day
  • section average heart rate corresponding to the first test day third concentration, hyperactivity
  • section average heart rate corresponding to the second test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the second test day first concentration, no hyperactivity
  • section average heart rate corresponding to the third test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the third test day first concentration, hyperactivity
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the third test day second concentration, hyperactivity
  • section average heart rate corresponding to the fourth test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the fourth test day first concentration, no hyperactivity
  • section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the fourth test day second concentration, no hyperactivity
  • hypothyroidism patient a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’
  • hypothyroidism prediction model In order to learn a hypothyroidism prediction model according to the second embodiment, clinical data of patients with hypothyroidism were used to obtain [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), corresponding to the first examination day.
  • Learning data sets in the following format can be created: average heart rate for the interval, thyroid hormone concentration corresponding to the second test date, and diagnosis result (deterioration or not) corresponding to the first test date.
  • the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the first test day second concentration, decline
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day
  • section average heart rate corresponding to the first test day third concentration, decline
  • section average heart rate corresponding to the second test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the second test day first concentration, not low
  • section average heart rate corresponding to the third test day [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the third test day, first concentration, not low]
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the third test day second concentration, not low
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'decreased'
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the first test day second concentration, decline
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day
  • section average heart rate corresponding to the first test day third concentration, decline
  • section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day
  • section average heart rate corresponding to the first test day fourth concentration, decline
  • section average heart rate corresponding to the second test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the second test day first concentration, not low
  • section average heart rate corresponding to the third test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the third test day first concentration, decline
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the third test day second concentration, decline
  • section average heart rate corresponding to the third test day - section average heart rate corresponding to the fourth test day
  • section average heart rate corresponding to the third test day fourth concentration, decline
  • section average heart rate corresponding to the fourth test day section average heart rate corresponding to the first test day
  • section average heart rate corresponding to the fourth test day first concentration, not low
  • section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the second test day
  • section average heart rate corresponding to the fourth test day second concentration, not low
  • the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.
  • the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day.
  • a value or a combination thereof can be used as an input value.
  • the section average heart rate corresponding to the first inspection day is subtracted from the section average heart rate corresponding to the second inspection day, and the section heart rate corresponding to the first inspection day
  • the standard deviation of the heart rates corresponding to the second test date is subtracted from the standard deviation of the heart rates and the hormone concentration of the second test day can be used as input values, and the diagnosis result corresponding to the first test date can be used as a labeling value.
  • classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.
  • the learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.
  • a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.
  • clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), (first examination day Standard deviation of section heart rates corresponding to - standard deviation of section heart rates corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day].
  • Learning data sets can be created.
  • the diagnosis result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnosis result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnostic result for the third test was 'normal'
  • the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:
  • the diagnostic result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'hyperactivity'.
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test.
  • hypothyroidism patient a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’
  • the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'decreased'
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.
  • the explanation was based on using the difference value of the standard deviation as the input value, but the calculation was made between the section heart rates corresponding to the X test date and the section heart rates corresponding to the Y test day.
  • the difference value of the standard deviation of the section heart rates, the difference value of the relative standard deviation of the section heart rates, the difference value of the skewness of the section heart rates, the difference value of the kurtosis of the section heart rates, and Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of corresponding section heart rates and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, may be used as an input value.
  • the section average heart rate corresponding to the first inspection day is subtracted from the section average heart rate corresponding to the second inspection day, and the section heart rate corresponding to the first inspection day
  • the standard deviation of the section heart rates corresponding to the second test day is subtracted from the standard deviation of the section average heart rate corresponding to the first test day and the hormone concentration of the second test day are used as input values, and the Diagnosis results can be used as labeling values.
  • the fourth embodiment differs only in that it uses “segment average heart rate on the first test day” as an input value and most other things are the same, so detailed description will be omitted here.
  • the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day.
  • a value or a combination thereof can be used as an input value.
  • the explanation was based on using the difference value of the standard deviation as the input value, but the calculation was made between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day.
  • the difference value of the standard deviation of the section heart rates, the difference value of the relative standard deviation of the section heart rates, the difference value of the skewness of the section heart rates, the difference value of the kurtosis of the section heart rates, and Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of corresponding section heart rates and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, may be used as an input value.
  • the rate of change of the interval average heart rate of the first examination day with respect to the second examination day i.e., (interval average heart rate corresponding to the first examination day - the second examination day) (section average heart rate corresponding to) / (section average heart rate corresponding to the second test day)
  • the hormone concentration of the second test day are used as input values, and the diagnosis result corresponding to the first test day can be used as a labeling value.
  • the fifth embodiment differs only in that it uses "rate of change in section average heart rate” instead of “change amount in section average heart rate” and most other things are the same, so a detailed description is given here. Decided to omit it.
  • Rate of change from heart rate on the reference day average heart rate in the section on the target day, hormone value on the reference day
  • the rate of change of the interval average heart rate of the first test day with respect to the second test day i.e., (interval average heart rate corresponding to the first test day - the second (section average heart rate corresponding to the test day) / (section average heart rate corresponding to the second test day)
  • the section average heart rate corresponding to the first test day and the hormone concentration of the second test day are used as input values, and on the first test day
  • the corresponding diagnostic result can be used as a labeling value.
  • the sixth embodiment differs only in that it uses “segment average heart rate on the first test day” as an input value and most other things are the same, so detailed description will be omitted here.
  • the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day.
  • a value or a combination thereof can be used as an input value.
  • the rate of change of the interval average heart rate of the first test day with respect to the second test day i.e., (interval average heart rate corresponding to the first test day - 2 Average heart rate of the section corresponding to the test day) / (average heart rate of the section corresponding to the second test day)
  • the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day i.e., (section corresponding to the first test day) Standard deviation of heart rates / average heart rate of the section corresponding to the first test day) - (standard deviation of the heart rates of the section corresponding to the second test day / average heart rate of the section corresponding to the second test day)
  • the hormone concentration of the second test day are input values.
  • the diagnosis result corresponding to the first inspection date can be used as a labeling value.
  • the seventh embodiment differs only in using "the amount of change in the relative standard deviation of section heart rates" as an input value, and most other things are the same, so detailed description will be omitted here. .
  • the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day.
  • the difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.
  • the rate of change of the interval average heart rate of the first test day with respect to the second test day i.e., (interval average heart rate corresponding to the first test day - 2 Average heart rate of the section corresponding to the test day) / (average heart rate of the section corresponding to the second test day)
  • the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day i.e., (section corresponding to the first test day) Standard deviation of heart rates / section average heart rate corresponding to the first test day) - (standard deviation of section heart rates corresponding to the second test day / section average heart rate corresponding to the second test day)
  • section average heart rate corresponding to the first test day i.e., (interval average heart rate corresponding to the first test day - 2 Average heart rate of the section corresponding to the test day) / (average heart rate of the section corresponding to the second test day)
  • the hormone concentration on the second test day can be used as an input value
  • the eighth embodiment differs only in that it further uses “segment average heart rate on the first reference day” as an input value, and most other things are the same, so detailed description will be omitted here.
  • the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day. A value or a combination thereof can be used as an input value.
  • the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day.
  • the difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.
  • Rate of change (or amount of change) from baseline heart rate, hormone values on baseline date, changes in distribution, and differences between test days
  • the rate of change of the interval average heart rate of the first examination day with respect to the second examination day i.e., (interval average heart rate corresponding to the first examination day - the second examination day) (section average heart rate corresponding to the second test day) / (section average heart rate corresponding to the second test day)
  • the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day i.e., (segment heart rate corresponding to the first test day) standard deviation of section heart rates corresponding to the first test day) - (standard deviation of section heart rates corresponding to the second test day / section average heart rate corresponding to the second test day)
  • hormone concentration on the second test day and first test day The difference (days) between and the second inspection date can be used as an input value, and the diagnosis result corresponding to the first inspection date can be used as a labeling value.
  • the ninth embodiment differs only in that the difference between the first inspection date and the second inspection date is used as an input value, and most other things are the same, so detailed description will be omitted here.
  • the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test date and the section heart rates corresponding to the Y test day
  • the difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.
  • the rate of change from the heart rate on the reference day and the amount of change in the relative standard deviation it is also possible to use the amount of change from the heart rate on the reference day and the amount of change in the standard deviation.
  • the difference value of the standard deviation of the section heart rates which can be calculated between the section heart rates corresponding to the The difference value of relative standard deviation, the difference value of skewness of section heart rates, the difference value of kurtosis of section heart rates, and the distribution of section heart rates corresponding to the Any one value selected from the group consisting of JS divergence, which represents the difference between distributions, or a combination thereof, can be used as an input value.
  • the combination value of the variables disclosed in the following [Table 1] can be further used as an input value.
  • the first section average heart rate, first standard deviation, first skewness, first kurtosis, first TSH hormone concentration, first free T4 concentration, first T4 concentration, first free T3 concentration, first T3 Concentration, first TRH concentration, and diagnosis results for thyroid dysfunction for the first reference date are corresponded, and on the second reference date, the second section average heart rate, second standard deviation, second skewness, second kurtosis, and second
  • the diagnosis results for dysthyroidism for the TSH hormone concentration, second free T4 concentration, second T4 concentration, second free T3 concentration, second T3 concentration, second TRH concentration and second reference date correspond to the above
  • Zone average heart rate change rate (segment average heart rate on the second reference day - section average heart rate on the first reference day) / section average heart rate on the second reference day
  • Section harmonic average heart rate change rate (section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day) / section harmonic average heart rate on the second reference day
  • Standard deviation change standard deviation of zone heart rates on the second reference day - standard deviation of zone heart rates on the first reference day
  • Relative standard deviation change (standard deviation of section heart rates on the second reference day / section average heart rate on the second reference day) - (standard deviation of section heart rates on the first reference day / section average heart rate on the first reference day)
  • the combination value between the variables is one value selected from among the variables (hereinafter referred to as the first value) and another value selected from the variables (hereinafter referred to as the second value). It means one of a multiplied value, a value obtained by dividing a first value by a second value, and a value obtained by dividing a second value by a first value.
  • the combination value may be the concentration value of free T4 on the second reference date divided by JS Divergence.
  • the input value may be as follows.
  • the labeling value may be the diagnosis result on the first inspection day.
  • the label value is 'hyperthyroidism'
  • the diagnosis result on the first test day is 'normal'
  • the label value is can be labeled as 'not hyperthyroid'
  • training a hypothyroidism prediction model if the diagnosis result on the first test day is 'hypothyroidism', the label value is 'hypothyroidism', and the diagnosis result on the first test day is 'hypothyroidism'. If it is ‘normal’, the label value can be labeled as ‘not deteriorated’.
  • the target's personal information in order to train a learning model, can be used as an input value in addition to the input values described in the above-described first to tenth embodiments.
  • the subject's gender and/or age may be used as input values.
  • the rate of change of the interval average heart rate of the first test day with respect to the second test day i.e., (corresponds to the first test day) average heart rate of the section corresponding to the second test day - average heart rate of the section corresponding to the second test day) / (average heart rate of the section corresponding to the second test day)
  • average heart rate of the section corresponding to the first test day hormone concentration, gender and age of the second test day
  • the thyroid hormone concentration corresponding to the first test date is used.
  • the value can be used as an input value.
  • the value related to the thyroid hormone concentration corresponding to the first test date can be obtained based on the output value of the thyroid hormone prediction model.
  • the value related to the thyroid hormone concentration may be the thyroid hormone concentration value of the target date estimated based on the output value of the thyroid hormone prediction model.
  • the value related to the thyroid hormone concentration may be a change value or amount of change in the thyroid hormone concentration on the target day with respect to the thyroid hormone concentration on the reference day estimated based on the output value of the thyroid hormone prediction model.
  • a value related to thyroid hormone concentration may be an output value of a thyroid hormone prediction model.
  • the value regarding the thyroid hormone concentration corresponding to the first test date may be one or more values. For example, a first value generated based on the output value of a thyroid hormone concentration prediction model specialized for hyperthyroidism, a second value generated based on the output value of a thyroid hormone concentration prediction model specialized for normal thyroid function, and a second value generated based on the output value of a thyroid hormone concentration prediction model specialized for normal thyroid function, and The learning model may be trained using all third values generated based on the output value of the thyroid hormone concentration prediction model, but is not limited to this.
  • classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.
  • the learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.
  • the 12th embodiment when learning a hyperthyroidism prediction model, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once in the diagnosis results can be used as learning data. .
  • a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.
  • clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the first examination day.
  • Learning data sets in the following format may be generated: a value for the corresponding estimated thyroid hormone concentration, a thyroid hormone concentration corresponding to the second test date, and a diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date.
  • the diagnosis result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnosis result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnostic result for the third test was 'normal'
  • the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:
  • the diagnostic result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'hyperactivity'.
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test.
  • the diagnosis result for the first test is 'hyperactivity'
  • the hormone concentration at this time was the first concentration
  • the diagnosis result for the second test is 'normal'.
  • the hormone concentration was the second concentration
  • the diagnostic result for the third test was 'decreased'.
  • the learning data that can be obtained from the clinical data of the first hyperactivity disorder patient Sets can be created as follows:
  • hypothyroidism patient a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’
  • hypothyroidism prediction model In order to learn a hypothyroidism prediction model according to the 12th embodiment, clinical data of patients with hypothyroidism were used to determine [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), corresponding to the first examination day.
  • Learning data sets in the following format may be generated: a value for the estimated thyroid hormone concentration, a thyroid hormone concentration corresponding to the second test date, and a diagnosis result (low or not low) corresponding to the first test date.
  • the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:
  • the hormone concentration at this time was the first concentration
  • the diagnostic result for the second test is 'normal'
  • the hormone concentration was the second concentration
  • the diagnosis result for the third test was 'decreased'
  • the hormone concentration was the third concentration
  • the diagnosis result for the fourth test was 'normal'.
  • the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.
  • the learning data that can be obtained from the clinical data of the first hypothyroidism patient Sets can be created as follows:
  • the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.
  • the core idea disclosed in the 12th embodiment is that the “value related to the estimated thyroid hormone concentration corresponding to the first test date” is used as input, and other information input to the learning model can be added or omitted.
  • the learning model may be implemented in a form that further uses a value related to the estimated thyroid hormone concentration corresponding to the first test date as input to the input data disclosed in the first to eleventh embodiments.
  • Example 5 in order to learn a learning model, clinical data of patients with hyperactivity were used to calculate [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / (second test day) Learning data sets in the following format are generated: section average heart rate corresponding to the test date)), estimated thyroid hormone concentration value corresponding to the first test date, thyroid hormone concentration corresponding to the second test date, and diagnosis result corresponding to the first test day. It can be.
  • Example 9 in order to learn a learning model, clinical data of patients with hyperactivity were used to calculate [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / (second test day) Average heart rate of the section corresponding to the test date)), difference between the first test date and the second test date (days), estimated thyroid hormone concentration value corresponding to the first test day, thyroid hormone concentration corresponding to the second test day, Learning data sets in the same format as [corresponding diagnostic results] can be created.
  • learning data for hyperthyroidism or non-hyperthyroidism is generated from clinical data of patients diagnosed with hyperthyroidism at least once in the diagnosis results.
  • learning data about hyperthyroidism can be generated from clinical data of patients who have never been diagnosed with hyperthyroidism as a result of the diagnosis.
  • learning data for hypothyroidism or non-hypothyroidism is generated using clinical data of patients diagnosed with hypothyroidism at least once in the diagnosis results.
  • learning data about hypothyroidism can be generated from clinical data of patients who have never been diagnosed with hypothyroidism.
  • FIG. 1 is a diagram illustrating a system for predicting thyroid dysfunction according to an embodiment disclosed by the present application.
  • the system 1 includes a plurality of heart rate measuring devices (10), a plurality of user terminals (20), and a server (30).
  • the plurality of heart rate measurement devices 10 the plurality of user terminals 20, and the server 30 will be described in more detail.
  • the plurality of heart rate measurement devices 10 can measure the heart rate of a subject (user).
  • the heart rate measurement devices 10 may measure the subject's heart rate (beats per minute) per unit time. That is, the heart rate measurement devices 10 can measure the heart rate per minute of the subject.
  • the plurality of heart rate measurement devices 10 may store the measured heart rates of the target. At this time, when the heart rate measurement devices 10 store the plurality of heart rates, they may store the time points at which the heart rate (heart rate per unit time) was measured by matching them together.
  • the plurality of heart rate measurement devices 10 may transmit the stored heart rates to the plurality of user terminals 20 and/or the server 30.
  • the plurality of heart rate measurement devices 10 may receive prediction results regarding thyroid dysfunction from the server 30 or the user terminals 20.
  • the heart rate measurement device may be a wearable watch, a wearable band, or a patch type device.
  • the plurality of user terminals 20 transmit information to the server 30 and receive information from the server 30 through various networks.
  • the plurality of user terminals 20 may receive the measured heart rates from the heart rate measurement device 10 and store the received heart rates.
  • the plurality of user terminals 20 may preprocess the stored heart rates. For example, the user terminals 20 may select heart rates measured in the resting state from among all stored heart rates. For another example, the user terminals 20 may select the above-described zone heart rates from all the stored heart rates based on a specific day. As another example, the user terminals 20 may calculate the average of heart rates in selected sections based on the specific day. For another example, the user terminals 20 may calculate the standard deviation, relative standard deviation, skewness, kurtosis, etc. of the heart rates in selected sections based on the specific day.
  • the plurality of user terminals 20 can obtain the test date of the user's hormone concentration and obtain test results (for example, hormone concentration and/or doctor's diagnosis results, etc.) corresponding to the test date. .
  • the plurality of user terminals 20 may obtain the inspection date and its inspection results through user input, or may obtain them by receiving them from the outside through a network.
  • the plurality of user terminals 20 may transmit the stored heart rates and/or the preprocessed results to the server 30 .
  • the plurality of user terminals 20 may transmit the test date and test results to the server 30.
  • the plurality of user terminals 20 may receive prediction results regarding thyroid dysfunction processed by the server 30 from the server 30 .
  • the plurality of user terminals 20 may be at least one of a smartphone, tablet PC, laptop, and desktop.
  • the server 30 transmits information to the plurality of user terminals 20 through various networks, and also receives information from the plurality of user terminals 20.
  • the server 30 may receive all or part of the heart rates measured from the plurality of user terminals 20. At this time, the server 30 may preprocess the received heart rates.
  • the server 30 may receive a result preprocessed by the user terminal 20 from the plurality of user terminals 20 .
  • the server 30 can obtain the hormone concentration test date and test results for each user.
  • the test date and test results may be received from the user terminal 20 or from another external device.
  • the server 30 may obtain a prediction result for the target's thyroid dysfunction based on the preprocessed results, etc.
  • the server 30 may transmit the prediction result for thyroid dysfunction to the heart rate measurement device 10 and/or the plurality of user terminals 20.
  • terminal software In order to perform communication between the user terminals 20 and the server 30, terminal software needs to be installed on the plurality of user terminals 20, and server software must be installed on the server 30. needs to be
  • a plurality of prediction models can be used to predict thyroid dysfunction based on the preprocessed results.
  • the plurality of prediction models may be driven by software installed on the server 30.
  • the plurality of prediction models may be driven by terminal software installed on the user terminals 20.
  • some of the plurality of learning models may be executed by the user terminals 20 and the remaining portions may be executed by the server 30.
  • FIG. 2 is a block diagram of a heart rate measurement device disclosed by the present application.
  • the heart rate measurement device 10 includes an output unit 110, a communication unit 120, a memory 130, a heart rate measurement unit 140, and a controller 150.
  • the output unit 110 outputs various information according to control commands from the controller 150.
  • the output unit 110 may include a display 112 that visually outputs information to the user.
  • it may include a speaker that audibly outputs information to the user and a vibration motor that tactilely outputs information to the user.
  • the communication unit 120 may include a wireless communication module and/or a wired communication module.
  • the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.
  • the memory 130 stores executable codes that can be read by the controller 150, processed result values, and necessary data.
  • the memory 130 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc.
  • the memory 130 may store the above-described terminal software, and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above. Furthermore, the memory 130 may store the heart rate of the target obtained through the heart rate measuring unit 140 and the measurement time of the heart rate.
  • the heart rate measurement unit 140 can measure the heart rate per minute of the subject (user).
  • the method of measuring a subject's heart rate per minute is widely known, so a detailed description thereof will be omitted.
  • Controller 150 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 130. Specifically, the controller 150 may process information according to terminal software running on the heart rate measurement device 10, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 150 controls the overall operation of the heart rate measurement device 10.
  • the heart rate measuring device 10 may include a user input unit.
  • the heart rate measurement device 10 may receive various information required for operation of the heart rate measurement device 10 from the user through the user input unit.
  • Figure 3 is a block diagram of a user terminal disclosed by the present application.
  • the user terminal 20 disclosed by the present application includes an output unit 210, a communication unit 220, a memory 230, and a controller 250.
  • the output unit 210 outputs various information according to control commands from the controller 250.
  • the output unit 210 may include a display 212 that visually outputs information to the user.
  • it may include a speaker that audibly outputs information to the user and a vibration motor that tactilely outputs information to the user.
  • the communication unit 220 may include a wireless communication module and/or a wired communication module.
  • the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.
  • the memory 230 stores executable codes that can be read by the controller 250, processed results, and necessary data.
  • the memory 230 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc.
  • the memory 230 may store the above-described terminal software, and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above.
  • Controller 240 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 230. Specifically, the controller 240 may process information according to terminal software running on the user terminal 20, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 240 controls the overall operation of the user terminal 20.
  • the user terminal 20 may include a user input unit.
  • the user terminal 20 can receive various information required for operation of the user terminal 20 from the user through the user input unit.
  • FIG. 4 is a block diagram of the server disclosed by this application.
  • the server 30 disclosed by the present application includes a communication unit 310, a memory 320, and a controller 330.
  • the communication unit 310 may include a wireless communication module and/or a wired communication module.
  • the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.
  • the memory 320 stores executable codes that can be read by the controller 330, processed results, and necessary data.
  • the memory 320 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc.
  • the memory 320 may store the above-described server software and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above. Furthermore, the memory 320 may store heart rate and/or pre-processing results for each user from the user terminal 20.
  • the controller 330 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 320. Specifically, the controller 330 may process information according to server software running on the server 30, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 330 controls the overall operation of the server 30.
  • the method for predicting thyroid dysfunction disclosed by the present application can be performed by the above-described system 1.
  • Figure 5 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.
  • the method for predicting thyroid dysfunction for a subject disclosed by the present application acquires a trigger signal (S100), and obtains heart rates in the section corresponding to the target date determined by the trigger signal.
  • a trigger signal may be obtained (S100).
  • the trigger signal refers to a signal requesting to obtain a prediction result for the subject's thyroid dysfunction.
  • the trigger signal may be generated by the user's input.
  • the user requests to obtain a prediction result for thyroid dysfunction through the user input unit of the user terminal 20, and in response to the user's request, the user terminal 20 generates the trigger signal.
  • the user requests to obtain a prediction result for thyroid dysfunction through the user input unit of the heart rate measurement device 10 that is communicating with the user terminal 20, and responds to the user's request.
  • the heart rate measurement device 10 or the user terminal 20 can generate the trigger signal.
  • the trigger signal may be automatically generated by the system 1 regardless of the user's request.
  • software installed on the heart rate measuring device 10, the user terminal 20, or the server 30 may generate the trigger signal according to a predetermined rule. If the trigger signal is generated by the heart rate measurement device 10 or the server 30, the generated trigger signal may be transmitted to the user terminal 20.
  • the target date may be determined by the trigger signal.
  • the day on which the trigger signal was generated may be determined as the target date.
  • a specific day may be designated by the user's input or the system 1, in which case the designated specific day is the target. It can be decided by work.
  • Zone heart rates corresponding to the target date may be obtained according to a predetermined period (S110).
  • the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23. , 24, 25, 26, 27, 28, 29, and 30 days.
  • the zone heart rates correspond to resting heart rates corresponding to July 21st, resting heart rates corresponding to July 20th, and July 19th. Resting heart rates corresponding to July 18, Resting heart rates corresponding to July 17, Resting heart rates corresponding to July 16, Resting heart rates corresponding to July 15, 07 Resting heart rates corresponding to the 14th of the month, resting heart rates corresponding to the 13th of July, and resting heart rates corresponding to the 12th of July are obtained as the zone heart rates.
  • the zone heart rates are selected from all heart rates stored corresponding to days determined by the determined target date and the predetermined period. Based on the measurement times corresponding to the heart rates, heart rates at which the measurement times are within the resting phase of the subject (i.e., the user) are selected as resting heart rates.
  • the resting period can be determined in various ways.
  • the rest period may be determined as a section in which the user's movement is below a reference value through the heart rate measurement device 10 or a motion sensor built into the user terminal 20.
  • the rest period may be determined as a section in which the user is determined to be resting or sleeping through various sensors built into the heart rate measurement device 10 or the user terminal 20.
  • the rest period may be determined by the sleep period (sleep start time and sleep end time) directly input by the user through the heart rate measurement device 10 or the user terminal 20.
  • the heart rates of the sections corresponding to the target date may be preprocessed (S120).
  • the average of heart rates in the sections corresponding to the target date may be calculated.
  • the harmonic average of heart rates in the section corresponding to the target date may be calculated.
  • various indicators related to the distribution of heart rates in the sections corresponding to the target date may be calculated.
  • the standard deviation of heart rates in the section corresponding to the target date may be calculated.
  • the skewness of heart rates in the sections corresponding to the target date may be calculated.
  • kurtosis of heart rate intervals corresponding to the target date may be calculated.
  • the relative standard deviation of heart rates in the zones corresponding to the target date may be calculated.
  • two or more of the average of section heart rates corresponding to the target date, the harmonic average of section heart rates corresponding to the target day, and the index regarding the distribution of section heart rates corresponding to the reference date may be calculated.
  • some heart rate values included in the zone heart rates may be removed as noise.
  • the section average heart rate can be calculated using the section heart rate from which noise has been removed.
  • the distribution of zone heart rates can be calculated using the zone heart rates from which noise has been removed.
  • Noise referred to in this specification refers to a heart rate value that is determined not to be used according to a predetermined standard, and is not a heart rate value that is removed after confirming that it is not a measurement of the actual heart rate.
  • a method of removing noise from the section heart rate may be in the form of not using heart rates that differ by more than a threshold based on the median of the heart rates of a preset section length.
  • the preset section length may correspond to time. For example, if the preset section length is 10 minutes, the median value is calculated based on all heart rate values obtained for 10 minutes, and the heart rate to be removed as noise can be determined by comparing the calculated median value with each heart rate value.
  • the number of heart rate values actually obtained in 10 minutes may change depending on the specifications of the heart rate measurement device or may change depending on the heart rate measurement environment.
  • the analysis may be performed by dividing the entire section heart rate equally into preset section lengths.
  • the analysis sections divided by preset section lengths are connected to each other. Can have overlapping areas.
  • the heart rate from 0 minutes to 10 minutes is set as the first analysis section, and the heart rate from 10 minutes to 20 minutes is set as the first analysis section.
  • noise analysis can be performed with 11 ((60 minutes/5 minutes)-1) analysis sections.
  • heart rates that differ by more than a threshold based on the median of the heart rates of the section set based on the number of single heart rate values may not be used. For example, if the number of single heart rate values is 10, the median value is calculated based on 10 consecutive heart rates without considering whether they were acquired for 30 minutes or 10 minutes, and the calculated median value and each By comparing heart rate values, you can determine the heart rate to be removed as noise.
  • a commonly used method may be applied to remove noise from section heart rates.
  • a form of removing noise from section heart rates can be used when preprocessing section heart rates of a target day, when preprocessing section heart rates of a reference day, as well as when generating learning data.
  • the section heart rate corresponding to the target date is calculated from the section heart rates corresponding to the target date. Only the mean and standard deviation of the values can be obtained. However, all or some of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.
  • the reference date for the object may be determined (S130).
  • the reference date may be selected from at least one test date on which the subject underwent a hormone test. That is, the reference date is selected from among the subject's past hormone test dates.
  • the reference date may be selected from at least one or more days on which the subject's hormone concentration is stored correspondingly. That is, the reference date may be selected from days in which the subject's past hormone concentration is stored.
  • Only one such reference date may be selected, but is not limited thereto. That is, multiple reference dates may be selected.
  • the reference date may be determined as the closest day based on the target date.
  • the hormone concentration corresponding to the determined reference date can be obtained.
  • the hormone concentration may be one or a combination thereof selected from the group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and TRH concentration.
  • the hormone concentrations corresponding to the determined reference date only the hormone concentration values used for learning the prediction model described later can be selectively selected. For example, when the hormone concentration corresponding to the determined reference date includes all of the concentration of T4, concentration of free T4, concentration of T3, concentration of free T3, concentration of TSH, and concentration of TRH, learning of the prediction model If only the concentrations of free T4 and TSH are used, only the concentrations of free T3 and TSH can be obtained.
  • hormone concentrations corresponding to each of the multiple reference days can be obtained.
  • the stored preprocessing result may be obtained.
  • preprocessing of the zone heart rates corresponding to the reference date may be performed.
  • zone heart rates corresponding to the reference date and preprocessing thereof are similar to the acquisition of zone heart rates corresponding to the target date and preprocessing thereof, detailed description thereof will be omitted here.
  • the preprocessing result of the section heart rates for the reference date is the average of the section heart rates corresponding to the reference date, the harmonic average of the section heart rates corresponding to the reference date, and the index of the distribution of section heart rates corresponding to the reference date. It can contain at least one.
  • the indicators regarding the distribution of section heart rates corresponding to the reference date include the standard deviation of section heart rates corresponding to the target date, the skewness of section heart rates corresponding to the target date, and the target date. It may include at least one of kurtosis of corresponding section heart rates and relative standard deviation of section heart rates corresponding to the target date.
  • some heart rate values included in the zone heart rates may be removed as noise.
  • the section average heart rate can be calculated using the section heart rate from which noise has been removed.
  • the distribution of zone heart rates can be calculated using the zone heart rates from which noise has been removed.
  • preprocessing results of section heart rates corresponding to each of the plurality of reference days may be obtained.
  • the various preprocessing results of the heart rates in the section corresponding to the reference date only preprocessing results related to the input value used for learning the prediction model, which will be described later, can be obtained.
  • the average of the section heart rates corresponding to the reference date is calculated from the section heart rates corresponding to the reference date. and only the standard deviation can be obtained.
  • all or some of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.
  • the difference between the preprocessing result for the zone heart rates of the target day and the preprocessing result for the zone heart rates of the reference day may be one of the following or a combination of these.
  • differences in preprocessing results for each of the multiple reference dates may be obtained. For example, the difference between the first preprocessing result determined between the first reference date and the target date and the difference between the second preprocessing result determined between the second reference date and the target date may be obtained.
  • the differences in the preprocessing results only the differences in the preprocessing results related to the input values used for learning the prediction model described later can be obtained.
  • the average of section heart rates corresponding to the target date average heart rate of the section on the target day
  • the rate of change with respect to the average of section heart rates corresponding to the reference date segment average heart rate of the reference date
  • the change in the standard deviation of section heart rates corresponding to the reference date of the standard deviation of section heart rates corresponding to the target date can be obtained.
  • differences in all or part of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.
  • combination values of variables may be calculated.
  • the first section average heart rate, first standard deviation, first skewness, first kurtosis, first TSH hormone concentration, first free T4 concentration, first T4 concentration, first free T3 concentration, first T3 Concentration, first TRH concentration, and diagnosis results for dysthyroidism for the first reference date are corresponded, and on the second reference date, the second section average heart rate, second standard deviation, second skewness, second kurtosis, and second
  • diagnosis results for dysthyroidism for the TSH hormone concentration, second free T4 concentration, second T4 concentration, second free T3 concentration, second T3 concentration, second TRH concentration and second reference date correspond to the above
  • the variables are calculated as follows.
  • section average heart rate section average heart rate on the second reference day - section average heart rate on the first reference day
  • Zone average heart rate change rate (section average heart rate on the second reference day - section average heart rate on the first reference day) / section average heart rate on the second reference day,
  • Section harmonic average heart rate change (section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day) / section harmonic average heart rate on the second reference day,
  • Standard deviation change standard deviation of zone heart rates on the second reference day - standard deviation of zone heart rates on the first reference day
  • Relative standard deviation change (standard deviation of section heart rates on the second reference day / section average heart rate on the second reference day) - (standard deviation of section heart rates on the first reference day / section average heart rate on the first reference day),
  • the combination value between the variables is one value selected from among the variables (hereinafter referred to as the first value) and another value selected from the variables (hereinafter referred to as the second value). It means one of a multiplied value, a value obtained by dividing a first value by a second value, and a value obtained by dividing a second value by a first value.
  • the combination value may be the concentration value of free T4 on the second reference date divided by JS Divergence.
  • a day gap between the target date and the reference date may be obtained.
  • the subject's personal information that is, the user's personal information
  • the user's personal information may be obtained.
  • values regarding thyroid hormone concentration corresponding to the target date can be obtained.
  • a value related to the thyroid hormone concentration obtained based on the output value of the thyroid hormone prediction model or the output value of the thyroid hormone prediction model may be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the amount of change in the average of the heart rates in each section can be entered, and accordingly, a result can be obtained as to whether the subject has hyperthyroidism or is not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day).
  • the amount of change in the average of the heart rates in each section (average heart rate in the section on the reference day) and the average heart rate in the section corresponding to the target date can be entered, and a result is obtained as to whether the subject has hyperthyroidism or not hyperthyroidism. It can be.
  • a prediction result regarding hyperthyroidism can be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the amount of change in the average of section heart rates (average section heart rate of the reference day) and the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date The difference can be entered, and a result can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the amount of change in the average of the section heart rates (the average heart rate of the section on the reference day), the average of the section heart rates corresponding to the target date, and the index of the distribution of section heart rates corresponding to the target date and the index corresponding to the reference date
  • the difference between the indices regarding the distribution of zone heart rates can be input, and accordingly a result can be obtained as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the rate of change for the average of the heart rates in the zone can be input, and accordingly, a result as to whether the subject has hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day).
  • the rate of change for the average of the heart rates in the zone (average heart rate in the zone on the reference day) and the average of the heart rates in the zone corresponding to the target date can be input, and accordingly, the result of whether the subject is hyperthyroid or not hyperthyroid is given. can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (segment average heart rate of the target day).
  • the rate of change in the average of the zone heart rates (average heart rate of the zone on the reference day), and the difference between the index of the distribution of zone heart rates corresponding to the target day and the index of the distribution of zone heart rates corresponding to the reference day. can be entered, and accordingly, a result can be obtained as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the rate of change for the average of the section heart rates (the average heart rate of the section on the reference day), the average of the section heart rates corresponding to the target date and the distribution of the section heart rates corresponding to the target day, and the index of the section heart rates corresponding to the reference date
  • Differences between indicators regarding distribution can be entered, and results can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day).
  • the difference between and the difference (day gap) between the target date and the reference date can be entered, and accordingly, a result as to whether the target is hyperthyroid or not hyperthyroid can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.
  • the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day)
  • the rate of change (or amount of change) of the average of the zone heart rates (average heart rate of the zone on the base day), an indicator of the distribution of the zone heart rates corresponding to the target date, and an indicator of the distribution of the zone heart rates corresponding to the base day.
  • Differences between variables and combination values between variables can be input, and results can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.
  • the following values can be input into the learned hyperthyroidism prediction model, and accordingly, a result of whether hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • the following values can be input into the learned hyperthyroidism prediction model, and accordingly, a result of whether hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.
  • hypothyroidism Similar to what was described in obtaining prediction results for hyperthyroidism through a hyperthyroidism prediction model, prediction results for hypothyroidism can be obtained through a hypothyroidism prediction model. A detailed description of this will be omitted.
  • the prediction results for dysthyroidism considering the prediction results for hyperthyroidism (A) and the prediction results for hypothyroidism (B) can be obtained as shown in [Table 2] below.
  • the output value (probability value) of the hyperthyroidism prediction model and the output value (probability value) of the hypothyroidism prediction model probability value
  • the result outputting a higher probability value can be adopted as the final result for thyroid dysfunction.
  • the hormone concentration corresponding to the reference date is obtained (S130)
  • the preprocessing results for the heart rates of the section corresponding to the reference date are obtained (S140)
  • acquiring a trigger signal (S100) is performed by the user terminal 20, and obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), on the target date Preprocess the corresponding section heart rates (S120), obtain the hormone concentration corresponding to the reference date (S130), obtain preprocessing results for the section heart rates corresponding to the reference date (S140), and obtain the preprocessing results for the section heart rates corresponding to the target date (S140).
  • Obtaining the difference between the preprocessing results of the section heart rates and the preprocessing results of the section heart rates corresponding to the reference date (S150), processing the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, Obtaining results on thyroid dysfunction for the subject (S160) may be performed by the server 30.
  • distributed execution it is not limited to the form of distributed execution described above, and distributed execution may be performed in more diverse forms.
  • the clinical data used in the experimental examples and comparative examples described below were collected by the method described above, and a total of 1,027 clinical data sets collected from a total of 297 patients were used, of which hormone concentration As a result of the diagnosis, there were a total of 168 clinical data sets classified as “hyperthyroidism,” a total of 801 clinical data sets classified as “normal,” and a total of 58 clinical data sets classified as “hypothyroidism.”
  • a learning data set was prepared in the following manner.
  • the value obtained by subtracting the section average heart rate corresponding to the test day when the diagnosis result was normal from the section average heart rate corresponding to the test day when the diagnosis result was not normal is used as an input value (e.g., the amount of change in section average heart rate). and the diagnostic result (increased or decreased) corresponding to the test date when the diagnostic result was not normal was used as the labeling value.
  • the value obtained by subtracting the section average heart rate corresponding to one test day (the other normal test day) from the section average heart rate corresponding to one test day (the normal test day) was used as the input value. , not high or not low were used as labeling values.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 495 data obtained from 157 patients, and the total number of learning data sets created from this was 1,199.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 437 clinical data obtained from a total of 121 patients, and a total of 1,138 test data sets were created.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #1. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day] Training data sets in the same format were created, and these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used were either free T4 alone or both free T4 and TSH.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the inventors of the present application determined the time when the hormone level was "normal” as the reference date, as in Comparative Example #1, and did not use the difference between the average heart rate of the section between the reference date and the date of analysis, and determined whether it was "normal” or "thyroid”. It is possible to sufficiently predict whether a person has “hyperthyroidism” by determining the reference date regardless of whether it is “hyperthyroidism” and by using the difference between the average heart rate of the section between the reference date and the analysis target date along with the hormone concentration corresponding to the reference date. In fact, it was found that the results such as accuracy and sensitivity were much higher. This not only made it possible to obtain more learning data sets from the same level of clinical data, but also more accurately diagnosed thyroid dysfunction. It has become possible to secure a predictive model that can make predictions.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #2. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, thyroid hormone concentration corresponding to the second test day, corresponding to the first test day. Training data sets in the same format as diagnostic results (hyperactivity or non-hyperactivity disorder) were created, and these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used were either free T4 alone or both free T4 and TSH.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the inventors of the present application have found that when predicting hyperthyroidism, the accuracy increases slightly when the section average heart rate (current section average heart rate) of the target day is used as an input value in addition to the change in section average heart rate.
  • the section average heart rate current section average heart rate
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in Learning Method Example #3 described above. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day) , thyroid hormone concentration corresponding to the second examination day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first examination day] were created, and these learning data sets were used to train the model. .
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used included both free T4 and TSH concentrations as input values.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the inventors of the present application found that in predicting hyperthyroidism, when the change in the standard deviation of the section heart rate was used as an input value in addition to the change in the section average heart rate, the accuracy increased slightly.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #4.
  • standard deviation and skewness were used as indicators related to distribution. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day) , (skewness of segment heart rates corresponding to the first examination day - skewness of segment heart rates corresponding to the second examination day), thyroid hormone concentration corresponding to the second examination day, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first examination day )], training data sets in the same format were created, and these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used included both free T4 and TSH concentrations as input values.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • Experimental Example #5 compared to Experimental Example #4, the training data sets and test data sets were the same except that standard deviation and kurtosis were used instead of standard deviation and skewness as indices related to distribution. was created.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • Experimental Example #6 compared to Experimental Example #4, used the same learning data sets except that standard deviation and JS Divergence were used instead of using the difference in standard deviation and skewness due to the difference in indicators related to the distribution. and test data sets were created.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #5. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Learning data sets in the following formats were created: (relative standard deviation of heart rates in the interval corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, and diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day. , these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used were based on both free T4 and TSH concentrations.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • Experimental Example #7 is the same as Experimental Example #3, except that instead of using the change amount of the section average heart rate and the change amount of the standard deviation, the rate of change of the section average heart rate and the change amount of the relative standard deviation were used. This is an experiment conducted under the following conditions. The inventors of the present application found that the rate of change in section average heart rate and the change in relative standard deviation could be used as input values.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in Learning Method Example #9 described above. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Relative standard deviation of zone heart rates corresponding to the second test day), difference between the second test day and the first test day (day gap), thyroid hormone concentration corresponding to the second test day, diagnostic result (hyperactivity disorder) corresponding to the first test day or not) were created and these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used were based on both free T4 and TSH concentrations.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #10. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Relative standard deviation of interval heart rates corresponding to the second test day), difference between the second test day and the first test day (day gap), thyroid hormone concentration corresponding to the second test day, combination value of variables, corresponding to the first test day Training data sets in the same format as diagnostic results (hyperactivity or non-hyperactivity disorder) were created, and these training data sets were used to train the model.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • hormone levels used were based on both free T4 and TSH concentrations.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • first value second value accuracy responsiveness specificity PPV NPV One Heart rate average rate of change ⁇ Second reference date TSH 82.97 87.84 82.05 48.15 97.26 2 relative standard deviation change x Heart rate skewness variation 82.54 85.14 82.05 47.37 96.68 3 Second reference date TSH x relative standard deviation change 84.48 87.84 83.85 50.78 97.32 4 Heart rate average rate of change x JS Divergence 84.48 89.19 83.59 50.77 97.60 5 Second reference date free T4 ⁇ Second reference date TSH 83.84 89.19 82.82 49.62 97.58 6 JS Divergence ⁇ Second reference date TSH 83.41 87.84 82.56 48.87 97.28 7 Heart rate kurtosis change ⁇ relative standard deviation change 83.41 89.19 82.31 48.89 97.57 8 Heart rate average rate of change ⁇ Heart rate
  • first value second value accuracy responsiveness specificity PPV NPV 40 Second reference date TSH x Heart rate kurtosis change 83.84 87.84 83.08 49.62 97.30 41 Heart rate average rate of change x Heart rate skewness variation 84.48 89.19 83.59 50.77 97.60 42 Heart rate average rate of change x Heart rate kurtosis change 84.70 89.19 83.85 51.16 97.61 43 JS Divergence ⁇ Heart rate kurtosis change 84.05 86.49 83.59 50.00 97.02 44 Second reference date free T4 ⁇ JS Divergence 85.34 90.54 84.36 52.34 97.92 45 Heart rate kurtosis change ⁇ Second reference date free T4 83.62 87.84 82.82 49.24 97.29 46 relative standard deviation change ⁇ Heart rate kurtosis change 83.41 89.19 82.31 48.89 97
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #11.
  • (1) [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), concentration of free T4 corresponding to the second test day, subject's age, subject's gender, first test day Learning was conducted using learning data sets such as (2) [((((section average heart rate corresponding to the first test day) - section average corresponding to the second test day).
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #10.
  • hormone levels used were based on both free T4 and TSH concentrations.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.
  • the inventors of the present application found that, in predicting hyperthyroidism, the highest accuracy was obtained when 15 days was used as a predetermined period for calculating zone heart rates.
  • the model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #12.
  • the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.
  • a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.
  • the clinical data used for learning was 464 data obtained from 126 patients, and the total number of learning data sets created from this was 1,416. Some of these were used as test data sets, and the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #12 was tested using the test data sets described above.
  • a hyperthyroidism prediction model was learned using the aforementioned training data sets.
  • the inventors of the present application were able to confirm that when predicting hyperthyroidism, the performance of both accuracy, sensitivity, specificity, PPV, and NPV was improved when the target date was used together with the value of the estimated thyroid hormone level. In addition, the performance improvement observed in hyperthyroidism is expected to be similar in predicting hypothyroidism.
  • the value obtained by subtracting the section average heart rate corresponding to the second test day from the section average heart rate corresponding to the first test day and the hormone concentration of the second test day It can be used as an input value, and the hormone concentration corresponding to the first test date can be used as a labeling value.
  • the learning model may be light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc.
  • the thyroid hormone concentration prediction model consists of a TRH concentration prediction model, a TSH concentration prediction model, a T4 concentration prediction model, a free T4 concentration prediction model, a T3 concentration prediction model, and a free T3 concentration prediction model on the target day. Contains at least one selected from the group.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, It can be one of the following days: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30.
  • the labeling value may be a concentration value corresponding to the type among a plurality of hormone concentrations corresponding to the first test date.
  • the labeling value when learning a free T4 concentration prediction model, the free T4 concentration value corresponding to the first inspection date is used as the labeling value.
  • the TSH concentration value corresponding to the first test date is used as the labeling value.
  • the obtained clinical data were used to calculate [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), second inspection day.
  • Learning data sets in the following format can be created: thyroid hormone concentration corresponding to , free T4 concentration corresponding to the first test date.
  • the thyroid hormone concentration corresponding to the second test day may be one selected from the group consisting of TRH concentration, TSH concentration, T4 concentration, free T4 concentration, T3 concentration, and free T3 concentration, or a combination thereof.
  • the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration
  • the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the third TSH concentration.
  • the learning data sets that can be obtained from the patient's clinical data may be as follows.
  • the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration
  • the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the learning data sets that can be obtained from the patient's clinical data are It could be like this:
  • the fourteenth embodiment disclosed by the present application compared to the above-described thirteenth embodiment, on the first examination day instead of the interval average heart rate corresponding to the first examination day minus the interval average heart rate corresponding to the second examination day
  • the change rate of the section average heart rate on the second test day is used as an input value, and the rest is the same as in the 13th embodiment.
  • the relative standard deviation of the zone heart rates corresponding to the first inspection day is subtracted from the relative standard deviation of the zone heart rates corresponding to the second inspection day.
  • the value is further used as an input value, and the rest is the same as the 14th embodiment.
  • Rate of change from heart rate on the reference day relative standard deviation change, skewness change, kurtosis change, JS Divergence and hormone value on the reference day
  • an index regarding the distribution of zone heart rates corresponding to the first inspection day and the distribution of zone heart rates corresponding to the second inspection day The value of the difference between the indicators is further used as an input value, and the rest is the same as the 15th embodiment.
  • Rate of change from heart rate on the reference day relative standard deviation change, skewness change, kurtosis change, JS Divergence, hormone value on the reference date, and Day Gap
  • the value related to the difference (day gaps) between the first inspection date and the second inspection date is further used as an input value, and the remaining Same as Example 16.
  • a value related to thyroid function corresponding to the first examination date This can be used as input value.
  • Values related to thyroid function corresponding to the first test date can be obtained based on the output value of the thyroid function classification model.
  • the value regarding thyroid function may be a value representing the thyroid function on the target date estimated based on the output value of the thyroid function classification model.
  • the value related to thyroid function may be the output value of a thyroid function classification model.
  • the value regarding thyroid function corresponding to the first test date may be one or more values.
  • a learning model can be trained using both the first value generated based on the output value of a thyroid function classification model specialized for hyperthyroidism and the second value generated based on the output value of a thyroid function classification model specialized for hypothyroidism.
  • it is not limited to this.
  • the learning model may be light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc.
  • the thyroid hormone concentration prediction model consists of a TRH concentration prediction model, a TSH concentration prediction model, a T4 concentration prediction model, a free T4 concentration prediction model, a T3 concentration prediction model, and a free T3 concentration prediction model on the target day. Contains at least one selected from the group.
  • resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, It can be one of the following days: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30.
  • the labeling value may be a concentration value corresponding to the type among a plurality of hormone concentrations corresponding to the first test date.
  • the labeling value when learning a free T4 concentration prediction model, the free T4 concentration value corresponding to the first inspection date is used as the labeling value.
  • the TSH concentration value corresponding to the first test date is used as the labeling value.
  • the obtained clinical data were used to calculate [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), first inspection day.
  • Learning data sets in the following format can be created: values for thyroid function corresponding to , thyroid hormone concentration corresponding to the second examination date, and free T4 concentration corresponding to the first examination day.
  • the thyroid hormone concentration corresponding to the second test day may be one selected from the group consisting of TRH concentration, TSH concentration, T4 concentration, free T4 concentration, T3 concentration, and free T3 concentration, or a combination thereof.
  • the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration
  • the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the third TSH concentration.
  • the learning data sets that can be obtained from the patient's clinical data may be as follows.
  • the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration
  • the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the second TSH concentration.
  • the learning data sets that can be obtained from the patient's clinical data are It could be like this:

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Pathology (AREA)
  • Cardiology (AREA)
  • Physics & Mathematics (AREA)
  • Primary Health Care (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Biophysics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Physiology (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

Disclosed in one aspect of the present application is a method by which one or more processors predict thyroid dysfunction of a subject. The method for predicting thyroid dysfunction of a subject may comprise: acquiring a trigger signal; determining a target date on the basis of the acquired trigger signal; inputting, into a first prediction model, i) a relationship between first heart rate information determined for the target date and second heart rate information determined for a reference date and ii) a reference date hormone concentration corresponding to the reference date, so as to acquire a provisional prediction result for thyroid dysfunction corresponding to the target date; inputting, into a second prediction model, i) the relationship between the first heart rate information and the second heart rate information, ii) the reference date hormone concentration and iii) the provisional prediction result, so as to acquire a result for a predicted hormone concentration corresponding to the target date; and inputting, into a third prediction model, i) the relationship between the first heart rate information and the second heart rate information, ii) the reference date hormone concentration and iv) the result for a predicted hormone concentration corresponding to the target date, so as to acquire a prediction result for thyroid dysfunction.

Description

대상에 대한 갑상선기능이상증의 예측방법 및 시스템Method and system for predicting thyroid dysfunction for a subject

본 출원은 환자의 심박수를 분석하고, 그 분석결과에 기초하여 갑상선의 기능이상에 대해 예측하는 방법 및 그 방법을 수행하기 위한 장치 및 시스템에 관한 것이다.This application relates to a method of analyzing a patient's heart rate and predicting thyroid dysfunction based on the analysis results, and a device and system for performing the method.

갑상선은 목의 앞부분에 위치하는 내분비기관으로, 뇌에 있는 뇌하수체에서 분비되는 갑상선자극호르몬의 신호를 받아 갑상선 호르몬을 만들어 내는 일을 한다. The thyroid gland is an endocrine organ located in the front of the neck. Its job is to produce thyroid hormones by receiving signals from thyroid-stimulating hormone secreted by the pituitary gland in the brain.

갑상선에서 생성되는 갑상선 호르몬은 우리 몸의 대사 속도를 조절하는 역할을 한다. 이러한 갑상선 호르몬이 정상 수준보다 더 많이 분비되면, 신체의 대사가 비정상적으로 빠르게 진행되어 평소보다 더위를 많이 느끼거나 가슴이 두근거리는 것을 느끼게 되는 등의 증상이 나타나게 된다. 반면, 이러한 갑상선 호르몬이 정상 수준보다 더 적게 분비되면, 추위를 심하게 느끼거나 피로감 및 무력감을 느끼게 되고, 맥박이 느려지는 등의 증상이 나타나게 된다.Thyroid hormones produced by the thyroid gland play a role in regulating our body's metabolic rate. When these thyroid hormones are secreted at a higher than normal level, the body's metabolism progresses abnormally quickly, causing symptoms such as feeling hotter than usual or feeling heart palpitations. On the other hand, when these thyroid hormones are secreted at a lower than normal level, symptoms such as feeling extremely cold, fatigue and helplessness, and a slow pulse appear.

갑상선의 기능이 정상적일 때에는, 갑상선 호르몬의 분비가 문제없이 이루어지지만, 갑상선의 기능에 이상이 생기게 되면, 갑상선 호르몬의 분비가 줄어들거나 늘어나게 되는데, 갑상선 호르몬의 분비가 줄어들어서 생기는 일련의 질환들을 갑상선기능저하증(hypothyroidism)이라고 하며, 갑상선 호르몬의 분비가 늘어서 생기는 일련의 질환들을 갑상선기능항진증(hyperthyroidism)이라고 한다. 아울러, 이 두 질환을 갑상선기능이상증(thyroid dysfunction)이라고 한다.When the function of the thyroid gland is normal, the secretion of thyroid hormones occurs without a problem, but when there is an abnormality in the function of the thyroid gland, the secretion of thyroid hormones decreases or increases. A series of diseases caused by decreased secretion of thyroid hormones are caused by the thyroid gland. It is called hypothyroidism, and a series of diseases caused by increased secretion of thyroid hormones are called hyperthyroidism. Additionally, these two diseases are called thyroid dysfunction.

현재, 갑상선의 기능의 이상을 확인하기 위한 유일한 진단 방법은 환자의 혈액속에 있는 갑상선 호르몬의 농도를 확인하여, 갑상선 호르몬의 농도가 정상범위 이내인지, 아니면 높은 수준인지 혹은 낮은 수준인지를 임상병리학적으로 확인하는 것이 유일한 방법이다.Currently, the only diagnostic method to confirm abnormalities in thyroid function is to check the concentration of thyroid hormones in the patient's blood and determine whether the concentration of thyroid hormones is within the normal range, high level, or low level using clinical pathology. The only way is to check.

한편, 갑상선기능항진증과 갑상선기능저하증을 치료하는 데에는 여러가지 방법들이 있지만, 가장 널리 사용되고 있는 치료방법은 약물치료방법이다. 갑상선항진증을 겪고 있는 환자들과 갑상선기능저하증을 겪고 있는 환자들은 대부분 일정 기간 동안의 약물 복용을 통해 정상 수준의 갑상선의 호르몬 생성 및 분비 기능을 되찾게 된다. 그러나, 갑상선기능항진증과 갑상선기능저하증은 완치율이 높지 않고 재발율이 매우 높은 질환으로 갑상선기능항진증 또는 갑상선기능저하증이 재발하는지 여부에 대해서 지속적으로 모니터링을 할 필요가 있다.Meanwhile, there are various methods to treat hyperthyroidism and hypothyroidism, but the most widely used treatment method is drug treatment. Most patients suffering from hyperthyroidism and patients suffering from hypothyroidism regain normal levels of thyroid hormone production and secretion function by taking medication for a certain period of time. However, hyperthyroidism and hypothyroidism are diseases that do not have a high cure rate and have a very high recurrence rate, so there is a need to continuously monitor whether hyperthyroidism or hypothyroidism recurs.

그러나, 전술한 바와 같이, 갑상선의 기능을 확인하기 위한 유일한 진단방법은 채혈을 통한 호르몬의 농도를 확인하는 방법이기 때문에, 갑상선의 기능 이상 여부를 모니터링하기 위해서는 환자가 주기적으로 병원에 방문하여야 하는 문제가 있다.However, as mentioned above, the only diagnostic method to check the function of the thyroid gland is to check the concentration of hormones through blood sampling, so the patient must visit the hospital periodically to monitor whether the thyroid function is abnormal. There is.

이에, 채혈을 통한 호르몬 수치의 확인 이외의 방법으로 보다 더 정확하게 갑상선의 기능에 이상이 있는지 여부를 확인할 수 있는 기술의 개발이 요구되고 있다.Accordingly, there is a need for the development of technology that can more accurately confirm whether there is an abnormality in thyroid function by methods other than confirmation of hormone levels through blood collection.

즉, 환자들이, 병원 직접 방문 없이도, 개개인이 보다 간편하고 신속하게 갑상선의 기능 이상에 관한 모니터링을 할 수 있도록 하며, 필요한 경우 환자의 내원을 유도하는 방법의 개발이 요구되고 있다.In other words, there is a need to develop a method that allows patients to monitor thyroid dysfunction more easily and quickly without having to visit a hospital in person, and to encourage patients to visit the hospital when necessary.

본 출원에 의해 개시되는 내용들이 해결하고자 하는 과제는 전문적인 의학 진단 기기가 아닌 일반인들이 사용할 수 있는 개인용 전자기기(personal electronic device)를 이용하여 획득된 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측할 수 있는 예측 모델을 제공하는 것이다.The problem to be solved by the contents disclosed by this application is to predict whether thyroid function is abnormal based on heart rate information obtained using a personal electronic device that can be used by the general public rather than a professional medical diagnostic device. It provides a predictive model that can be predicted.

본 출원에 의해 개시되는 내용들이 해결하고자 하는 다른 과제는 전술한 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측할 수 있는 예측 모델의 학습 방법을 제공하는 것이다.Another problem to be solved by the contents disclosed by the present application is to provide a method of learning a prediction model that can predict whether thyroid function is abnormal based on the above-described heart rate information.

본 출원에 의해 개시되는 내용들이 해결하고자 하는 또 다른 과제는 전술한 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측하는 모델을 이용하여 일반인들이, 의사의 도움없이 그리고 병원 직접 방문 없이, 지속적으로 갑상선의 기능 이상에 대해 모니터링할 수 있는 방법 및 시스템을 제공하는 것이다.Another problem to be solved by the contents disclosed by this application is to use a model to predict whether thyroid function is abnormal based on the above-mentioned heart rate information, so that the general public can continuously monitor the thyroid gland without the help of a doctor and without visiting the hospital directly. The purpose is to provide a method and system for monitoring malfunctions.

본 출원이 해결하고자 하는 과제가 상술한 과제로 제한되는 것은 아니며, 언급되지 아니한 과제들은 본 명세서 및 첨부된 도면으로부터 본 출원에 의해 개시되는 기술이 속하는 기술분야에서 통상의 지식을 가진 자에게 명확하게 이해될 수 있을 것이다.The problem to be solved by this application is not limited to the above-mentioned problems, and problems not mentioned are clearly apparent to those skilled in the art from this specification and the attached drawings to which the technology disclosed by this application belongs. It will be understandable.

본 출원의 일 양태에 의하면, 하나 이상의 프로세서에 의해 수행되는 대상에 대한 갑상선기능이상증의 예측 방법이 개시된다. 상기 대상에 대한 갑상선기능이상증의 예측방법은, 트리거 신호를 획득함; 상기 획득된 트리거 신호에 기초하여 결정된 대상일을 결정함; i) 대상일에 대해 결정되는 제1 심박수 정보와 기준일에 대해 결정되는 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 제1 예측모델에 입력하여 상기 대상일에 대응되는 갑상선기능이상에 대한 가예측 결과를 획득함-상기 제1 심박수 정보 및 상기 제2 심박수 정보는 상기 대상에 관한 심박수 정보임-; i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iii) 상기 가예측 결과를 제2 예측모델에 입력하여 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 획득함; 및 i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iv) 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 제3 예측모델에 입력하여 갑상선기능이상에 대한 예측 결과를 획득함;을 포함할 수 있다.According to one aspect of the present application, a method for predicting dysthyroidism for a subject performed by one or more processors is disclosed. The method for predicting thyroid dysfunction for the subject includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; i) the relationship between the first heart rate information determined for the target date and the second heart rate information determined for the reference date, and ii) the thyroid hormone concentration corresponding to the target date by inputting the reference day hormone concentration corresponding to the reference date into the first prediction model Obtaining a tentative prediction result for dysfunction - the first heart rate information and the second heart rate information are heart rate information about the subject; i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the tentative prediction result is input into a second prediction model to produce a result for the predicted hormone concentration corresponding to the target date. acquired; And i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iv) the results of the predicted hormone concentration corresponding to the target date are input into a third prediction model to determine thyroid dysfunction. It may include obtaining a prediction result for.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 가예측 결과를 획득함에 있어서, 상기 대상일에 대응되는 가예측 호르몬 농도에 대한 결과가 더 입력되고, 상기 가예측 호르몬 농도는, i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계 및 ii) 상기 기준일 호르몬 농도를 제4 예측모델에 입력하여 획득될 수 있다.In some embodiments disclosed by the present application, in obtaining the tentative prediction result, a result for the tentative predicted hormone concentration corresponding to the target date is further input, and the tentative predicted hormone concentration is: i) the It can be obtained by inputting the relationship between the first heart rate information and the second heart rate information and ii) the hormone concentration on the reference day into the fourth prediction model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 예측 호르몬 농도에 대한 결과는 대상일에 대응되는 갑상선 호르몬 농도 예측값일 수 있다.In some embodiments disclosed by the present application, the result of the predicted hormone concentration may be a predicted thyroid hormone concentration value corresponding to the target date.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 예측 호르몬 농도에 대한 결과는 대상일의 갑상선 호르몬 농도와 상기 기준일 호르몬 농도 사이의 관계에 대응될 수 있다.In some embodiments disclosed by the present application, the result for the predicted hormone concentration may correspond to the relationship between the thyroid hormone concentration on the target day and the hormone concentration on the reference day.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 관계는, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량 및 7) 상기 기준일에 대응되는 구간 심박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence로 이루어진 그룹에서 선택된 하나 또는 그들의 조합일 수 있다.In some embodiments disclosed by the present application, the relationship is: 1) the amount of change in the average of section heart rates of the target day with respect to the reference date, 2) the rate of change of the average of section heart rates of the target day with respect to the reference date , 3) the amount of change in the standard deviation of the target date with respect to the reference date, 4) the amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) the above It may be one or a combination thereof selected from the group consisting of the amount of change in kurtosis of the target date with respect to the reference date and 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 예측 호르몬 농도는, 1) 상기 대상의 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 2) 상기 대상의 사요오드티로닌(tetraiodothyronine, T4)의 농도, 3) 상기 대상의 혈청에 유리된 T4 (free T4)의 농도, 4) 상기 대상의 삼요오드티로닌(triiodothyronine, T3)의 농도, 5) 상기 대상의 혈청에 유리된 T3 (free T3)의 농도 및 6) 상기 대상의 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 그들의 조합에 대응될 수 있다.In some embodiments disclosed by the present application, the predicted hormone concentration is 1) the concentration of the subject's thyroid stimulating hormone (TSH), 2) the subject's tetraiodothyronine (T4) ) concentration, 3) the concentration of free T4 (free T4) in the serum of the subject, 4) the concentration of triiodothyronine (T3) in the subject, 5) the concentration of free T3 (free T4) in the serum of the subject It may correspond to one or a combination thereof selected from the group consisting of the concentration of T3) and 6) the concentration of Thyrotropin-Releasing Hormone (TRH) of the subject.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제1 예측모델에는, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량, 7) 상기 기준일에 대응되는 구간 심박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence, 8) 상기 기준일에 대응되는 상기 대상의 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 9) 상기 기준일에 대응되는 상기 대상의 사요오드티로닌(tetraiodothyronine, T4)의 농도, 10) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T4 (free T4)의 농도, 11) 상기 기준일에 대응되는 상기 대상의 삼요오드티로닌(triiodothyronine, T3)의 농도, 12) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T3 (free T3)의 농도 및 13) 상기 기준일에 대응되는 상기 대상의 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 제1 값 및 제2 값을 곱하거나 나눈 조합 값이 더 입력될 수 있다.In some embodiments disclosed by the present application, the first prediction model includes 1) the change in the average of the heart rates of the target day on the target day with respect to the reference date, 2) the heart rates of the section on the target day with respect to the reference date. The rate of change in the average, 3) the amount of change in the standard deviation of the target date with respect to the reference date, 4) the amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) The amount of change in kurtosis of the target date with respect to the reference date, 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date, 8) Thyroid stimulation of the subject corresponding to the reference date Concentration of thyroid stimulating hormone (TSH), 9) Concentration of tetraiodothyronine (T4) in the subject corresponding to the reference date, 10) Free T4 (free) in the serum of the subject corresponding to the reference date Concentration of T4), 11) Concentration of triiodothyronine (T3) of the subject corresponding to the reference date, 12) Concentration of free T3 (free T3) in the serum of the subject corresponding to the reference date, and 13 ) A combination value obtained by multiplying or dividing the first value and the second value selected from the group consisting of the concentration of Thyrotropin-Releasing Hormone (TRH) of the subject corresponding to the reference date may be further input.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제2 예측모델에는 상기 조합 값이 입력되지 않고, 상기 제3 예측모델에는 상기 조합 값이 더 입력될 수 있다.In some embodiments disclosed by the present application, the combination value may not be input to the second prediction model, and the combination value may be further input to the third prediction model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 가예측 결과를 획득함에 있어서, 상기 대상일에 대응되는 구간심박수들의 표준편차, 왜도, 첨도 및 상기 대상일과 상기 기준일 사이의 일차(day gap)로 이루어진 그룹에서 선택된 하나 또는 그들의 조합이 더 입력될 수 있다.In some embodiments disclosed by the present application, in obtaining the tentative prediction result, the standard deviation, skewness, kurtosis, and difference between the target date and the reference date of the section heart rates corresponding to the target date (day gap) ) or a combination thereof may be further entered.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 대상일에 대응되는 구간 심박수들은 상기 대상일을 기준으로 미리 결정된 기간에 대응되는 모든 휴지기 심박수(resting heart rate)이고, 상기 기준일에 대응되는 구간 심박수들은 상기 기준일을 기준으로 상기 미리 결정된 기간에 대응되는 모든 휴지기 심박수일 수 있다.In some embodiments disclosed by the present application, the section heart rates corresponding to the target date are all resting heart rates corresponding to a predetermined period based on the target date, and the section corresponding to the reference date The heart rates may be all resting heart rates corresponding to the predetermined period based on the reference date.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제1 예측모델은 갑상선 기능 항진증 또는 갑상선 기능 항진증 아님으로 분류하는 모델이고, 상기 제3 예측모델은 갑상선 기능 항진증 또는 갑상선 기능 항진증 아님으로 분류하는 모델일 수 있다.In some embodiments disclosed by the present application, the first prediction model is a model that classifies hyperthyroidism or not hyperthyroidism, and the third prediction model is a model that classifies hyperthyroidism or not hyperthyroidism. It could be a model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제2 예측모델은 갑상선 기능 항진증으로 확인된 환자들의 심박수 정보 및 그때의 호르몬 수치를 학습 데이터 세트로 하여 학습된 모델일 수 있다.In some embodiments disclosed by the present application, the second prediction model may be a model learned using the heart rate information and hormone levels of patients confirmed to have hyperthyroidism as a learning data set.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제1 예측모델은 갑상선 기능 항진증, 정상 또는 갑상선 기능 저하증으로 분류하는 모델이고, 상기 제3 예측모델은 갑상선 기능 항진증, 정상 또는 갑상선 기능 저하증으로 분류하는 모델일 수 있다.In some embodiments disclosed by the present application, the first prediction model is a model that classifies hyperthyroidism, normal function, or hypothyroidism, and the third prediction model is a model that classifies hyperthyroidism, normal function, or hypothyroidism. It may be a classification model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제1 예측모델은 갑상선 기능 항진증으로 확인된 환자들의 심박수 정보 및 그때의 갑상선 기능 이상 여부를 학습 데이터 세트로 하여 학습된 모델이고, 갑상선 기능 저하증으로 확인된 환자들의 심박수 정보 및 그때의 갑상선 기능 이상 여부를 학습 데이터 세트로 하여 학습된 제5 예측 모델은 i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 입력받아 갑상선기능이상에 대한 가예측 결과를 획득하고, 상기 제2 예측 모델은 상기 제5 예측 모델에서 출력된 상기 가예측 결과를 더 입력받아 예측 호르몬 농도에 대한 결과를 획득할 수 있다.In some embodiments disclosed by the present application, the first prediction model is a model learned using as a learning data set the heart rate information of patients confirmed to have hyperthyroidism and whether there is an abnormality in thyroid function at that time, and the first prediction model is a model learned with hypothyroidism The fifth prediction model learned using the heart rate information of the patients identified as a learning data set and the presence or absence of thyroid function abnormalities at that time is i) the relationship between the first heart rate information and the second heart rate information, and ii) corresponding to the reference date. The hormone concentration on the reference day is input to obtain a tentative prediction result for thyroid dysfunction, and the second prediction model can further receive the tentative prediction result output from the fifth prediction model to obtain a result for the predicted hormone concentration. there is.

본 출원의 일 양태에 따르면 하나 이상의 프로세서에 의해 수행되는 대상에 대한 갑상선기능이상증의 예측 방법이 개시된다. 상기 대상에 대한 갑상선기능이상증 예측방법은, 트리거 신호를 획득함; 상기 획득된 트리거 신호에 기초하여 결정된 대상일을 결정함; i) 대상일에 대해 결정되는 대상일 심박수에 관한 제1 심박수 정보와 기준일에 대해 결정되는 기준일 심박수에 관한 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 제1 예측모델에 입력하여 상 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 획득함-상기 제1 심박수 정보 및 상기 제2 심박수 정보는 상기 대상에 관한 심박수 정보임-; i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iii) 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 제2 예측모델에 입력하여 갑상선기능이상에 대한 예측 결과를 획득함;을 포함할 수 있다.According to one aspect of the present application, a method for predicting dysthyroidism for a subject performed by one or more processors is disclosed. The method for predicting thyroid dysfunction for the subject includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; i) the relationship between the first heart rate information regarding the target date heart rate determined for the target date and the second heart rate information regarding the reference date heart rate determined for the reference date, and ii) the reference day hormone concentration corresponding to the reference date to the first prediction model Input to obtain a result of the predicted hormone concentration corresponding to the target date - the first heart rate information and the second heart rate information are heart rate information related to the target; i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the results of the predicted hormone concentration corresponding to the target date are input into the second prediction model to determine thyroid dysfunction. Obtaining a prediction result may include.

본 출원에 의해 개시되는 내용들에 의하여, 전문적인 의학 진단 기기가 아닌 일반인들이 사용할 수 있는 개인용 전자기기(personal electronic device)를 이용하여 획득된 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측할 수 있는 예측 모델을 제공될 수 있다.According to the contents disclosed by this application, it is possible to predict whether there is an abnormality in thyroid function based on heart rate information obtained using a personal electronic device that can be used by the general public rather than a professional medical diagnostic device. A prediction model may be provided.

본 출원에 의해 개시되는 내용들에 의하여, 전술한 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측할 수 있는 예측 모델의 학습 방법이 제공된다.According to the contents disclosed by the present application, a method of learning a prediction model that can predict whether the thyroid function is abnormal based on the above-described heart rate information is provided.

본 출원에 의해 개시되는 내용들에 의하여, 전술한 심박수 정보에 기초하여 갑상선의 기능 이상 여부를 예측하는 모델을 이용하여 일반인들이, 의사의 도움없이 그리고 병원 직접 방문 없이, 지속적으로 갑상선의 기능 이상에 대해 모니터링할 수 있는 방법 및 시스템을 제공된다.According to the contents disclosed by the present application, the general public can continuously detect abnormal thyroid function without the help of a doctor and without visiting the hospital directly, using a model that predicts abnormality of thyroid function based on the above-mentioned heart rate information. A method and system for monitoring is provided.

도 1은 본 출원에 의해 개시되는 일실시예에 따른 갑상선기능이상증을 예측하기 위한 시스템을 도시하는 도면이다.1 is a diagram illustrating a system for predicting thyroid dysfunction according to an embodiment disclosed by the present application.

도 2는 본 출원에 의해 개시되는 심박수 측정 디바이스의 블록도이다.Figure 2 is a block diagram of a heart rate measurement device disclosed by the present application.

도 3은 본 출원에 의해 개시되는 사용자 단말기의 블록도이다.Figure 3 is a block diagram of a user terminal disclosed by the present application.

도 4는 본 출원에 의해 개시되는 서버의 블록도이다.Figure 4 is a block diagram of the server disclosed by this application.

도 5는 본 출원에 의해 개시되는 갑상선기능이상증을 예측하는 방법을 설명하기 위한 흐름도이다.Figure 5 is a flow chart to explain the method for predicting thyroid dysfunction disclosed by the present application.

도 6은 본 출원에 의해 개시되는 갑상선기능이상증을 예측하는 방법을 설명하기 위한 흐름도이다.Figure 6 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.

도 7은 갑상선 기능 이상 예측 모듈을 설명하기 위한 블록도이다.Figure 7 is a block diagram for explaining the thyroid dysfunction prediction module.

도 8 및 도 9는 학습시킬 때의 임상데이터를 선별함으로써 항진증, 정상 및/또는 저하증에 특화된 갑상선 기능 분류모델 및/또는 갑상선호르몬농도 예측모델을 포함하는 갑상선 기능 이상 예측 모듈을 설명하기 위한 블록도이다.Figures 8 and 9 are block diagrams for illustrating a thyroid function abnormality prediction module including a thyroid function classification model and/or thyroid hormone concentration prediction model specialized for hyperactivity, normal and/or hypothyroidism by selecting clinical data during learning. am.

도 10 및 도 11은 갑상선기능이상증을 예측하는 방법을 설명하기 위한 도면이다.Figures 10 and 11 are diagrams for explaining a method for predicting thyroid dysfunction.

도 12는 본 출원에 의해 개시되는 갑상선기능이상증을 예측하는 방법을 설명하기 위한 흐름도이다.Figure 12 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.

--

본 출원의 상술한 목적, 특징들 및 장점은 첨부된 도면과 관련된 다음의 상세한 설명을 통해 보다 분명해질 것이다. 다만, 본 출원은 다양한 변경을 가할 수 있고 여러 가지 실시예들을 가질 수 있는 바, 이하에서는 특정 실시예들을 도면에 예시하고 이를 상세히 설명하고자 한다.The above-described objects, features and advantages of the present application will become more apparent through the following detailed description in conjunction with the accompanying drawings. However, since the present application can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail below.

명세서 전체에 걸쳐서 동일한 참조번호들은 원칙적으로 동일한 구성요소들을 나타낸다. 또한, 각 실시예의 도면에 나타나는 동일한 사상의 범위 내의 기능이 동일한 구성요소는 동일한 참조부호를 사용하여 설명하며, 이에 대한 중복되는 설명은 생략하기로 한다.Like reference numerals throughout the specification in principle refer to the same elements. In addition, components with the same function within the scope of the same idea shown in the drawings of each embodiment will be described using the same reference numerals, and overlapping descriptions thereof will be omitted.

본 출원과 관련된 공지 기능 혹은 구성에 대한 구체적인 설명이 본 출원의 요지를 불필요하게 흐릴 수 있다고 판단되는 경우 그 상세한 설명을 생략한다. 또한, 본 명세서의 설명 과정에서 이용되는 숫자(예를 들어, 제1, 제2 등)는 하나의 구성요소를 다른 구성요소와 구분하기 위한 식별기호에 불과하다.If it is determined that a detailed description of a known function or configuration related to the present application may unnecessarily obscure the gist of the present application, the detailed description will be omitted. In addition, numbers (eg, first, second, etc.) used in the description of this specification are merely identifiers to distinguish one component from another component.

또한, 이하의 실시예에서 사용되는 구성요소에 대한 접미사 "모듈" 및 "부"는 명세서 작성의 용이함만이 고려되어 부여되거나 혼용되는 것으로서, 그 자체로 서로 구별되는 의미 또는 역할을 갖는 것은 아니다.In addition, the suffixes “module” and “part” for components used in the following examples are given or used interchangeably only for the ease of writing the specification, and do not have distinct meanings or roles in themselves.

이하의 실시예에서, 단수의 표현은 문맥상 명백하게 다르게 뜻하지 않는 한, 복수의 표현을 포함한다.In the following examples, singular terms include plural terms unless the context clearly dictates otherwise.

이하의 실시예에서, 포함하다 또는 가지다 등의 용어는 명세서상에 기재된 특징, 또는 구성요소가 존재함을 의미하는 것이고, 하나 이상의 다른 특징들 또는 구성요소가 부가될 가능성을 미리 배제하는 것은 아니다.In the following embodiments, terms such as include or have mean the presence of features or components described in the specification, and do not exclude in advance the possibility of adding one or more other features or components.

도면에서는 설명의 편의를 위하여 구성 요소들이 그 크기가 과장 또는 축소될 수 있다. 예컨대, 도면에서 나타난 각 구성의 크기 및 두께는 설명의 편의를 위해 임의로 나타낸 것으로, 본 발명이 반드시 도시된 바에 한정되지 않는다.In the drawings, the sizes of components may be exaggerated or reduced for convenience of explanation. For example, the size and thickness of each component shown in the drawings are arbitrarily shown for convenience of explanation, and the present invention is not necessarily limited to what is shown.

어떤 실시예가 달리 구현 가능한 경우에 특정한 프로세스의 순서는 설명되는 순서와 다르게 수행될 수도 있다. 예를 들어, 연속하여 설명되는 두 프로세스가 실질적으로 동시에 수행될 수도 있고, 설명되는 순서와 반대의 순서로 진행될 수 있다.If an embodiment can be implemented differently, the order of specific processes may be performed differently from the order described. For example, two processes described in succession may be performed substantially simultaneously, or may proceed in an order opposite to that in which they are described.

이하의 실시예에서, 구성 요소 등이 연결되었다고 할 때, 구성 요소들이 직접적으로 연결된 경우뿐만 아니라 구성요소들 중간에 구성 요소들이 개재되어 간접적으로 연결된 경우도 포함한다. 예컨대, 본 명세서에서 구성 요소 등이 전기적으로 연결되었다고 할 때, 구성 요소 등이 직접 전기적으로 연결된 경우뿐만 아니라, 그 중간에 구성 요소 등이 개재되어 간접적으로 전기적 연결된 경우도 포함한다.In the following embodiments, when components are connected, this includes not only the case where the components are directly connected, but also the case where the components are indirectly connected by intervening between the components. For example, in this specification, when components, etc. are said to be electrically connected, this includes not only cases where the components are directly electrically connected, but also cases where components, etc. are interposed and indirectly electrically connected.

본 출원의 일 양태에 의하면, 대상(subject)에 대한 갑상선기능이상증(thyroid dysfunction)의 예측방법이 개시된다. 상기 대상에 대한 갑상선기능이상증의 예측방법은, 트리거 신호를 획득함; 상기 획득된 트리거 신호에 기초하여 결정된 대상일(target date)을 결정함; 상기 결정된 대상일에 대응되는 구간 심박수들을 획득함; 상기 대상에 대하여, 상기 획득된 대상일에 대응되는 구간 심박수들에 대한 제1 전처리 결과를 획득함 - 이때, 상기 제1 전처리 결과는 상기 대상일에 대응되는 구간 심박수들의 평균 및 분포(distribution)에 관한 적어도 하나의 지표를 포함함 - ; 상기 대상에 대하여, 기준일에 대응되는 갑상선에 관련된 적어도 하나 이상의 호르몬 농도를 획득함; 상기 대상에 대하여, 상기 기준일에 대응되는 구간 심박수들에 대한 제2 전처리 결과를 획득함 - 이때, 상기 제2 전처리 결과는 상기 기준일에 대응되는 구간 심박수들의 평균 및 분포(distribution)에 관한 적어도 하나의 지표를 포함함 - ; 상기 제2 전처리 결과에 대한 상기 제1 전처리 결과의 차이를 획득함; 및 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 값들에 기초하여 획득된 갑상선기능이상증에 대한 예측결과를 획득함;을 포함한다. 이때, 상기 갑상선기능이상증에 대한 예측결과는 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 입력값들이 갑상선기능이상 예측 모델에 의해 처리된 결과일 수 있다.According to one aspect of the present application, a method for predicting thyroid dysfunction for a subject is disclosed. The method for predicting thyroid dysfunction for the subject includes obtaining a trigger signal; determining a target date determined based on the obtained trigger signal; Obtaining zone heart rates corresponding to the determined target date; For the object, a first preprocessing result is obtained for the zone heart rates corresponding to the acquired target date - At this time, the first preprocessing result is the average and distribution of the zone heart rates corresponding to the target day. - Contains at least one indicator about ; For the subject, obtain the concentration of at least one hormone related to the thyroid gland corresponding to the reference date; For the object, a second preprocessing result for the section heart rates corresponding to the reference date is obtained - At this time, the second preprocessing result is at least one information about the average and distribution of the section heart rates corresponding to the reference date. Contains indicators - ; Obtaining a difference between the first preprocessing result and the second preprocessing result; and obtaining a prediction result for thyroid dysfunction obtained based on values including the difference and at least one hormone concentration corresponding to the reference date. At this time, the prediction result for thyroid dysfunction may be a result of processing input values including the difference and at least one hormone concentration corresponding to the reference date by a thyroid dysfunction prediction model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 적어도 하나의 지표는 상기 대상일에 대응되는 구간 심박수들의 표준편차, 왜도 및 첨도를 포함하고, 상기 기준일에 대응되는 구간 심박수들의 분포에 관한 적어도 하나의 지표는 상기 기준일에 대응되는 구간 심박수들의 표준편차, 왜도 및 첨도를 포함할 수 있다.In some embodiments disclosed by the present application, at least one indicator regarding the distribution of zone heart rates corresponding to the target date includes standard deviation, skewness, and kurtosis of zone heart rates corresponding to the target date, At least one indicator regarding the distribution of section heart rates corresponding to the reference date may include standard deviation, skewness, and kurtosis of section heart rates corresponding to the reference date.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 제2 전처리 결과에 대한 상기 제1 전처리 결과의 차이는, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량 및 7) 상기 기준일에 대응되는 구간 삼박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence로 이루어진 그룹에서 선택된 하나 또는 그들의 조합일 수 있다.In some embodiments disclosed by the present application, the difference between the first preprocessing result and the second preprocessing result is 1) the amount of change in the average of the heart rates of the intervals of the target day with respect to the reference date, 2) the reference date , 3) the change in the standard deviation of the target day with respect to the reference date, 4) the change in the relative standard deviation of the target date with respect to the reference date, 5) with respect to the reference date. A group consisting of the amount of change in skewness of the target date, 6) the amount of change in kurtosis of the target date with respect to the reference date, and 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date. It may be one selected from or a combination thereof.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도는 1) 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 2) 사요오드티로닌(tetraiodothyronine, T4)의 농도, 3) 혈청에 유리된 T4 (free T4)의 농도, 4) 삼요오드티로닌(triiodothyronine, T3)의 농도, 5) 혈청에 유리된 T3 (free T3)의 농도 및 6) 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수 있다.In some embodiments disclosed by the present application, the concentration of at least one hormone corresponding to the reference date is 1) the concentration of thyroid stimulating hormone (TSH), 2) tetraiodothyronine (T4) concentration, 3) concentration of free T4 (free T4) in serum, 4) concentration of triiodothyronine (T3), 5) concentration of free T3 (free T3) in serum, and 6) thyroid-stimulating hormone. -It may be one selected from the group consisting of the concentration of Thyrotropin-Releasing Hormone (TRH) or a combination thereof.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 대상에 대한 갑상선기능이상증의 예측방법은, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량, 7) 상기 기준일에 대응되는 구간 삼박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence, 8) 상기 기준일에 대응되는 상기 대상의 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 9) 상기 기준일에 대응되는 상기 대상의 사요오드티로닌(tetraiodothyronine, T4)의 농도, 10) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T4 (free T4)의 농도, 11) 상기 기준일에 대응되는 상기 대상의 삼요오드티로닌(triiodothyronine, T3)의 농도, 12) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T3 (free T3)의 농도 및 13) 상기 기준일에 대응되는 상기 대상의 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 제1 값 및 제2 값을 획득함; 및 상기 i) 상기 제1 값과 상기 제2 값을 곱한 값, ii) 상기 제1 값을 상기 제2 값으로 나눈 값 및 iii) 상기 제2 값을 상기 제1값으로 나눈 값 중 하나를 획득함;을 더 포함할 수 있다. In some embodiments disclosed by the present application, the method for predicting thyroid dysfunction for the subject includes: 1) the amount of change in the average of the heart rates of the section of the subject day with respect to the reference date, 2) the subject with respect to the reference date The rate of change in the average of the daily heart rates, 3) the amount of change in the standard deviation of the target day with respect to the reference date, 4) the amount of change in the relative standard deviation of the target day with respect to the reference date, 5) the amount of change in the target day with respect to the reference date Amount of change in skewness, 6) Amount of change in kurtosis of the target date with respect to the reference date, 7) JS Divergence between section heart rates corresponding to the reference date and section heart rates corresponding to the target date, 8) Corresponding to the reference date concentration of thyroid stimulating hormone (TSH) in the subject, 9) concentration of tetraiodothyronine (T4) in the subject corresponding to the reference date, 10) serum of the subject corresponding to the reference date Concentration of free T4 (free T4), 11) Concentration of triiodothyronine (T3) in the subject corresponding to the reference date, 12) Free T3 (free) in the serum of the subject corresponding to the reference date Obtaining the first and second values from the group consisting of the concentration of T3) and 13) the concentration of Thyrotropin-Releasing Hormone (TRH) of the subject corresponding to the reference date; and obtain one of i) a value obtained by multiplying the first value and the second value, ii) a value obtained by dividing the first value by the second value, and iii) a value obtained by dividing the second value by the first value. It may further include;

이때, 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 값들에 기초하여 획득된 갑상선기능이상증에 대한 예측결과를 획득함은, 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도, 상기 차이 및 상기 획득된 하나의 값을 포함하는 값들에 기초하여 획득된 갑상선기능이상증에 대한 예측결과를 획득하는 것이고, 상기 갑상선기능이상증에 대한 예측결과는 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도, 상기 차이 및 상기 획득된 하나의 값을 포함하는 입력값들이 갑상선기능이상 예측 모델에 의해 처리된 결과일 수 있다.At this time, obtaining a prediction result for thyroid dysfunction obtained based on values including at least one hormone concentration corresponding to the reference date and the difference includes at least one hormone concentration corresponding to the reference date, the difference, and A prediction result for thyroid dysfunction is obtained based on values including the obtained single value, and the prediction result for thyroid dysfunction is the concentration of at least one hormone corresponding to the reference date, the difference, and Input values including the obtained single value may be the result of processing by a thyroid dysfunction prediction model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 대상에 대한 갑상선기능이상증의 예측방법은, 상기 대상일과 상기 기준일 사이의 차이(day gap)를 획득함;을 더 포함할 수 있다.In some embodiments disclosed by the present application, the method for predicting thyroid dysfunction for the subject may further include obtaining a difference (day gap) between the subject date and the reference date.

이때, 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 값들에 기초하여 획득된 갑상선기능이상증에 대한 예측결과를 획득함은, 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도, 상기 차이 및 상기 대상일과 상기 기준일 사이의 차이(day gap)를 포함하는 값들에 기초하여 획득된 갑상선기능이상증에 대한 예측결과를 획득하는 것이고, 상기 갑상선기능이상증에 대한 예측결과는 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도, 상기 차이 및 상기 대상일과 상기 기준일 사이의 차이를 포함하는 입력값들이 갑상선기능이상 예측 모델에 의해 처리된 결과일 수 있다.At this time, obtaining a prediction result for thyroid dysfunction obtained based on values including at least one hormone concentration corresponding to the reference date and the difference includes at least one hormone concentration corresponding to the reference date, the difference, and A prediction result for thyroid dysfunction is obtained based on values including the difference (day gap) between the target date and the reference date, and the prediction result for thyroid dysfunction is at least one corresponding to the reference date. Input values including hormone concentration, the difference, and the difference between the target date and the reference date may be the result of being processed by a thyroid dysfunction prediction model.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 대상일에 대응되는 구간 심박수들은 상기 대상일을 기준으로 미리 결정된 기간에 대응되는 모든 휴지기 심박수(resting heart rate)이고, 상기 기준일에 대응되는 구간 심박수들은 상기 기준일을 기준으로 상기 미리 결정된 기간에 대응되는 모든 휴지기 심박수일 수 있다.In some embodiments disclosed by the present application, the section heart rates corresponding to the target date are all resting heart rates corresponding to a predetermined period based on the target date, and the section corresponding to the reference date The heart rates may be all resting heart rates corresponding to the predetermined period based on the reference date.

이때, 상기 미리 결정된 기간은 8, 9, 10, 11, 12, 13, 14, 15, 16 및 17일 중에서 선택된 어느 하나일 수 있다.At this time, the predetermined period may be any one selected from 8, 9, 10, 11, 12, 13, 14, 15, 16, and 17 days.

본 출원에 의해 개시되는 몇몇 실시예들에 있어서, 상기 갑상선기능이상증 예측모델은 갑상선기능항진증 예측모델 및 갑상선기능저하증 예측모델을 포함할 수 있다.In some embodiments disclosed by the present application, the dysthyroidism prediction model may include a hyperthyroidism prediction model and a hypothyroidism prediction model.

이때, 상기 갑상선기능이상증에 대한 예측결과는 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 입력값들이 상기 갑상선기능항진증 예측모델에 의해 처리된 제1 예측결과와 상기 기준일에 대응되는 적어도 하나 이상의 호르몬 농도 및 상기 차이를 포함하는 입력값들이 상기 갑상선기능저하증 예측모델에 의해 처리된 제2 예측결과를 고려하여 결정될 수 있다.At this time, the prediction result for dysthyroidism is the first prediction result processed by the hyperthyroidism prediction model and the input values including the difference and at least one hormone concentration corresponding to the reference date and the reference date. Input values including at least one hormone concentration and the difference may be determined by considering the second prediction result processed by the hypothyroidism prediction model.

1. 용어의 정의1. Definition of terms

심박수(heart rate)heart rate

본 명세서에서, 용어 "심박수(heart rate)"는 단위시간당 심장박동 수를 의미한다. 예를 들어, 심박수는 분당 심장박동수(beats per minute, bpm)로 측정될 수 있다. 한편, 본 명세서에서, 특별한 언급이 있거나 그 용어를 일반적인 과학적 설명을 목적으로 사용하는 경우가 아니라면, "심박수"는 그 심박수가 측정된 시점(estimation time point)과 대응되거나 연계되어 있는 것으로 본다.As used herein, the term “heart rate” refers to the number of heart beats per unit time. For example, heart rate may be measured in beats per minute (bpm). Meanwhile, in this specification, unless specifically mentioned or the term is used for the purpose of general scientific explanation, “heart rate” is considered to correspond to or be linked to the estimation time point at which the heart rate was measured.

휴지기(resting period)resting period

본 명세서에서, 용어 "휴지기(resting period)"는 사람의 신체적인 움직임이 일정 수준 이하인 구간을 의미한다. 일반적으로 "휴지기"는 대상이 깨어 있으면서 움직임이 없는 기간을 의미하지만, 본 명세서에서 용어 "휴지기"는 대상이 수면을 취하고 있는 기간도 포함하여 지칭한다.In this specification, the term “resting period” refers to a period in which a person's physical movement is below a certain level. Generally, “rest period” refers to a period in which the subject is awake and does not move, but in this specification, the term “rest period” also refers to a period in which the subject is sleeping.

휴지기 심박수들(resting heart rates)Resting heart rates

본 명세서에서, 용어 "휴지기 심박수들(resting heart rates)"는 그 심박수가 측정된 시점이 사람의 휴지기(resting period) 내에 있는 심박수들을 의미한다. As used herein, the term “resting heart rates” refers to heart rates that are within a person's resting period when the heart rate is measured.

특히, 본 명세서에서 "X일에 대응되는 휴지기 심박수들"이라 함은 X일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 휴지기에 대응되는 모든 심박수들을 의미할 수 있다. In particular, in this specification, "resting heart rates corresponding to day It can mean heart rates.

다만, i) 휴지기는 일반적으로 사람의 수면시간을 포함한다는 점, ii) 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 휴지기 심박수들"은 X일의 바로 전날(이하, X-1일이라 함)에 잠들기 시작한 시각부터 X-1일의 오후 11시 59분 59초까지의 시간에 대응되는 휴지기 심박수들을 포함할 수 있다. 이와 유사하게, "X일에 대응되는 휴지기 심박수들"는 X일에 잠들기 시작한 시각부터 X일의 오후 11시 59분 59초까지의 시간에 대응되는 휴지기 심박수들을 포함하지 않을 수 있다. 즉, X일에 잠들기 시작한 시각부터 X일의 오후 11시 59분 59초까지의 시간에 대응되는 휴지기 심박수들은 전술한 바와 같이 "X일의 다음날(이하, X+1일이라 함)에 대응되는 휴지기 심박수들"에 포함될 수 있다.However, taking into account that i) the resting period generally includes a person's sleeping time, and ii) considering people's sleeping habits, there are many people who start sleeping before 12:00:00 a.m. In this specification, “resting heart rates corresponding to day Corresponding resting heart rates may be included. Similarly, “resting heart rates corresponding to day In other words, the resting heart rates corresponding to the time from the time you start falling asleep on Day X to 11:59:59 PM on Day It may be included in “resting heart rates.”

구간 심박수들(period heart rates)Period heart rates

본 명세서에서, 용어 "구간 심박수들(period heart rates)"는 2일 이상의 기간에 대응되는 모든 휴지기 심박수들을 의미한다. 본 명세서에서 용어 "구간 휴지기 심박수들(period resting heart rates)"는 "구간 심박수들"과 동일한 의미로 사용될 수 있다.As used herein, the term “period heart rates” means all resting heart rates corresponding to a period of two or more days. As used herein, the term “period resting heart rates” may be used interchangeably with “period resting heart rates.”

구간 평균 심박수(period average heart rate)period average heart rate

본 명세서에서, 용어 "구간 평균 심박수(period average heart rate)"는 2일이상의 기간 동안 측정된 휴지기 심박수들(즉, 구간 심박수들)의 평균값을 의미한다. 예를 들어, 5일간의 구간 평균 심박수는 제1일(first day)에 대응되는 휴지기 심박수들, 제2일(second day)에 대응되는 휴지기 심박수들, 제3일(third day)에 대응되는 휴지기 심박수들, 제4일(fourth day)에 대응되는 휴지기 심박수들 및 제5일(fifth day)에 대응되는 휴지기 심박수들의 평균값을 의미한다. 본 명세서에서 용어 "구간 심박수들의 평균(average of period heart rates)"는 "구간 평균 심박수"와 동일한 의미로 사용될 수 있다.As used herein, the term “period average heart rate” refers to the average value of resting heart rates (i.e., zone heart rates) measured over a period of two or more days. For example, the average heart rate over a 5-day period is the resting heart rates corresponding to the first day, the resting heart rates corresponding to the second day, and the resting heart rates corresponding to the third day. It means the average value of heart rates, resting heart rates corresponding to the fourth day, and resting heart rates corresponding to the fifth day. In this specification, the term “average of period heart rates” may be used with the same meaning as “average period heart rate.”

구간 조화평균 심박수(Section harmonic average heart rate ( period harmonic average heart rate)period harmonic average heart rate)

본 명세서에서, 용어 "구간 조화평균 심박수(period harmonic average heart rate)"는 2일이상의 기간 동안 측정된 휴지기 심박수들(즉, 구간 심박수들)의 조화평균값을 의미한다. 예를 들어, 5일간의 구간 조화평균 심박수는 제1일(first day)에 대응되는 휴지기 심박수들, 제2일(second day)에 대응되는 휴지기 심박수들, 제3일(third day)에 대응되는 휴지기 심박수들, 제4일(fourth day)에 대응되는 휴지기 심박수들 및 제5일(fifth day)에 대응되는 휴지기 심박수들의 조화평균값을 의미한다. 본 명세서에서 용어 "구간 심박수들의 조화평균(harmonic average of period heart rates)"는 "구간 조화평균 심박수"와 동일한 의미로 사용될 수 있다.As used herein, the term “period harmonic average heart rate” refers to the harmonic average value of resting heart rates (i.e., interval heart rates) measured over a period of 2 days or more. For example, the 5-day interval harmonic average heart rate is the resting heart rates corresponding to the first day, the resting heart rates corresponding to the second day, and the resting heart rates corresponding to the third day. It means the harmonic average value of resting heart rates, resting heart rates corresponding to the fourth day, and resting heart rates corresponding to the fifth day. In this specification, the term “harmonic average of period heart rates” may be used with the same meaning as “period harmonic average heart rate.”

갑상선 호르몬thyroid hormones

본 명세서에서, 용어 "갑상선 호르몬(thyroid hormone)"은 갑상선(thyroid)에서 분비되는 호르몬인 티록신(thyroxine, T4) 또는 사요오드티로닌(tetraiodothyronine), T4가 혈청에 유리된 형태인 혈청 유리 T4(free T4) 뿐만 아니라, T4로부터 요오드가 떨어져 나가도록 전환된 삼요오드티로닌(triiodothyronine, T3), T3가 혈청에 유리된 형태인 혈청 유리 T3(free T3), 뇌하수체에서 분비되는 갑상선 자극 호르몬(Thyroid Stimulating Hormone, TSH), 그리고 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)도 포함하는 의미이다.In this specification, the term "thyroid hormone" refers to thyroxine (T4) or tetraiodothyronine, which are hormones secreted by the thyroid gland, and serum free T4 (T4 in the form of T4 free in serum). free T4), as well as triiodothyronine (T3), which is converted to remove iodine from T4, serum free T3 (free T3), which is a form of T3 that is free in the serum, and thyroid-stimulating hormone (Thyroid) secreted by the pituitary gland. It also includes Stimulating Hormone (TSH) and Thyrotropin-Releasing Hormone (TRH).

한편, 본 명세서에서, 용어 "호르몬"은 용어 "갑상선 호르몬"과 동일한 의미로 사용될 수 있다.Meanwhile, in this specification, the term “hormone” may be used with the same meaning as the term “thyroid hormone.”

갑상선기능항진증(hyperthyroidism)hyperthyroidism

본 명세서에서, 용어 "갑상선기능항진증(hyperthyroidism)"은 갑상선에서 분비되는 호르몬들이 과다하게 생성, 분비되거나 그 외의 요인으로 혈중 갑상선호르몬 농도가 과도하게 높아지는 상태 및 그로 인하여 발생되는 임상병리학적인 증상들을 모두 일컫는다. In this specification, the term "hyperthyroidism" refers to a condition in which the concentration of thyroid hormones in the blood is excessively high due to excessive production or secretion of hormones secreted by the thyroid gland or other factors, as well as all clinicopathological symptoms resulting therefrom. It refers to

갑상선기능저하증(hypothyroidism)hypothyroidism

본 명세서에서, 용어 "갑상선기능저하증(hypothyroidism)"은 갑상선에서 분비되는 호르몬들이 과소하게 생성, 분비되거나 그 외의 요인으로 혈중 갑상선호르몬 농도가 과도하게 낮아지는 상태 및 그로 인하여 발생되는 임상병리학적인 증상들을 모두 일컫는다. In this specification, the term "hypothyroidism" refers to a condition in which hormones secreted by the thyroid gland are produced or secreted excessively or the concentration of thyroid hormones in the blood is excessively low due to other factors, and the clinical-pathological symptoms that occur therefrom. It refers to everyone.

갑상선기능이상증(thyroid dysfunction)thyroid dysfunction

본 명세서에서, 용어 "갑상선기능이상증(thyroid dysfunction)"은 전술한 "갑상선기능저하증" 및 "갑상선기능항진증"을 모두 포함하는 의미이다.In this specification, the term “thyroid dysfunction” includes both the aforementioned “hypothyroidism” and “hyperthyroidism.”

깊은 수면 비율(deep sleep rate)deep sleep rate

본 명세서에서, 용어 "깊은 수면 비율(deep sleep rate)"는 대상의 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “deep sleep rate” refers to the ratio of deep sleep time to the subject's total sleep time.

본 명세서에서 "X일에 대응되는 깊은 수면 비율"이라 함은 X일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, “deep sleep ratio corresponding to day This may mean the proportion it occupies.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 깊은 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 깊은 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 당일(즉, X일)에 잠들기 시작한 시각까지의 시간에 대응되는 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미할 수 있다.However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "deep sleep ratio corresponding to It can refer to the ratio of deep sleep time to the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X. Alternatively, in this specification, “deep sleep ratio corresponding to day This may mean the proportion it occupies.

구간 깊은 수면 비율(period deep sleep rate)period deep sleep rate

본 명세서에서, 용어 "구간 깊은 수면 비율(period deep sleep rate)"는 2일 이상의 기간 동안 측정된 대상의 총 수면 시간 중 총 깊은 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “period deep sleep rate” refers to the ratio of total deep sleep time to a subject's total sleep time measured over a period of two or more days.

본 명세서에서 "X일에 대응되는 Y일간의 구간 깊은 수면 비율"이라 함은 X-Y+1일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, "deep sleep ratio for Y days corresponding to X days" includes the period from 12:00:00 AM on Day It may refer to the proportion of deep sleep time among the total sleep time.

예를 들어, 5일간의 구간 깊은 수면 비율은, 제1일(first day)에 대응되는 수면 시간, 제2일(second day)에 대응되는 수면 시간, 제3일(third day)에 대응되는 수면 시간, 제4일(fourth day)에 대응되는 수면 시간 및 제5일(fifth day)에 대응되는 수면 시간을 합하여 총 수면 시간으로 계산하고, 제1일(first day)에 대응되는 깊은 수면 시간, 제2일(second day)에 대응되는 깊은 수면 시간, 제3일(third day)에 대응되는 깊은 수면 시간, 제4일(fourth day)에 대응되는 깊은 수면 시간 및 제5일(fifth day)에 대응되는 깊은 수면 시간을 합하여 총 깊은 수면 시간으로 계산한 후, 총 수면 시간에서 총 깊은 수면 시간의 비율로 계산될 수 있다.For example, the deep sleep ratio for a 5-day period is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day. Time, the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day are added to calculate the total sleep time, the deep sleep time corresponding to the first day, Deep sleep time corresponding to the second day, deep sleep time corresponding to the third day, deep sleep time corresponding to the fourth day, and deep sleep time corresponding to the fifth day. The total deep sleep time can be calculated by adding the corresponding deep sleep time, and then calculated as the ratio of the total deep sleep time to the total sleep time.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 Y일간의 구간 깊은 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 깊은 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 Y일간의 구간 깊은 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠들기 시작한 시각까지의 총 수면 시간에 대한 총 깊은 수면 시간의 비율로 계산될 수 있으나, 이에 한정되지 않는다. However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "proportion of deep sleep for Y days corresponding to X days" may mean the ratio of deep sleep time to the total sleep time corresponding to the time from the time you started falling asleep the day before on day X-Y+1 to waking up on day X. Alternatively, in this specification, "interval deep sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.

얕은 수면 비율(light sleep rate)light sleep rate

본 명세서에서, 용어 "얕은 수면 비율(light sleep rate)"는 대상의 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “light sleep rate” refers to the proportion of light sleep time among the subject's total sleep time.

본 명세서에서 "X일에 대응되는 얕은 수면 비율"이라 함은 X일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, “light sleep ratio corresponding to day This may mean the proportion it occupies.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 얕은 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 얕은 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 당일(즉, X일)에 잠들기 시작한 시각까지의 시간에 대응되는 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미할 수 있다.However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "light sleep ratio corresponding to day It can refer to the proportion of light sleep time among the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X. Alternatively, in this specification, “light sleep ratio corresponding to day This may mean the proportion it occupies.

구간 section shallow 은 수면 비율(period light sleep rate)period light sleep rate

본 명세서에서, 용어 "구간 얕은 수면 비율(period light sleep rate)"는 2일 이상의 기간 동안 측정된 대상의 총 수면 시간 중 총 얕은 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “period light sleep rate” refers to the proportion of the total light sleep time of a subject's total sleep time measured over a period of two or more days.

본 명세서에서 "X일에 대응되는 Y일간의 구간 얕은 수면 비율"이라 함은 X-Y+1일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, “light sleep ratio for Y days corresponding to X days” includes the period from 12:00:00 AM on Day It may refer to the proportion of light sleep time among the total sleep time.

예를 들어, 5일간의 구간 얕은 수면 비율은, 제1일(first day)에 대응되는 수면 시간, 제2일(second day)에 대응되는 수면 시간, 제3일(third day)에 대응되는 수면 시간, 제4일(fourth day)에 대응되는 수면 시간 및 제5일(fifth day)에 대응되는 수면 시간을 합하여 총 수면 시간으로 계산하고, 제1일(first day)에 대응되는 얕은 수면 시간, 제2일(second day)에 대응되는 얕은 수면 시간, 제3일(third day)에 대응되는 얕은 수면 시간, 제4일(fourth day)에 대응되는 얕은 수면 시간 및 제5일(fifth day)에 대응되는 얕은 수면 시간을 합하여 총 얕은 수면 시간으로 계산한 후, 총 수면 시간에서 총 얕은 수면 시간의 비율로 계산될 수 있다.For example, the light sleep ratio for a 5-day interval is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day. Time, the total sleep time is calculated by adding the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day, the light sleep time corresponding to the first day, Light sleep time corresponding to the second day, light sleep time corresponding to the third day, light sleep time corresponding to the fourth day, and light sleep time corresponding to the fifth day. The total light sleep time can be calculated by adding the corresponding light sleep time, and then calculated as the ratio of the total light sleep time to the total sleep time.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 Y일간의 구간 얕은 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 얕은 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 Y일간의 구간 얕은 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠들기 시작한 시각까지의 총 수면 시간에 대한 총 얕은 수면 시간의 비율로 계산될 수 있으나, 이에 한정되지 않는다. However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "percentage of light sleep for Y days corresponding to X days" may mean the ratio of light sleep time to the total sleep time corresponding to the time from the time you started falling asleep the day before on day X-Y+1 to waking up on day X. Alternatively, in this specification, "interval light sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.

REM 수면 비율(REM sleep rate)REM sleep rate

본 명세서에서, 용어 "REM 수면 비율(REM sleep rate)"는 대상의 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “REM sleep rate” refers to the ratio of REM sleep time to the subject's total sleep time.

본 명세서에서 "X일에 대응되는 REM 수면 비율"이라 함은 X일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, “REM sleep ratio corresponding to day This may mean the proportion it occupies.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 REM 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 REM 수면 비율"은 X일의 바로 전날에 잠들기 시작한 시각부터 당일(즉, X일)에 잠들기 시작한 시각까지의 시간에 대응되는 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미할 수 있다.However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "REM sleep ratio corresponding to day It can refer to the proportion of REM sleep time among the total sleep time corresponding to the time from the time you started falling asleep the previous day to waking up on day X. Alternatively, in this specification, “REM sleep ratio corresponding to day This may mean the proportion it occupies.

구간 REM 수면 비율(period REM sleep rate)period REM sleep rate

본 명세서에서, 용어 "구간 REM 수면 비율(period REM sleep rate)"는 2일 이상의 기간 동안 측정된 대상의 총 수면 시간 중 총 REM 수면 시간이 차지하는 비율을 의미한다. As used herein, the term “period REM sleep rate” refers to the ratio of the total REM sleep time to the subject's total sleep time measured over a period of 2 days or more.

본 명세서에서 "X일에 대응되는 Y일간의 구간 REM 수면 비율"이라 함은 X-Y+1일의 오전 12시 00분 00초부터 X일의 오후 11시 59분 59초까지의 기간에 포함된 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미할 수 있다. In this specification, "REM sleep ratio for Y days corresponding to It may refer to the proportion of REM sleep time among the total sleep time.

예를 들어, 5일간의 구간 REM 수면 비율은, 제1일(first day)에 대응되는 수면 시간, 제2일(second day)에 대응되는 수면 시간, 제3일(third day)에 대응되는 수면 시간, 제4일(fourth day)에 대응되는 수면 시간 및 제5일(fifth day)에 대응되는 수면 시간을 합하여 총 수면 시간으로 계산하고, 제1일(first day)에 대응되는 REM 수면 시간, 제2일(second day)에 대응되는 REM 수면 시간, 제3일(third day)에 대응되는 REM 수면 시간, 제4일(fourth day)에 대응되는 REM 수면 시간 및 제5일(fifth day)에 대응되는 REM 수면 시간을 합하여 총 REM 수면 시간으로 계산한 후, 총 수면 시간에서 총 REM 수면 시간의 비율로 계산될 수 있다.For example, the REM sleep ratio for a 5-day period is the sleep time corresponding to the first day, the sleep time corresponding to the second day, and the sleep corresponding to the third day. Time, the sleep time corresponding to the fourth day and the sleep time corresponding to the fifth day are added to calculate the total sleep time, the REM sleep time corresponding to the first day, REM sleep time corresponding to the second day, REM sleep time corresponding to the third day, REM sleep time corresponding to the fourth day, and REM sleep time corresponding to the fifth day. The total REM sleep time can be calculated by adding the corresponding REM sleep time, and then calculated as the ratio of the total REM sleep time to the total sleep time.

다만, 사람들의 수면 습관들을 고려할 때, 오전 12시 00분 00초 이전에 수면을 시작하는 사람들이 많이 있다는 점 등을 고려하여, 본 명세서에서 "X일에 대응되는 Y일간의 구간 REM 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠에서 깨기까지의 시간에 대응되는 총 수면 시간 중 REM 수면 시간이 차지하는 비율을 의미할 수 있다. 또는, 본 명세서에서 "X일에 대응되는 Y일간의 구간 REM 수면 비율"은 X-Y+1일의 바로 전날에 잠들기 시작한 시각부터 X일에 잠들기 시작한 시각까지의 총 수면 시간에 대한 총 REM 수면 시간의 비율로 계산될 수 있으나, 이에 한정되지 않는다. However, considering people's sleeping habits and the fact that there are many people who start sleeping before 12:00:00 a.m., in this specification, "REM sleep ratio for Y days corresponding to X days" may mean the ratio of REM sleep time to the total sleep time corresponding to the time from the time you started falling asleep on the day before X-Y+1 to waking up on day X. Alternatively, in this specification, "REM sleep ratio for Y days corresponding to day It may be calculated as a ratio of time, but is not limited to this.

2. 학습 방법2. How to learn

개요outline

본 출원의 출원인은 2017년에 전술한 종래기술의 문제를 해결하고 시장의 니즈에 부합하는 기술을 개발하고자, 심박수와 갑상선기능항진증의 상관관계에 기초하여, 환자의 갑상선 기능이 "정상"이라고 판단된 시점에 대응되는 심박수, 현재의 심박수를 주요 입력변수로 하여 환자의 현재상태가 갑상선의 기능이 정상인 상태인지 혹은 현재의 상태가 갑상선의 기능에 이상이 있는 상태인지(즉, 갑상선기능항진증 혹은 갑상선기능저하증 인지 여부)를 판단하는 방법 및 시스템에 관한 기술 개발을 완료하고 이에 관한 특허출원을 진행한 바 있다.In 2017, the applicant of this application determined that the patient's thyroid function was "normal" based on the correlation between heart rate and hyperthyroidism in order to solve the problems of the prior art and develop technology that meets the needs of the market. The heart rate corresponding to the point in time and the current heart rate are used as the main input variables to determine whether the patient's current state is a state in which thyroid function is normal or a state in which there is an abnormality in thyroid function (i.e., hyperthyroidism or thyroid function). We have completed the development of technology and applied for a patent on a method and system to determine whether a patient has hypofunction.

그러나, 본 출원의 출원인이 개발하였던 기술에는 몇 가지 문제점이 존재하였다.However, there were several problems with the technology developed by the applicant of this application.

첫째, 사람마다 갑상선 기능이 정상인 상태에서의 심박수(미리 정해진 기간에 대한 구간 평균 심박수)가 다 다르기 때문에, 단순히 정상 상태일 때의 심박수와 현재 상태의 심박수를 인자(factor)로 하여 갑상선기능이상여부를 출력하는 함수를 도출하는 데에 어려움이 있었다.First, because each person's heart rate (average heart rate for a predetermined period) when thyroid function is normal is different for each person, the heart rate in normal state and the current heart rate are simply used as factors to determine whether thyroid function is abnormal. There was difficulty in deriving a function that outputs .

둘째, "정상"일 때의 심박수가 인자로 사용되기 때문에, 호르몬 수치를 검사한 날 "정상"이 아닌 "갑상선기능항진증" 또는 "갑상선기능저하증"으로 판단된 경우에는, 전술한 기술을 적용할 수 없다는 문제점이 있었다.Second, since the heart rate when “normal” is used as a factor, if the hormone level is judged to be “hyperthyroidism” or “hypothyroidism” instead of “normal” on the day of testing, the aforementioned technique cannot be applied. There was a problem that it couldn't be done.

전술한 문제점들로 인하여, 종래에 개발하였던 기술은 낮은 정확도 및 사용 상의 불편함 등으로 인해 지속적인 모니터링의 필요성을 요구하는 시장의 니즈를 충족시키기에는 부족함이 있었다.Due to the above-mentioned problems, previously developed technologies were insufficient to meet the needs of the market requiring continuous monitoring due to low accuracy and inconvenience in use.

이러한 문제점들을 극복하기 위하여, 본 출원인들은 본 명세서에 의해 개시되는 머신 러닝 기반의 갑상선기능이상 예측 모델을 제안한다.To overcome these problems, the present applicants propose a machine learning-based thyroid dysfunction prediction model disclosed by this specification.

또한, 본 출원인들은 갑상선기능이상 여부를 예측하는 예측모델을 학습시킴에 있어서, 개인화된 예측모델을 만들어 내기에는 여러 문제점이 있었기 때문에, 예측모델을 학습시키기 위한 데이터들의 속성을 분석하고, 이로부터 예측의 정확도를 높이기 위한 목적을 가지고 모델을 다양한 방식으로 학습을 시켜보았고, 이로부터 예측모델의 정확도를 높이기 위한 몇몇 유의미한 학습방법들을 확인하였고, 본 명세서를 통해 정확도를 높이기 위한 학습방법들을 개시한다.In addition, since the present applicants had several problems in creating a personalized prediction model when learning a prediction model that predicts thyroid dysfunction, they analyzed the properties of the data to learn the prediction model and made predictions from this. With the purpose of increasing the accuracy of the model, we trained the model in various ways. From this, we identified several meaningful learning methods to increase the accuracy of the prediction model, and through this specification, we disclose learning methods to increase the accuracy.

임상 데이터의 수집Collection of clinical data

본 출원에 의해 개시되는 예측 모델을 학습시키기 위한 기본적인 임상 데이터들을 수집한 방법에 대하여 먼저 설명한다.First, a method of collecting basic clinical data for learning the prediction model disclosed by this application will be described.

(1) 갑상선기능항진증 또는 갑상선기능저하증을 겪고 있는 환자들이 선별된다.(1) Patients suffering from hyperthyroidism or hypothyroidism are selected.

(2) 선별된 환자들에 대한 개인 정보들이 획득된다. 예를 들어, 선별된 환자들의 성명, 성별, 나이, 몸무게, 키 등등을 포함한 환자들의 개인정보들이 획득된다.(2) Personal information about selected patients is obtained. For example, personal information of selected patients is obtained, including name, gender, age, weight, height, etc.

(3) 선별된 환자들 각각에게, 손목에 착용가능하고 환자들의 심박수를 측정할 수 있는 기능이 탑재된 웨어러블 디바이스가 제공된다.(3) Each selected patient is provided with a wearable device that can be worn on the wrist and is equipped with a function that can measure the patient's heart rate.

(4) 선별된 환자들 각각에 대하여, 선별된 환자들이 병원에 방문하여 갑상선 호르몬 검사를 하면, 환자들에게 제공되었던 웨어러블 디바이스에 의해 측정된 심박수들 중 상기 검사일에 대응되는 구간 심박수들(period heart rates)이 확보된다.(4) For each selected patient, when the selected patients visit the hospital and undergo a thyroid hormone test, among the heart rates measured by the wearable device provided to the patients, the period heart rate corresponding to the test date is calculated. rates) are secured.

(5) 선별된 환자들 각각에 대하여, 상기 검사일에 시행된 갑상선 호르몬들의 농도값들이 확보된다.(5) For each selected patient, the concentration values of thyroid hormones performed on the test date are obtained.

(6) 선별된 환자들 각각에 대하여, 상기 갑상선 호르몬 검사 결과에 기초하여, 의사의 상기 환자에 대한 진단 결과가 확보된다. 즉, 상기 갑상선 호르몬 검사 결과, 상기 환자가 갑상선기능항진증, 갑상선기능저하증 및 정상 중 어떤 상태인지 여부가 확보된다.(6) For each of the selected patients, based on the thyroid hormone test results, a doctor's diagnosis result for the patient is secured. That is, as a result of the thyroid hormone test, it is confirmed whether the patient has hyperthyroidism, hypothyroidism, or normal condition.

결과적으로, 한 명의 환자가 갑상선 호르몬 검사를 1회 받게 되면, i) 환자의 개인정보, ii) 검사일, iii) 검사일에 대응되는 구간 심박수, iv) 검사일에 시행된 검사에 따른 갑상선 호르몬들의 농도, v) 상기 검사일에 시행된 검사 결과에 따른 갑상선기능이상여부에 대한 의사의 진단결과를 포함하는 하나의 데이터 세트가 확보될 수 있다.As a result, when a patient undergoes a thyroid hormone test once, i) the patient's personal information, ii) the test date, iii) the zone heart rate corresponding to the test date, iv) the concentration of thyroid hormones according to the test performed on the test day, v) One data set containing the doctor's diagnosis of thyroid dysfunction according to the results of the test performed on the above test date can be secured.

즉, 100명의 환자들로부터 한 환자 당 평균 4번의 호르몬 검사결과에 따른 데이터 세트들이 확보된다면, 총 400개의 데이터세트들이 확보될 수 있으며, 300명의 환자들로부터 한 환자 당 평균 3번의 호르몬 검사결과에 따른 데이터 세트들이 확보된다면, 총 900개의 데이터 세트들이 확보될 수 있다.In other words, if data sets based on an average of 4 hormone test results per patient are obtained from 100 patients, a total of 400 data sets can be obtained, and if data sets are obtained based on an average of 3 hormone test results per patient from 300 patients, If the corresponding data sets are secured, a total of 900 data sets can be secured.

갑상선기능이상을 겪고 있는 환자들은 3개월에 한 번씩 병원에 방문하여 호르몬 수치검사를 위한 채혈검사를 할 것을 권고 받는다. 따라서, 기본적으로 전술한 방식으로 데이터 세트들을 충분히 확보를 하려면, 데이터 세트의 수집 대상이 되는 환자들의 수를 늘리거나 혹은 환자 당 데이터 세트를 확보하는 기간을 수년 이상으로 늘려야 한다. 즉, 많은 비용이 투입되거나 혹은 많은 시간이 투입되어야만 갑상선기능이상에 관련된 많은 임상 데이터들이 확보될 수 있는 것이 현재의 실정이다.Patients suffering from thyroid dysfunction are advised to visit the hospital once every three months and have a blood sample taken to test hormone levels. Therefore, in order to secure sufficient data sets in the above-described manner, the number of patients subject to data set collection must be increased or the period for securing data sets per patient must be extended to several years or more. In other words, the current situation is that a lot of clinical data related to thyroid dysfunction can be secured only when a lot of money or a lot of time is invested.

학습방법 실시예#1. 기준일 심박수와의 차이값과 기준일의 호르몬값Learning method example #1. Difference value from heart rate on reference day and hormone value on reference day

본 출원에 의해 개시되는 제1 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다. According to the first embodiment disclosed by the present application, in order to learn a learning model, the value obtained by subtracting the section average heart rate corresponding to the second test day from the section average heart rate corresponding to the first test day and the hormone concentration of the second test day It can be used as an input value, and the diagnosis result corresponding to the first inspection date can be used as a labeling value.

이때, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등을 포함하는 분류 모델들이 사용될 수 있다. At this time, classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.

상기 학습 모델은 갑상선기능항진증을 예측하기 위한 항진증 예측 모델(hyperthyroidism prediction model)과 갑상선기능저하증을 예측하기 위한 저하증 예측 모델(hypothyroidism prediction model)을 포함한다.The learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.

제1 실시예에 따라 항진증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.When learning a hyperthyroidism prediction model according to the first embodiment, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능항진증 진단을 받은 적이 있는 환자는 '항진증 환자(hyperthyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.'

한편, 구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.Meanwhile, in calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.

제1 실시예에 따라, 항진증 예측 모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.According to the first embodiment, in order to learn a hyperactivity disorder prediction model, clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the second examination day. Learning data sets in the following format can be created: corresponding thyroid hormone concentration, diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date.

예를 들어, 제1 항진증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with hyperactivity disorder 1, the diagnosis result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnosis result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), third concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), first concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), third concentration, no hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 항진아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), first concentration, no hyperactivity], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 항진아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), second concentration, no hyperactivity],

다른 예를 들어, 제2 항진증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hyperactivity disorder, the diagnostic result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'hyperactivity'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that the concentration was the same, learning datasets that can be obtained from the clinical data of the second hyperactivity disorder patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), third concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), fourth concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 항진아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), first concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 항진아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), third concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the fourth test day), fourth concentration, no hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), first concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), second concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 항진], [(section average heart rate corresponding to the 3rd test day - section average heart rate corresponding to the 4th test day), 4th concentration, hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 항진아님], [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the first test day), first concentration, no hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 항진아님], 및 [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the second test day), second concentration, no hyperactivity], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 항진아님],[(interval average heart rate corresponding to the 4th test day - interval average heart rate corresponding to the 3rd test day), 3rd concentration, no hyperactivity],

제1 실시예에 따라, 저하증 예측 모델을 학습시킬 때 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.According to the first embodiment, not all clinical data obtained when learning a hypothyroidism prediction model can be used, and only clinical data of patients who have been diagnosed with hypothyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능저하증 진단을 받은 적이 있는 환자는 '저하증 환자(hypothyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’

제1 실시예에 따라, 저하증 예측 모델을 학습시키기 위하여, 저하증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(저하 또는 저하아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.According to the first embodiment, in order to learn a hypothyroidism prediction model, clinical data of patients with hypothyroidism were used to calculate [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the second examination day. Learning data sets in the following format can be created: corresponding thyroid hormone concentration, diagnostic result (low or not low) corresponding to the first test date.

예를 들어, 제1 저하증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with first hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), third concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), first concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), third concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 저하아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), first concentration, not low], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 저하아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), second concentration, not low],

다른 예를 들어, 제2 저하증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'decreased'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), third concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the fourth test day), fourth concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 저하아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), first concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 저하아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), third concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the fourth test day), fourth concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), first concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), second concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the fourth test day), fourth concentration, decline],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 농도, 저하아님], [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the first test day), first concentration, not low],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 농도, 저하아님], 및 [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the second test day), second concentration, not low], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 농도, 저하아님],[(interval average heart rate corresponding to the 4th test day - interval average heart rate corresponding to the 3rd test day), 3rd concentration, not low],

제1 실시예에 있어서, 호르몬 농도는 T4의 농도, free T4의 농도, T3의 농도, free T3의 농도, TSH의 농도 및 갑상샘자극호르몬방출호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수도 있다.In the first embodiment, the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.

학습방법 실시예#2. 기준일 심박수와의 차이값, 대상일 구간 평균 심박수와 기준일의 호르몬값Learning method example #2. Difference value from the heart rate on the reference day, average heart rate in the section on the target day, and hormone value on the reference day

본 출원에 의해 개시되는 제2 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값, 제1 검사일의 호르몬 농도 및 제2 검사일에 대응되는 구간 평균 심박수를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다. According to the second embodiment disclosed by the present application, in order to learn a learning model, the section average heart rate corresponding to the first test day is subtracted from the section average heart rate corresponding to the second test day, the hormone concentration on the first test day, and The section average heart rate corresponding to the second test date can be used as an input value, and the diagnosis result corresponding to the first test date can be used as a labeling value.

이때, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등을 포함하는 분류 모델들이 사용될 수 있다.At this time, classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.

상기 학습 모델은 갑상선기능항진증을 예측하기 위한 항진증 예측 모델(hyperthyroidism prediction model)과 갑상선기능저하증을 예측하기 위한 저하증 예측 모델(hypothyroidism prediction model)을 포함한다.The learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.

제2 실시예에 따라, 항진증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.According to the second embodiment, when learning a hyperthyroidism prediction model, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once in the diagnosis results can be used as learning data. .

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능항진증 진단을 받은 적이 있는 환자는 '항진증 환자(hyperthyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.'

한편, 구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.Meanwhile, in calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.

제2 실시예에 따라, 항진증 예측 모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.According to a second embodiment, in order to learn a hyperactivity disorder prediction model, clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the first examination day. Learning data sets in a format such as the corresponding section average heart rate, thyroid hormone concentration corresponding to the second test date, and diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day can be created.

예를 들어, 제1 항진증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with hyperactivity disorder 1, the diagnosis result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnosis result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the first test day, third concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제1 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the second test day, first concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제3 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the second test day, third concentration, no hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제1 농도, 항진아님], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the third test day, first concentration, no hyperactivity], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제2 농도, 항진아님],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the third test day, second concentration, no hyperactivity],

다른 예를 들어, 제2 항진증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hyperactivity disorder, the diagnostic result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'hyperactivity'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that the concentration was the same, learning datasets that can be obtained from the clinical data of the second hyperactivity disorder patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the first test day, third concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제4 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the first test day, fourth concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제1 농도, 항진아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the second test day, first concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제3 농도, 항진아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the second test day, third concentration, no hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제4 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the second test day, fourth concentration, no hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제1 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the third test day, first concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제2 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the third test day, second concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제4 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the third test day, fourth concentration, hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제1 농도, 항진아님], [(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the fourth test day, first concentration, no hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제2 농도, 항진아님], 및 [(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the fourth test day, second concentration, no hyperactivity], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제3 농도, 항진아님],[(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), section average heart rate corresponding to the 4th test day, 3rd concentration, no hyperactivity],

제2 실시예에 따라 저하증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.When learning a hypothyroidism prediction model according to the second embodiment, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hypothyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능저하증 진단을 받은 적이 있는 환자는 '저하증 환자(hypothyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’

제2 실시예에 따라 저하증 예측 모델을 학습시키기 위하여, 저하증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(저하 또는 저하아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.In order to learn a hypothyroidism prediction model according to the second embodiment, clinical data of patients with hypothyroidism were used to obtain [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), corresponding to the first examination day. Learning data sets in the following format can be created: average heart rate for the interval, thyroid hormone concentration corresponding to the second test date, and diagnosis result (deterioration or not) corresponding to the first test date.

예를 들어, 제1 저하증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with first hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제3 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the first test day, third concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제1 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the second test day, first concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제3 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the second test day, third concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제1 농도, 저하아님], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the third test day, first concentration, not low], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제2 농도, 저하아님],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the third test day, second concentration, not low],

다른 예를 들어, 제2 저하증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'decreased'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제3 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the first test day, third concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제4 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the first test day, fourth concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제1 농도, 저하아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the second test day, first concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제3 농도, 저하아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), section average heart rate corresponding to the second test day, third concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 평균 심박수, 제4 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the second test day, fourth concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제1 농도, 저하], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the third test day, first concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제2 농도, 저하], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the third test day, second concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제3 검사일에 대응되는 구간 평균 심박수, 제4 농도, 저하], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the fourth test day), section average heart rate corresponding to the third test day, fourth concentration, decline],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제1 농도, 저하아님], [(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the first test day), section average heart rate corresponding to the fourth test day, first concentration, not low],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제2 농도, 저하아님], 및 [(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the fourth test day, second concentration, not low], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제4 검사일에 대응되는 구간 평균 심박수, 제3 농도, 저하아님],[(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), section average heart rate corresponding to the 4th test day, 3rd concentration, not low],

제2 실시예에 있어서, 호르몬 농도는 T4의 농도, free T4의 농도, T3의 농도, free T3의 농도, TSH의 농도 및 갑상샘자극호르몬방출호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수도 있다.In the second embodiment, the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.

전술한 제2 실시예에 대한 설명에서는 입력값으로 대상일에 대응되는 구간 평균 심박수를 활용하는 것을 기준으로 설명하였으나, 대상일에 대응되는 구간 평균 심박수, 대상일에 대응되는 구간 조화평균 심박수, 대상일에 대응되는 구간 심박수들의 표준편차, 대상일에 대응되는 구간 심박수들의 상대 표준편차, 대상일에 대응되는 구간 심박수들의 왜도 및 대상일에 대응되는 구간 심박수들의 첨도로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the second embodiment described above, the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day. A value or a combination thereof can be used as an input value.

학습방법 실시예#3. 기준일 심박수와의 차이값, 기준일의 호르몬값 및 분포의 변화Learning method example #3. Difference value from the heart rate on the reference day, changes in hormone values and distribution on the reference day

본 출원에 의해 개시되는 제3 실시예에 따르면 학습 모델을 학습시키기 위하여, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값, 제1 검사일에 대응되는 구간 심박수들의 표준편차에서 제2 검사일에 대응되는 구간 심박수들의 표준편차를 뺀 값 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다. According to the third embodiment disclosed by the present application, in order to learn a learning model, the section average heart rate corresponding to the first inspection day is subtracted from the section average heart rate corresponding to the second inspection day, and the section heart rate corresponding to the first inspection day The standard deviation of the heart rates corresponding to the second test date is subtracted from the standard deviation of the heart rates and the hormone concentration of the second test day can be used as input values, and the diagnosis result corresponding to the first test date can be used as a labeling value.

이때, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등을 포함하는 분류 모델들이 사용될 수 있다.At this time, classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.

상기 학습 모델은 갑상선기능항진증을 예측하기 위한 항진증 예측 모델(hyperthyroidism prediction model)과 갑상선기능저하증을 예측하기 위한 저하증 예측 모델(hypothyroidism prediction model)을 포함한다.The learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.

제3 실시예에 따라 항진증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.When learning a hyperthyroidism prediction model according to the third embodiment, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능항진증 진단을 받은 적이 있는 환자는 '항진증 환자(hyperthyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.'

한편, 구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.Meanwhile, in calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.

제3 실시예에 따라, 항진증 예측 모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.According to the third embodiment, in order to learn a hyperactivity disorder prediction model, clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), (first examination day Standard deviation of section heart rates corresponding to - standard deviation of section heart rates corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day]. Learning data sets can be created.

예를 들어, 제1 항진증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with hyperactivity disorder 1, the diagnosis result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnosis result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, not hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 항진아님], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not hyperactivity], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 항진아님],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, not hyperactivity],

다른 예를 들어, 제2 항진증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hyperactivity disorder, the diagnostic result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'hyperactivity'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that the concentration was the same, learning datasets that can be obtained from the clinical data of the second hyperactivity disorder patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 항진], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the fourth test day), 4 concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 항진아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 항진아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, not hyperactivity];

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 항진아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the fourth test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the fourth test day), 4 concentration, not hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 항진], [(section average heart rate corresponding to the 3rd test day - section average heart rate corresponding to the 4th test day), (standard deviation of section heart rates corresponding to the 1st test day - standard deviation of section heart rates corresponding to the 4th test day), 4 concentration, hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 항진아님], [(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the 4th test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 항진아님], 및 [(segment average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 2nd test day), (standard deviation of the section heart rates corresponding to the 4th test day - standard deviation of the section heart rates corresponding to the 2nd test day), 2 concentration, not hyperactivity], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 항진아님][(segment average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), (standard deviation of the section heart rates corresponding to the 4th test day - standard deviation of the section heart rates corresponding to the 3rd test day), 3 concentration, not hyperactivity]

제3 실시예에 따라 저하증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.When learning a hypothyroidism prediction model according to the third embodiment, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hypothyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능저하증 진단을 받은 적이 있는 환자는 '저하증 환자(hypothyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’

제3 실시예에 따라 저하증 예측 모델을 학습시키기 위하여, 저하증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(저하 또는 저하아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.In order to learn a hypothyroidism prediction model according to the third embodiment, clinical data of patients with hypothyroidism were used to obtain [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), (interval average heart rate corresponding to the first examination day), standard deviation of corresponding interval heart rates - standard deviation of interval heart rates corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnostic result (low or not low) corresponding to the first test day] Training data sets may be created.

예를 들어, 제1 저하증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with first hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, lowering];

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, lowering];

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 저하아님], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not lowering], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 저하아님],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, not low],

다른 예를 들어, 제2 저하증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'decreased'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, lowering];

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, lowering];

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 저하], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the fourth test day), 4 concentration, lowering],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 저하아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not low];

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 저하아님], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the third test day), 3 concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제2 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 저하아님],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the fourth test day), (standard deviation of section heart rates corresponding to the second test day - standard deviation of section heart rates corresponding to the fourth test day), 4 concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 저하], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, lowering],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 저하], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the third test day - standard deviation of section heart rates corresponding to the second test day), 2 concentration, lowering];

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), (제3 검사일에 대응되는 구간 심박수들의 표준편차 - 제4 검사일에 대응되는 구간 심박수들의 표준편차), 제4 농도, 저하], [(segment average heart rate corresponding to the 3rd test day - section average heart rate corresponding to the 4th test day), (standard deviation of the section heart rates corresponding to the 3rd test day - standard deviation of the section heart rates corresponding to the 4th test day), 4 concentration, lowering],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제1 검사일에 대응되는 구간 심박수들의 표준편차), 제1 농도, 저하아님], [(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the first test day), (standard deviation of section heart rates corresponding to the 4th test day - standard deviation of section heart rates corresponding to the first test day), 1 concentration, not low];

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 농도, 저하아님], 및 [(segment average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 2nd test day), (standard deviation of the section heart rates corresponding to the 4th test day - standard deviation of the section heart rates corresponding to the 2nd test day), 2 concentration, not lowering], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), (제4 검사일에 대응되는 구간 심박수들의 표준편차 - 제3 검사일에 대응되는 구간 심박수들의 표준편차), 제3 농도, 저하아님][(segment average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), (standard deviation of the section heart rates corresponding to the 4th test day - standard deviation of the section heart rates corresponding to the 3rd test day), 3 concentration, not low]

전술한 제3 실시예에 대한 설명에서는 입력값으로 표준편차의 차이값을 활용하는 것을 기준으로 설명하였으나, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the third embodiment described above, the explanation was based on using the difference value of the standard deviation as the input value, but the calculation was made between the section heart rates corresponding to the X test date and the section heart rates corresponding to the Y test day. The difference value of the standard deviation of the section heart rates, the difference value of the relative standard deviation of the section heart rates, the difference value of the skewness of the section heart rates, the difference value of the kurtosis of the section heart rates, and Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of corresponding section heart rates and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, may be used as an input value.

학습방법 실시예#4. 기준일 심박수와의 차이값, 대상일의 구간 평균 심박수, 기준일의 호르몬값 및 분포의 변화Learning method example #4. Difference value from the heart rate on the reference day, average heart rate in the section on the target day, changes in hormone values and distribution on the reference day

본 출원에 의해 개시되는 제4 실시예에 따르면 학습 모델을 학습시키기 위하여, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값, 제1 검사일에 대응되는 구간 심박수들의 표준편차에서 제2 검사일에 대응되는 구간 심박수들의 표준편차를 뺀 값, 제1 검사일에 대응되는 구간 평균 심박수 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다.According to the fourth embodiment disclosed by the present application, in order to learn a learning model, the section average heart rate corresponding to the first inspection day is subtracted from the section average heart rate corresponding to the second inspection day, and the section heart rate corresponding to the first inspection day The standard deviation of the section heart rates corresponding to the second test day is subtracted from the standard deviation of the section average heart rate corresponding to the first test day and the hormone concentration of the second test day are used as input values, and the Diagnosis results can be used as labeling values.

제4 실시예는, 제3 실시예와 비교할 때, "제1 검사일의 구간 평균 심박수"를 입력값으로 더 사용한다는 것만 상이하고 다른 것들 대부분 동일하기 때문에, 여기에서는 자세한 설명을 생략하기로 한다.Compared to the third embodiment, the fourth embodiment differs only in that it uses “segment average heart rate on the first test day” as an input value and most other things are the same, so detailed description will be omitted here.

전술한 제4 실시예에 대한 설명에서는 입력값으로 대상일에 대응되는 구간 평균 심박수를 활용하는 것을 기준으로 설명하였으나, 대상일에 대응되는 구간 평균 심박수, 대상일에 대응되는 구간 조화평균 심박수, 대상일에 대응되는 구간 심박수들의 표준편차, 대상일에 대응되는 구간 심박수들의 상대 표준편차, 대상일에 대응되는 구간 심박수들의 왜도 및 대상일에 대응되는 구간 심박수들의 첨도로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the above-mentioned fourth embodiment, the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day. A value or a combination thereof can be used as an input value.

전술한 제4 실시예에 대한 설명에서는 입력값으로 표준편차의 차이값을 활용하는 것을 기준으로 설명하였으나, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the fourth embodiment described above, the explanation was based on using the difference value of the standard deviation as the input value, but the calculation was made between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day. The difference value of the standard deviation of the section heart rates, the difference value of the relative standard deviation of the section heart rates, the difference value of the skewness of the section heart rates, the difference value of the kurtosis of the section heart rates, and Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of corresponding section heart rates and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, may be used as an input value.

학습방법 실시예#5. 기준일 심박수와의 변화율, 기준일의 호르몬값Learning method example #5. Rate of change from heart rate on the reference day, hormone value on the reference day

본 출원에 의해 개시되는 제5 실시예에 따르면 학습 모델을 학습시키기 위하여, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)) 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다. According to the fifth embodiment disclosed by the present application, in order to learn a learning model, the rate of change of the interval average heart rate of the first examination day with respect to the second examination day (i.e., (interval average heart rate corresponding to the first examination day - the second examination day) (section average heart rate corresponding to) / (section average heart rate corresponding to the second test day)) and the hormone concentration of the second test day are used as input values, and the diagnosis result corresponding to the first test day can be used as a labeling value. there is.

제5 실시예는, 제1 실시예와 비교할 때, "구간 평균 심박수의 변화량"을 사용하는 대신 "구간 평균 심박수의 변화율"을 사용하는 것만 상이하고 다른 것들은 대부분 동일하기 때문에, 여기에서는 자세한 설명을 생략하기로 한다.Compared to the first embodiment, the fifth embodiment differs only in that it uses "rate of change in section average heart rate" instead of "change amount in section average heart rate" and most other things are the same, so a detailed description is given here. Decided to omit it.

학습방법 실시예#6. 기준일 심박수와의 변화율, 대상일의 구간 평균 심박수, 기준일의 호르몬값Learning method example #6. Rate of change from heart rate on the reference day, average heart rate in the section on the target day, hormone value on the reference day

본 출원에 의해 개시되는 제6 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일에 대응되는 구간 평균 심박수 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다.According to the sixth embodiment disclosed by the present application, in order to learn a learning model, the rate of change of the interval average heart rate of the first test day with respect to the second test day (i.e., (interval average heart rate corresponding to the first test day - the second (section average heart rate corresponding to the test day) / (section average heart rate corresponding to the second test day)), the section average heart rate corresponding to the first test day and the hormone concentration of the second test day are used as input values, and on the first test day The corresponding diagnostic result can be used as a labeling value.

제6 실시예는, 제5 실시예와 비교할 때, "제1 검사일의 구간 평균 심박수"를 입력값으로 더 사용한다는 것만 상이하고 다른 것들 대부분 동일하기 때문에, 여기에서는 자세한 설명을 생략하기로 한다.Compared to the fifth embodiment, the sixth embodiment differs only in that it uses “segment average heart rate on the first test day” as an input value and most other things are the same, so detailed description will be omitted here.

전술한 제6 실시예에 대한 설명에서는 입력값으로 대상일에 대응되는 구간 평균 심박수를 활용하는 것을 기준으로 설명하였으나, 대상일에 대응되는 구간 평균 심박수, 대상일에 대응되는 구간 조화평균 심박수, 대상일에 대응되는 구간 심박수들의 표준편차, 대상일에 대응되는 구간 심박수들의 상대 표준편차, 대상일에 대응되는 구간 심박수들의 왜도 및 대상일에 대응되는 구간 심박수들의 첨도로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the sixth embodiment described above, the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day. A value or a combination thereof can be used as an input value.

학습방법 실시예#7. 기준일 심박수와의 변화율, 기준일의 호르몬값 및 분포의 변화Learning method example #7. Rate of change from heart rate on the reference day, changes in hormone values and distribution on the reference date

본 출원에 의해 개시되는 제7 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제2 검사일에 대한 제1 검사일의 구간 심박수들의 상대 표준편차의 변화량 (즉, (제1 검사일에 대응되는 구간 심박수들의 표준편차 / 제1 검사일에 대응되는 구간 평균 심박수) - (제2 검사일에 대응되는 구간 심박수들의 표준편차 / 제2 검사일에 대응되는 구간 평균 심박수)) 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다. According to the seventh embodiment disclosed by the present application, in order to learn a learning model, the rate of change of the interval average heart rate of the first test day with respect to the second test day (i.e., (interval average heart rate corresponding to the first test day - 2 Average heart rate of the section corresponding to the test day) / (average heart rate of the section corresponding to the second test day)), the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day (i.e., (section corresponding to the first test day) Standard deviation of heart rates / average heart rate of the section corresponding to the first test day) - (standard deviation of the heart rates of the section corresponding to the second test day / average heart rate of the section corresponding to the second test day)) and the hormone concentration of the second test day are input values. , and the diagnosis result corresponding to the first inspection date can be used as a labeling value.

제7 실시예는, 제5 실시예와 비교할 때, "구간 심박수들의 상대표준편차의 변화량"을 입력값으로 더 사용하는 것만 상이하고 다른 것들은 대부분 동일하기 때문에, 여기에서는 자세한 설명을 생략하기로 한다.Compared to the fifth embodiment, the seventh embodiment differs only in using "the amount of change in the relative standard deviation of section heart rates" as an input value, and most other things are the same, so detailed description will be omitted here. .

전술한 제7 실시예에 대한 설명에서는 입력값으로 상대 표준편차의 차이값을 활용하는 것을 기준으로 설명하였으나, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the above-described seventh embodiment, the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day The difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.

학습방법 실시예#8. 기준일 심박수와의 변화율, 대상일의 구간 평균 심박수, 기준일의 호르몬값 및 분포의 변화Learning method example #8. Rate of change from the heart rate on the reference day, average heart rate in the section on the target day, changes in hormone values and distribution on the reference day

본 출원에 의해 개시되는 제8 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제2 검사일에 대한 제1 검사일의 구간 심박수들의 상대 표준편차의 변화량 (즉, (제1 검사일에 대응되는 구간 심박수들의 표준편차 / 제1 검사일에 대응되는 구간 평균 심박수) - (제2 검사일에 대응되는 구간 심박수들의 표준편차 / 제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일에 대응되는 구간 평균 심박수 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다.According to the eighth embodiment disclosed by the present application, in order to learn a learning model, the rate of change of the interval average heart rate of the first test day with respect to the second test day (i.e., (interval average heart rate corresponding to the first test day - 2 Average heart rate of the section corresponding to the test day) / (average heart rate of the section corresponding to the second test day)), the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day (i.e., (section corresponding to the first test day) Standard deviation of heart rates / section average heart rate corresponding to the first test day) - (standard deviation of section heart rates corresponding to the second test day / section average heart rate corresponding to the second test day)), section average heart rate corresponding to the first test day And the hormone concentration on the second test day can be used as an input value, and the diagnosis result corresponding to the first test day can be used as a labeling value.

제8 실시예는, 제7 실시예와 비교할 때, "제1 기준일의 구간 평균 심박수"를 입력값으로 더 사용하는 것만 상이하고 다른 것들은 대부분 동일하기 때문에, 여기에서는 자세한 설명을 생략하기로 한다.Compared to the seventh embodiment, the eighth embodiment differs only in that it further uses “segment average heart rate on the first reference day” as an input value, and most other things are the same, so detailed description will be omitted here.

전술한 제8 실시예에 대한 설명에서는 입력값으로 대상일에 대응되는 구간 평균 심박수를 활용하는 것을 기준으로 설명하였으나, 대상일에 대응되는 구간 평균 심박수, 대상일에 대응되는 구간 조화평균 심박수, 대상일에 대응되는 구간 심박수들의 표준편차, 대상일에 대응되는 구간 심박수들의 상대 표준편차, 대상일에 대응되는 구간 심박수들의 왜도 및 대상일에 대응되는 구간 심박수들의 첨도로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the above-described eighth embodiment, the explanation was based on using the section average heart rate corresponding to the target day as the input value, but the section average heart rate corresponding to the target day, the section harmonic average heart rate corresponding to the target day, Any one selected from the group consisting of the standard deviation of the section heart rates corresponding to the day, the relative standard deviation of the section heart rates corresponding to the target day, the skewness of the section heart rates corresponding to the target day, and the kurtosis of the section heart rates corresponding to the target day. A value or a combination thereof can be used as an input value.

전술한 제8 실시예에 대한 설명에서는 입력값으로 상대 표준편차의 차이값을 활용하는 것을 기준으로 설명하였으나, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the above-described eighth embodiment, the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test day and the section heart rates corresponding to the Y test day The difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.

학습방법 실시예#9. 기준일 심박수와의 변화율(또는 변화량), 기준일의 호르몬값, 분포의 변화 및 검사일 간의 차이Learning method example #9. Rate of change (or amount of change) from baseline heart rate, hormone values on baseline date, changes in distribution, and differences between test days

본 출원에 의해 개시되는 제9 실시예에 따르면 학습 모델을 학습시키기 위하여, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제2 검사일에 대한 제1 검사일의 구간 심박수들의 상대 표준편차의 변화량 (즉, (제1 검사일에 대응되는 구간 심박수들의 표준편차 / 제1 검사일에 대응되는 구간 평균 심박수) - (제2 검사일에 대응되는 구간 심박수들의 표준편차 / 제2 검사일에 대응되는 구간 평균 심박수)), 제2 검사일의 호르몬 농도 및 제1 검사일과 제2 검사일 사이의 차이(days)를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다.According to the ninth embodiment disclosed by the present application, in order to learn a learning model, the rate of change of the interval average heart rate of the first examination day with respect to the second examination day (i.e., (interval average heart rate corresponding to the first examination day - the second examination day) (section average heart rate corresponding to the second test day) / (section average heart rate corresponding to the second test day)), the amount of change in the relative standard deviation of the section heart rates of the first test day with respect to the second test day (i.e., (segment heart rate corresponding to the first test day) standard deviation of section heart rates corresponding to the first test day) - (standard deviation of section heart rates corresponding to the second test day / section average heart rate corresponding to the second test day)), hormone concentration on the second test day and first test day The difference (days) between and the second inspection date can be used as an input value, and the diagnosis result corresponding to the first inspection date can be used as a labeling value.

제9 실시예는, 제7 실시예와 비교할 때, 제1 검사일과 제2 검사일 사이의 차이를 입력값으로 더 활용한다는 것만 상이하고 다른 것들은 대부분 동일하기 때문에, 여기서는 자세한 설명을 생략하기로 한다.Compared to the seventh embodiment, the ninth embodiment differs only in that the difference between the first inspection date and the second inspection date is used as an input value, and most other things are the same, so detailed description will be omitted here.

전술한 제9 실시예에 대한 설명에서는 입력값으로 상대 표준편차의 차이값을 활용하는 것을 기준으로 설명하였으나, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.In the description of the ninth embodiment described above, the explanation was based on using the difference value of the relative standard deviation as the input value, but between the section heart rates corresponding to the X test date and the section heart rates corresponding to the Y test day The difference value of the standard deviation of the interval heart rates, the difference value of the relative standard deviation of the interval heart rates, the difference value of the skewness of the interval heart rates, the difference value of the kurtosis of the interval heart rates, and the X test date that can be calculated Any one value selected from the group consisting of JS divergence, which represents the difference between the distribution of section heart rates corresponding to and the distribution of section heart rates corresponding to the Y test date, or a combination thereof, can be used as an input value.

한편, 기준일 심박수와의 변화율 및 상대 표준편차의 변화량을 사용하는 대신, 기준일 심박수와의 변화량 및 표준편차의 변화량을 사용하는 것도 가능하다. 이 경우, 상대 표준편차의 변화량뿐만 아니라, 제X 검사일에 대응되는 구간 심박수들과 제Y 검사일에 대응되는 구간 심박수들 사이에 대해서 계산될 수 있는, 구간 심박수들의 표준편차의 차이값, 구간 심박수들의 상대 표준편차의 차이값, 구간 심박수들의 왜도(skewness)의 차이값, 구간 심박수들의 첨도(kurtois)의 차이값 및 제X 검사일에 대응되는 구간 심박수들 분포와 제Y 검사일에 대응되는 구간 심박수들의 분포 사이의 차이를 나타내는 JS divergence로 이루어진 그룹에서 선택된 어느 하나의 값 또는 이들의 조합이 입력값으로 활용될 수 있다.Meanwhile, instead of using the rate of change from the heart rate on the reference day and the amount of change in the relative standard deviation, it is also possible to use the amount of change from the heart rate on the reference day and the amount of change in the standard deviation. In this case, not only the change in relative standard deviation, but also the difference value of the standard deviation of the section heart rates, which can be calculated between the section heart rates corresponding to the The difference value of relative standard deviation, the difference value of skewness of section heart rates, the difference value of kurtosis of section heart rates, and the distribution of section heart rates corresponding to the Any one value selected from the group consisting of JS divergence, which represents the difference between distributions, or a combination thereof, can be used as an input value.

학습방법 실시예#10. 변수들의 조합Learning method example #10. combination of variables

본 출원에 의해 개시되는 제10 실시예에 따르면, 학습 모델을 학습시키기 위하여, 다음의 [표 1]에 개시된 변수들의 조합값(combination value)이 입력값으로 더 활용될 수 있다.According to the tenth embodiment disclosed by the present application, in order to train the learning model, the combination value of the variables disclosed in the following [Table 1] can be further used as an input value.

NoNo 변수variable 1One 구간 평균 심박수 변화량Average heart rate change for each section 22 구간 평균 심박수 변화율Zone average heart rate change rate 33 구간 조화평균 심박수 변화량Section harmonic average heart rate change 44 구간 조화평균 심박수 변화율Section harmonic average heart rate change rate 55 표준편차 변화량standard deviation change 66 상대 표준편차 변화량relative standard deviation change 77 구간 심박수의 왜도 변화량Skewness change in zone heart rate 88 구간 심박수의 첨도 변화량Change in kurtosis of interval heart rate 99 JS DivergenceJS Divergence 1010 제2 기준일의 TSH 농도TSH concentration on second reference day 1111 제2 기준일의 free T4 농도Free T4 concentration on second reference day 1212 제2 기준일의 T4 농도T4 concentration on second reference day 1313 제2 기준일의 free T3 농도Free T3 concentration on second reference day 1414 제2 기준일의 T3 농도T3 concentration on second reference day 1515 제2 기준일의 TRH 농도TRH concentration on second reference day

제1 기준일에 제1 구간 평균 심박수, 제1 표준편차, 제1 왜도, 제1 첨도, 제1 TSH 호르몬 농도, 제1 free T4 농도, 제1 T4 농도, 제1 free T3 농도, 제1 T3 농도, 제1 TRH 농도 및 제1 기준일에 대한 갑상선기능이상증에 대한 진단결과가 대응되어 있고, 제2 기준일에 제2 구간 평균 심박수, 제2 표준편차, 제2 왜도, 제2 첨도, 제2 TSH 호르몬 농도, 제2 free T4 농도, 제2 T4 농도, 제2 free T3 농도, 제2 T3 농도, 제2 TRH 농도 및 제2 기준일에 대한 갑상선기능이상증에 대한 진단결과가 대응되어 있을 때, 상기 변수들은 다음과 같이 산출된다.(1) 구간 평균 심박수 변화량 = 제2 기준일의 구간 평균 심박수 - 제1 기준일의 구간 평균 심박수On the first reference day, the first section average heart rate, first standard deviation, first skewness, first kurtosis, first TSH hormone concentration, first free T4 concentration, first T4 concentration, first free T3 concentration, first T3 Concentration, first TRH concentration, and diagnosis results for thyroid dysfunction for the first reference date are corresponded, and on the second reference date, the second section average heart rate, second standard deviation, second skewness, second kurtosis, and second When the diagnosis results for dysthyroidism for the TSH hormone concentration, second free T4 concentration, second T4 concentration, second free T3 concentration, second T3 concentration, second TRH concentration and second reference date correspond to the above The variables are calculated as follows: (1) Change in section average heart rate = section average heart rate on the second reference day - section average heart rate on the first reference day

(2) 구간 평균 심박수 변화율 = (제2 기준일의 구간 평균 심박수 - 제1 기준일의 구간 평균 심박수) / 제2 기준일의 구간 평균 심박수(2) Zone average heart rate change rate = (segment average heart rate on the second reference day - section average heart rate on the first reference day) / section average heart rate on the second reference day

(3) 구간 조화평균 심박수 변화량 = 제2 기준일의 구간 조화평균 심박수 - 제1 기준일의 구간 조화평균 심박수(3) Change in section harmonic average heart rate = section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day

(4) 구간 조화평균 심박수 변화율 = (제2 기준일의 구간 조화평균 심박수 - 제1 기준일의 구간 조화평균 심박수) / 제2 기준일의 구간 조화화평균 심박수(4) Section harmonic average heart rate change rate = (section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day) / section harmonic average heart rate on the second reference day

(5) 표준편차 변화량 = 제2 기준일의 구간 심박수들의 표준편차 - 제1 기준일의 구간심박수들의 표준편차(5) Standard deviation change = standard deviation of zone heart rates on the second reference day - standard deviation of zone heart rates on the first reference day

(6) 상대 표준편차 변화량 = (제2 기준일의 구간 심박수들의 표준편차 / 제2 기준일의 구간 평균 심박수) - (제1 기준일의 구간심박수들의 표준편차 / 제1 기준일의 구간 평균 심박수)(6) Relative standard deviation change = (standard deviation of section heart rates on the second reference day / section average heart rate on the second reference day) - (standard deviation of section heart rates on the first reference day / section average heart rate on the first reference day)

(7) 구간 심박수의 왜도 변화량 = 제2 기준일의 구간 심박수들의 왜도 - 제1 기준일의 구간 심박수들의 왜도(7) Skewness change in section heart rates = skewness of section heart rates on the second reference day - skewness of section heart rates on the first reference day

(8) 구간 심박수의 첨도 변화량 = 제2 기준일의 구간 심박수들의 첨도 - 제1 기준일의 구간 심박수들의 왜도(8) Change in kurtosis of section heart rates = kurtosis of section heart rates on the second reference day - skewness of section heart rates on the first reference day

(9) JS Divergence : 제2 기준일의 구간 심박수들과 제1 기준일의 구간 심박수들 사이에서 계산되는 JS Divergence(9) JS Divergence: JS Divergence calculated between the zone heart rates of the second reference day and the zone heart rates of the first reference day

이때, 상기 변수들간의 조합값은 상기 변수들 중에서 선택되는 하나의 값(이하, 제1 값이라 함)과 상기 상기 변수들 중에서 선택되는 다른 하나의 값(이하, 제2 값이라 함)을 서로 곱한 값, 제1 값을 제2 값으로 나눈 값 및 제2 값을 제1 값으로 나눈 값 중 하나를 의미한다. 예를 들어, 상기 조합값은 제2 기준일의 free T4의 농도값을 JS Divergence로 나눈값일 수 있다.At this time, the combination value between the variables is one value selected from among the variables (hereinafter referred to as the first value) and another value selected from the variables (hereinafter referred to as the second value). It means one of a multiplied value, a value obtained by dividing a first value by a second value, and a value obtained by dividing a second value by a first value. For example, the combination value may be the concentration value of free T4 on the second reference date divided by JS Divergence.

이론적으로, 상기 조합값들의 개수는 15*14/2 + 15*14 = 315개일 수 있다.Theoretically, the number of combination values may be 15*14/2 + 15*14 = 315.

한편, 변수들간의 조합값은 둘 이상이 사용될 수 있다.Meanwhile, two or more combination values between variables can be used.

제10 실시예에 따라 학습 모델을 학습시킬 때, 입력값은 다음과 같을 수 있다.When training a learning model according to the tenth embodiment, the input value may be as follows.

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 전술한 조합값들 중 적어도 하나][change in interval average heart rate, at least one of the hormone concentrations on the second test day, and at least one of the above-mentioned combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 표준편차 변화량, 전술한 조합값들 중 적어도 하나][change in section average heart rate, at least one of the hormone concentrations on the second test day, change in standard deviation of section average heart rate, and at least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of the hormone concentrations on the second test day, section average heart rate on the first reference day, and at least one of the above-mentioned combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 표준편차 변화량, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][change in section average heart rate, at least one of the hormone concentrations on the second test day, standard deviation change in section average heart rate, section average heart rate on the first reference day, at least one of the above-mentioned combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 표준편차 변화량, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][change in section average heart rate, at least one of the hormone concentrations on the second test day, standard deviation change in section average heart rate, section average heart rate on the first reference day, difference between the second reference date and the first reference day (day gap), above At least one of the combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 전술한 조합값들 중 적어도 하나][rate of change in interval average heart rate, at least one of the hormone concentrations on the second test day, and at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 상대 표준편차 변화량, 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in relative standard deviation of section average heart rate, and at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][rate of change in interval average heart rate, at least one of the hormone concentrations on the second test day, interval average heart rate on the first reference day, and at least one of the above-mentioned combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 상대 표준편차 변화량, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in relative standard deviation of section average heart rate, section average heart rate on the first reference day, and at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수의 상대 표준편차 변화량, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, relative standard deviation change in section average heart rate, section average heart rate on the first reference day, difference between the second reference date and the first reference date (day gap), At least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 전술한 조합값들 중 적어도 하나][At least one of the change in section average heart rate, hormone concentrations on the second test day, change in skewness or kurtosis of section average heart rates, or JS Divergence, at least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of hormone concentrations on the second test day, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, at least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of the hormone concentrations on the second test day, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, difference between the second reference date and the first reference date (day gap), at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 전술한 조합값들 중 적어도 하나][Rate of change in interval average heart rate, at least one of the hormone concentrations on the second test day, change in skewness or kurtosis of interval average heart rates, or JS Divergence, at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, at least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, difference between the second reference date and the first reference date (day gap), at least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of the hormone concentrations on the second test day, change in standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates, or JS Divergence, at least one of the above combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of the hormone concentrations on the second test day, change in standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, above At least one of the combination values]

[구간 평균 심박수의 변화량, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][Change in section average heart rate, at least one of the hormone concentrations on the second test day, change in standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, 2 Difference between the reference date and the first reference date (day gap), at least one of the above-mentioned combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 상대 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in relative standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates, or JS Divergence, at least one of the above combination values ]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 상대 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in relative standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, At least one of the above combination values]

[구간 평균 심박수의 변화율, 제2 검사일의 호르몬 농도들 중 적어도 하나, 구간 평균 심박수들의 상대 표준편차의 변화량, 구간 평균 심박수들의 왜도 또는 첨도의 변화량 또는 JS Divergence, 제1 기준일의 구간 평균 심박수, 제2 기준일과 제1 기준일 사이의 차이(day gap), 전술한 조합값들 중 적어도 하나][Rate of change in section average heart rate, at least one of the hormone concentrations on the second test day, change in relative standard deviation of section average heart rates, change in skewness or kurtosis of section average heart rates or JS Divergence, section average heart rate on the first reference day, Difference between the second reference date and the first reference date (day gap), at least one of the above-mentioned combination values]

한편, 제10 실시예에 따라 학습 모델을 학습시킬 때, 라벨링값은 제1 검사일의 진단결과일 수 있다. 예를 들어, 갑상선기능항진증 예측 모델을 학습시킬 때, 제1 검사일의 진단 결과가 '항진증'이면, 그 라벨값은 '항진'으로, 제1 검사일의 진단 결과가 '정상'이면, 그 라벨값은 '항진아님'으로 라벨할 수 있고, 갑상선기능저하증 예측 모델을 학습시킬 때, 제1 검사일의 진단 결과가 '저하증'이면, 그 라벨값은 '저하'로, 제1 검사일의 진단 결과가 '정상'이면, 그 라벨값은 '저하아님'으로 라벨할 수 있다.Meanwhile, when training the learning model according to the tenth embodiment, the labeling value may be the diagnosis result on the first inspection day. For example, when learning a hyperthyroidism prediction model, if the diagnosis result on the first test day is 'hyperthyroidism', the label value is 'hyperthyroidism', and if the diagnosis result on the first test day is 'normal', the label value is can be labeled as 'not hyperthyroid', and when training a hypothyroidism prediction model, if the diagnosis result on the first test day is 'hypothyroidism', the label value is 'hypothyroidism', and the diagnosis result on the first test day is 'hypothyroidism'. If it is ‘normal’, the label value can be labeled as ‘not deteriorated’.

학습방법 실시예#11. 성별 및 연령 등의 개인정보의 활용Learning method example #11. Use of personal information such as gender and age

본 출원에 의해 개시되는 제11 실시예에 따르면, 학습 모델을 학습시키기 위하여, 전술한 제1 실시예 내지 제10 실시예에서 설명한 입력값에 더하여, 대상의 개인정보를 입력값으로 사용할 수 있다. 예를 들어, 대상의 성별 및/또는 나이가 입력값으로 사용될 수 있다.According to the eleventh embodiment disclosed by the present application, in order to train a learning model, the target's personal information can be used as an input value in addition to the input values described in the above-described first to tenth embodiments. For example, the subject's gender and/or age may be used as input values.

예를 들어, 제1 실시예에서 설명한 입력값에 더하여 대상의 성별 및 나이를입력값으로 사용하는 경우, 제2 검사일에 대한 제1 검사일의 구간 평균 심박수의 변화율 (즉, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일에 대응되는 구간 평균 심박수, 제2 검사일의 호르몬 농도, 성별 및 나이를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 진단 결과를 라벨링값으로 사용할 수 있다.For example, when the gender and age of the subject are used as input values in addition to the input values described in the first embodiment, the rate of change of the interval average heart rate of the first test day with respect to the second test day (i.e., (corresponds to the first test day) average heart rate of the section corresponding to the second test day - average heart rate of the section corresponding to the second test day) / (average heart rate of the section corresponding to the second test day)), average heart rate of the section corresponding to the first test day, hormone concentration, gender and age of the second test day It can be used as an input value, and the diagnosis result corresponding to the first inspection date can be used as a labeling value.

제1 실시예 내지 제10 실시예에서 설명된 입력값들에 더하여 대상의 개인정보를 보다 더 사용하는 것에 대한 자세한 설명은 생략하기로 한다.A detailed description of further use of the subject's personal information in addition to the input values described in the first to tenth embodiments will be omitted.

한편, 대상의 개인정보로 성별과 나이를 모두 사용하는 것에 대하여 설명하였지만, 성별과 나이가 함께 사용되는 대신, 성별만 사용되거나 나이만 사용되는 실시예도 가능하며, 성별과 나이 이외의 다른 개인정보가 더 활용될 수도 있을 것이다.On the other hand, although the use of both gender and age as the target's personal information was explained, an embodiment in which only gender or age is used instead of gender and age together is also possible, and personal information other than gender and age is used. It could be used more.

학습방법 실시예#12. 추정된 대상일의 호르몬값의 활용Learning method example #12. Utilization of hormone values on estimated target date

본 출원에 의해 개시되는 제12 실시예에 따르면, 학습 모델을 학습시키기 위하여, 전술한 제1 실시예 내지 제11 실시예에서 설명하는 입력값에 더하여, 제1 검사일에 대응되는 갑상선 호르몬 농도에 관한 값이 입력값으로 사용할 수 있다. According to the twelfth embodiment disclosed by the present application, in order to learn a learning model, in addition to the input values described in the above-described first to eleventh embodiments, the thyroid hormone concentration corresponding to the first test date is used. The value can be used as an input value.

제1 검사일에 대응되는 갑상선 호르몬 농도에 관한 값은 갑상선 호르몬 예측모델의 출력값을 기초로 획득될 수 있다. 일 예로, 갑상선 호르몬 농도에 관한 값은 갑상선 호르몬 예측 모델의 출력값을 기초로 추정된 대상일의 갑상선 호르몬 농도 값일 수 있다. 다른 예로, 갑상선 호르몬 농도에 관한 값은 갑상선 호르몬 예측 모델의 출력값을 기초로 추정된 기준일의 갑상선 호르몬 농도에 대한 대상일의 갑상선 호르몬 농도의 변화값 내지는 변화량일 수 있다. 또 다른 예로, 갑상선 호르몬 농도에 관한 값은 갑상선 호르몬 예측 모델의 출력값 일 수 있다.The value related to the thyroid hormone concentration corresponding to the first test date can be obtained based on the output value of the thyroid hormone prediction model. As an example, the value related to the thyroid hormone concentration may be the thyroid hormone concentration value of the target date estimated based on the output value of the thyroid hormone prediction model. As another example, the value related to the thyroid hormone concentration may be a change value or amount of change in the thyroid hormone concentration on the target day with respect to the thyroid hormone concentration on the reference day estimated based on the output value of the thyroid hormone prediction model. As another example, a value related to thyroid hormone concentration may be an output value of a thyroid hormone prediction model.

제1 검사일에 대응되는 갑상선 호르몬 농도에 관한 값은 하나 이상의 값일 수 있다. 예를 들어, 갑상선 항진증에 특화된 갑상선 호르몬 농도 예측 모델의 출력값을 기초로 생성된 제1 값, 갑상선 기능 정상에 특화된 갑상선 호르몬 농도 예측 모델의 출력값을 기초로 생성된 제2 값, 및 갑상선 저하증에 특화된 갑상선 호르몬 농도 예측 모델의 출력값을 기초로 생성된 제3 값을 모두 이용하여 학습 모델을 학습시킬 수 있으나, 이에 한정되지 않는다.The value regarding the thyroid hormone concentration corresponding to the first test date may be one or more values. For example, a first value generated based on the output value of a thyroid hormone concentration prediction model specialized for hyperthyroidism, a second value generated based on the output value of a thyroid hormone concentration prediction model specialized for normal thyroid function, and a second value generated based on the output value of a thyroid hormone concentration prediction model specialized for normal thyroid function, and The learning model may be trained using all third values generated based on the output value of the thyroid hormone concentration prediction model, but is not limited to this.

이때, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등을 포함하는 분류 모델들이 사용될 수 있다.At this time, classification models including light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc. may be used as the learning model.

상기 학습 모델은 갑상선기능항진증을 예측하기 위한 항진증 예측 모델(hyperthyroidism prediction model)과 갑상선기능저하증을 예측하기 위한 저하증 예측 모델(hypothyroidism prediction model)을 포함한다.The learning model includes a hyperthyroidism prediction model for predicting hyperthyroidism and a hypothyroidism prediction model for predicting hypothyroidism.

제12 실시예에 따라, 항진증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.According to the 12th embodiment, when learning a hyperthyroidism prediction model, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once in the diagnosis results can be used as learning data. .

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능항진증 진단을 받은 적이 있는 환자는 '항진증 환자(hyperthyroidism patient)'라고 지칭할 수 있다. 한편, 구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hyperthyroidism at least once in the diagnostic results may be referred to as a 'hyperthyroidism patient.' Meanwhile, in calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, It can be one of the following days: 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30.

제12 실시예에 따라, 항진증 예측 모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 추정 갑상선 호르몬 농도에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.According to the twelfth embodiment, in order to learn a hyperactivity disorder prediction model, clinical data of patients with hyperactivity were used [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), on the first examination day. Learning data sets in the following format may be generated: a value for the corresponding estimated thyroid hormone concentration, a thyroid hormone concentration corresponding to the second test date, and a diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date.

예를 들어, 제1 항진증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with hyperactivity disorder 1, the diagnosis result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnosis result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hyperactivity disorder patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 항진아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), estimated value for the third concentration, first concentration, no hyperactivity], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 항진아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, not hyperactivity],

다른 예를 들어, 제2 항진증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hyperactivity disorder, the diagnostic result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'hyperactivity'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that the concentration was the same, learning datasets that can be obtained from the clinical data of the second hyperactivity disorder patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제4 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the fourth test day), estimated value for the first concentration, fourth concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 항진아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 항진아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제4 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the fourth test day), estimated value for the second concentration, fourth concentration, not hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), estimated value for the third concentration, first concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 항진], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제4 농도, 항진], [(section average heart rate corresponding to the third test day - section average heart rate corresponding to the fourth test day), estimated value for the third concentration, fourth concentration, hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제1 농도, 항진아님], [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the first test day), estimated value for the fourth concentration, first concentration, no hyperactivity],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제2 농도, 항진아님], 및 [(interval average heart rate corresponding to the 4th test day - interval average heart rate corresponding to the 2nd test day), estimated value for the 4th concentration, 2nd concentration, no hyperactivity], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제3 농도, 항진아님],[(interval average heart rate corresponding to the 4th test day - interval average heart rate corresponding to the 3rd test day), estimated value for the 4th concentration, 3rd concentration, not hyperactivity],

또 다른 예를 들어, 제1 항진증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '항진'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '저하'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 항진증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 3 data sets obtained from a patient with hyperactivity disorder 1, the diagnosis result for the first test is 'hyperactivity', the hormone concentration at this time was the first concentration, and the diagnosis result for the second test is 'normal'. ', and at this time, the hormone concentration was the second concentration, and the diagnostic result for the third test was 'decreased'. Assuming that the hormone concentration at this time was the third concentration, the learning data that can be obtained from the clinical data of the first hyperactivity disorder patient Sets can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, hyperactivity],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 항진], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not hyperactivity],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 항진아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not hyperactivity],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 항진아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), estimated value for the third concentration, first concentration, no hyperactivity], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 항진아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, not hyperactivity],

제12 실시예에 따라 저하증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있다.When learning a hypothyroidism prediction model according to the 12th embodiment, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hypothyroidism at least once can be used as learning data.

이하에서는, 설명의 편의를 위하여, 진단결과에서 적어도 한번 이상 갑상선기능저하증 진단을 받은 적이 있는 환자는 '저하증 환자(hypothyroidism patient)'라고 지칭할 수 있다.Hereinafter, for convenience of explanation, a patient who has been diagnosed with hypothyroidism at least once in the diagnostic results may be referred to as a ‘hypothyroidism patient.’

제12 실시예에 따라 저하증 예측 모델을 학습시키기 위하여, 저하증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 추정 갑상선 호르몬 농도에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(저하 또는 저하아님)]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.In order to learn a hypothyroidism prediction model according to the 12th embodiment, clinical data of patients with hypothyroidism were used to determine [(interval average heart rate corresponding to the first examination day - interval average heart rate corresponding to the second examination day), corresponding to the first examination day. Learning data sets in the following format may be generated: a value for the estimated thyroid hormone concentration, a thyroid hormone concentration corresponding to the second test date, and a diagnosis result (low or not low) corresponding to the first test date.

예를 들어, 제1 저하증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For example, there are a total of 3 data sets obtained from a patient with first hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. , At this time, the hormone concentration was the second concentration, the diagnostic result for the third test was 'normal', and assuming that the hormone concentration at this time was the third concentration, the learning datasets that can be obtained from the clinical data of the first hypothyroidism patient are It can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not lower],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 저하아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), estimate for the third concentration, first concentration, not lower], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 저하아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, not low],

다른 예를 들어, 제2 저하증 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였으며, 세번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제3 농도였으며, 네번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제4 농도였다고 가정한다면, 상기 제2 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 4 data sets obtained from patients with secondary hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. At this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'decreased'. At this time, the hormone concentration was the third concentration, and the diagnosis result for the fourth test was 'normal'. At this time, the hormone concentration was the fourth test. Assuming that it was the concentration, learning datasets that can be obtained from the clinical data of the second hypothyroidism patient can be created as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제4 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the fourth test day), estimated value for the first concentration, fourth concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 저하아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not lower],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 저하아님], [(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not low],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제4 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the fourth test day), estimated value for the second concentration, fourth concentration, not lower],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), estimated value for the third concentration, first concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, decline],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제4 농도, 저하], [(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the fourth test day), estimated value for the third concentration, fourth concentration, decline],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제1 농도, 저하아님], [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the first test day), estimated value for the fourth concentration, first concentration, not low],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제2 농도, 저하아님], 및 [(interval average heart rate corresponding to the fourth test day - interval average heart rate corresponding to the second test day), estimated value for the fourth concentration, second concentration, not lower], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제4 농도에 관한 추정값, 제3 농도, 저하아님],[(interval average heart rate corresponding to the 4th test day - interval average heart rate corresponding to the 3rd test day), estimated value for the 4th concentration, 3rd concentration, not low],

또 다른 예를 들어, 제1 저하증 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사에 대한 진단결과는 '저하'이며, 이때 호르몬 농도는 제1 농도였고, 두번째 검사에 대한 진단 결과는 '정상'이며, 이때 호르몬 농도는 제2 농도였고, 세번째 검사에 대한 진단 결과는 '항진'이며, 이때 호르몬 농도는 제3 농도였다고 가정한다면, 상기 제1 저하증 환자의 임상데이터로부터 확보할 수 있는 학습 데이터세트들은 다음과 같이 생성될 수 있다.For another example, there are a total of 3 data sets obtained from a patient with first hypothyroidism, and the diagnostic result for the first test is 'low', the hormone concentration at this time was the first concentration, and the diagnostic result for the second test is 'normal'. ', and at this time, the hormone concentration was the second concentration, and the diagnosis result for the third test was 'hyperactivity'. Assuming that the hormone concentration at this time was the third concentration, the learning data that can be obtained from the clinical data of the first hypothyroidism patient Sets can be created as follows:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제2 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the second test day), estimated value for the first concentration, second concentration, decline],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 농도에 관한 추정값, 제3 농도, 저하], [(interval average heart rate corresponding to the first test day - interval average heart rate corresponding to the third test day), estimated value for the first concentration, third concentration, decline],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제1 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the first test day), estimated value for the second concentration, first concentration, not lower],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 농도에 관한 추정값, 제3 농도, 저하아님],[(interval average heart rate corresponding to the second test day - interval average heart rate corresponding to the third test day), estimated value for the second concentration, third concentration, not low],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제1 농도, 저하아님], 및[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the first test day), estimate for the third concentration, first concentration, not lower], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 농도에 관한 추정값, 제2 농도, 저하아님],[(interval average heart rate corresponding to the third test day - interval average heart rate corresponding to the second test day), estimated value for the third concentration, second concentration, not low],

제12 실시예에 있어서, 호르몬 농도는 T4의 농도, free T4의 농도, T3의 농도, free T3의 농도, TSH의 농도 및 갑상샘자극호르몬방출호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수도 있다.In the twelfth embodiment, the hormone concentration is a group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and Thyrotropin-Releasing Hormone (TRH) concentration. It may be one selected from or a combination thereof.

제12 실시예를 개시함에 있어서, 설명의 편의를 위해,In disclosing the twelfth embodiment, for convenience of explanation,

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 추정 갑상선 호르몬 농도에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도]를 학습 데이터의 입력으로 사용하면서, 그에 대한 라벨링값으로 [제1 검사일에 대응되는 진단결과]을 사용하는 경우를 상정하여 설명하였으나,[(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), value related to the estimated thyroid hormone concentration corresponding to the first inspection day, thyroid hormone concentration corresponding to the second inspection day] as learning data The explanation was based on the assumption that [diagnosis result corresponding to the first examination date] is used as the labeling value while using it as an input.

제12 실시예에서의 개시하는 핵심적인 사상은 "제1 검사일에 대응되는 추정 갑상선 호르몬 농도에 관한 값"을 입력으로 이용한다는 점이며, 학습모델에 입력되는 그 외의 정보는 추가되거나 생략될 수 있다. 예를 들어, 제1 내지 제11 실시예에서 개시된 입력 데이터에 제1 검사일에 대응되는 추정 갑상선 호르몬 농도에 관한 값을 입력으로 더 이용하는 형태로 학습 모델이 구현될 수 있다.The core idea disclosed in the 12th embodiment is that the “value related to the estimated thyroid hormone concentration corresponding to the first test date” is used as input, and other information input to the learning model can be added or omitted. . For example, the learning model may be implemented in a form that further uses a value related to the estimated thyroid hormone concentration corresponding to the first test date as input to the input data disclosed in the first to eleventh embodiments.

실시예 5를 빗대어 설명하면, 학습모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일에 대응되는 추정 갑상선 호르몬 농도값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.To explain Example 5, in order to learn a learning model, clinical data of patients with hyperactivity were used to calculate [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / (second test day) Learning data sets in the following format are generated: section average heart rate corresponding to the test date)), estimated thyroid hormone concentration value corresponding to the first test date, thyroid hormone concentration corresponding to the second test date, and diagnosis result corresponding to the first test day. It can be.

실시예 9를 빗대어 설명하면, 학습모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일과 제2 검사일 사이의 차이(days), 제1 검사일에 대응되는 추정 갑상선 호르몬 농도값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.To explain Example 9, in order to learn a learning model, clinical data of patients with hyperactivity were used to calculate [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / (second test day) Average heart rate of the section corresponding to the test date)), difference between the first test date and the second test date (days), estimated thyroid hormone concentration value corresponding to the first test day, thyroid hormone concentration corresponding to the second test day, Learning data sets in the same format as [corresponding diagnostic results] can be created.

전술한 제1 내지 제12 실시예에 있어서, 항진증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있음을 전제로 학습 데이터 세트를 설명한바 있다.In the above-described first to twelfth embodiments, when learning a hyperthyroidism prediction model, not all obtained clinical data can be used, and only clinical data of patients who have been diagnosed with hyperthyroidism at least once in the diagnosis results are learned. The learning data set has been described on the premise that it can be used as data.

또한, 전술한 제1 내지 제12 실시예에 있어서, 저하증 예측 모델을 학습시킬 때, 확보한 모든 임상데이터가 활용될 수는 없고, 진단결과에 적어도 한번 이상 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들만 학습 데이터로 사용될 수 있음을 전제로 학습 데이터 세트를 설명한바 있다.In addition, in the above-described first to twelfth embodiments, when learning a hypothyroidism prediction model, not all obtained clinical data can be used, and clinical data of patients who have been diagnosed with hypothyroidism at least once in the diagnosis results The learning data set has been described on the premise that only those data can be used as learning data.

다만, 전술한 제1 내지 제12 실시예에 있어서도, 항진증 예측 모델을 학습시킬 때, 진단결과에 적어도 한번 이상의 갑상선기능항진증 진단을 받은 환자들의 임상 데이터들로 항진 또는 항진아님에 대한 학습 데이터가 생성되고, 진단결과에 한 번도 갑상선기능항진증 진단을 받지않은 환자들의 임상데이터들로 항진아님에 대한 학습데이터가 생성될 수 있음은 물론이다.However, even in the above-described first to twelfth embodiments, when learning the hyperthyroidism prediction model, learning data for hyperthyroidism or non-hyperthyroidism is generated from clinical data of patients diagnosed with hyperthyroidism at least once in the diagnosis results. Of course, learning data about hyperthyroidism can be generated from clinical data of patients who have never been diagnosed with hyperthyroidism as a result of the diagnosis.

또한, 전술한 제1 내지 제12 실시예에 있어서도, 저하증 예측 모델을 학습시킬 때, 진단결과에 적어도 한번 이상의 갑상선기능저하증 진단을 받은 환자들의 임상 데이터들로 저하 또는 저하아님에 대한 학습 데이터가 생성되고, 진단결과에 한 번도 갑상선기능저하증 진단을 받지않은 환자들의 임상데이터들로 저하아님에 대한 학습데이터가 생성될 수 있음은 물론이다.In addition, in the above-described first to twelfth embodiments, when learning the hypothyroidism prediction model, learning data for hypothyroidism or non-hypothyroidism is generated using clinical data of patients diagnosed with hypothyroidism at least once in the diagnosis results. Of course, learning data about hypothyroidism can be generated from clinical data of patients who have never been diagnosed with hypothyroidism.

한편, 전술한 제1 내지 제12 실시예에 있어서, 갑상선기능이상 예측모델을 학습시킴에 있어서 유의미한 정보들을 개시한 것이며, 만약, 변화량(차이값) 또는 변화율이 유의미한 정보로 특정된 경우 갑상선기능이상 예측 모델을 학습시키는 데이터로 변화량(차이값) 또는 변화율을 대신하여 제1 검사일에 대응되는 값과 제2 검사일에 대응되는 값을 각각 활용하는 것은 당연히 가능하다.Meanwhile, in the above-described first to twelfth embodiments, meaningful information is disclosed in learning the thyroid dysfunction prediction model, and if the amount of change (difference value) or change rate is specified as meaningful information, thyroid dysfunction It is naturally possible to use the value corresponding to the first inspection date and the value corresponding to the second inspection date instead of the change amount (difference value) or change rate as data for training the prediction model.

이하에서는, 본 출원에 의해 개시되는 예측 모델을 이용하여 대상(subject)의 갑상선기능이상증을 예측하는 방법 및 시스템에 대해서 설명한다.Below, a method and system for predicting thyroid dysfunction in a subject using the prediction model disclosed by this application will be described.

3. 전체 시스템3. Entire system

(1) 시스템의 하드웨어적 구성(1) System hardware configuration

도 1은 본 출원에 의해 개시되는 일실시예에 따른 갑상선기능이상증을 예측하기 위한 시스템을 도시하는 도면이다.1 is a diagram illustrating a system for predicting thyroid dysfunction according to an embodiment disclosed by the present application.

도 1을 참조하면, 상기 시스템(1)은 복수의 심박수 측정 디바이스들(heart rate measuring device, 10), 복수의 사용자 단말기들(user terminal, 20) 및 서버(30)를 포함한다.Referring to FIG. 1, the system 1 includes a plurality of heart rate measuring devices (10), a plurality of user terminals (20), and a server (30).

이하에서는, 상기 복수의 심박수 측정 디바이스들(10), 상기 복수의 사용자 단말기들(20) 및 상기 서버(30)에 대해 보다 구체적으로 설명한다.Hereinafter, the plurality of heart rate measurement devices 10, the plurality of user terminals 20, and the server 30 will be described in more detail.

(2) 심박수 측정 디바이스의 기능(2) Function of heart rate measurement device

상기 복수의 심박수 측정 디바이스들(10)은 대상(사용자)의 심박수(heart rate)를 측정할 수 있다. 예를 들어, 상기 심박수 측정 디바이스들(10)은 대상의 단위 시간당 심박수(beats per minute)를 측정할 수 있다. 즉, 상기 심박수 측정 디바이스들(10)은 대상의 분당 심박수를 측정할 수 있다.The plurality of heart rate measurement devices 10 can measure the heart rate of a subject (user). For example, the heart rate measurement devices 10 may measure the subject's heart rate (beats per minute) per unit time. That is, the heart rate measurement devices 10 can measure the heart rate per minute of the subject.

상기 복수의 심박수 측정 디바이스들(10)은 측정된 대상의 심박수들을 저장할 수 있다. 이때, 상기 심박수 측정 디바이스들(10)은, 상기 복수의 심박수들을 저장할 때, 상기 심박수(단위 시간당 심박수)가 측정된 시각(time point)를 함께 매칭시켜 저장할 수 있다.The plurality of heart rate measurement devices 10 may store the measured heart rates of the target. At this time, when the heart rate measurement devices 10 store the plurality of heart rates, they may store the time points at which the heart rate (heart rate per unit time) was measured by matching them together.

상기 복수의 심박수 측정 디바이스들(10)은 상기 저장된 심박수들을 상기 복수의 사용자 단말기들(20) 및/또는 상기 서버(30)로 전송할 수 있다.The plurality of heart rate measurement devices 10 may transmit the stored heart rates to the plurality of user terminals 20 and/or the server 30.

상기 복수의 심박수 측정 디바이스들(10)은 상기 서버(30) 또는 상기 사용자단말기들(20)로부터 갑상선기능이상증에 관한 예측결과를 수신할 수 있다.The plurality of heart rate measurement devices 10 may receive prediction results regarding thyroid dysfunction from the server 30 or the user terminals 20.

(3) 심박수 측정 디바이스의 종류(3) Types of heart rate measurement devices

상기 심박수 측정 디바이스는 웨어러블 워치(wearable watch), 웨어러블 밴드(wearable band) 또는 패치 타입 디바이스(patch type device)일 수 있다.The heart rate measurement device may be a wearable watch, a wearable band, or a patch type device.

(3) 사용자 단말기의 기능(3) Functions of the user terminal

상기 복수의 사용자 단말기들(20)은 각종 네트워크를 통해 상기 서버(30)에 정보를 전송하고, 또 상기 서버(30)로부터 정보를 수신한다. The plurality of user terminals 20 transmit information to the server 30 and receive information from the server 30 through various networks.

상기 복수의 사용자 단말기들(20)은 상기 심박수 측정 디바이스(10)로부터 상기 측정된 심박수들을 수신하고 또 상기 수신된 심박수들을 저장할 수 있다.The plurality of user terminals 20 may receive the measured heart rates from the heart rate measurement device 10 and store the received heart rates.

상기 복수의 사용자 단말기들(20)은 상기 저장된 심박수들을 전처리할 수 있다. 예를 들어, 상기 사용자 단말기들(20)은 저장되어 있는 모든 심박수들 중에서 휴지기에 측정된 심박수들을 선별할 수 있다. 다른 예를 들어, 상기 사용자 단말기들(20)은 특정한 날을 기준으로 상기 저장되어 있는 모든 심박수들 중에서 전술한 구간 심박수들을 선별할 수 있다. 또 다른 예를 들어, 상기 사용자 단말기들(20)은 상기 특정한 날을 기준으로 선별된 구간 심박수들의 평균을 산출할 수 있다. 또 다른 예를 들어, 상기 사용자 단말기들(20)은 상기 특정한 날을 기준으로 선별된 구간 심박수들의 표준편차, 상대 표준편차, 왜도, 첨도 등을 산출할 수 있다.The plurality of user terminals 20 may preprocess the stored heart rates. For example, the user terminals 20 may select heart rates measured in the resting state from among all stored heart rates. For another example, the user terminals 20 may select the above-described zone heart rates from all the stored heart rates based on a specific day. As another example, the user terminals 20 may calculate the average of heart rates in selected sections based on the specific day. For another example, the user terminals 20 may calculate the standard deviation, relative standard deviation, skewness, kurtosis, etc. of the heart rates in selected sections based on the specific day.

상기 복수의 사용자 단말기들(20)은 사용자의 호르몬 농도의 검사일을 획득할 수 있으며, 상기 검사일에 대응되는 검사결과(예를 들어, 호르몬 농도 및/또는 의사의 진단 결과 등)를 획득할 수 있다. 상기 복수의 사용자 단말기들(20)은 상기 검사일 및 그에 대한 검사결과를 사용자의 입력에 의해 획득할 수도 있으며, 네트워크를 통해 외부로부터 수신하는 방식으로 획득할 수도 있다.The plurality of user terminals 20 can obtain the test date of the user's hormone concentration and obtain test results (for example, hormone concentration and/or doctor's diagnosis results, etc.) corresponding to the test date. . The plurality of user terminals 20 may obtain the inspection date and its inspection results through user input, or may obtain them by receiving them from the outside through a network.

상기 복수의 사용자 단말기들(20)은 상기 저장된 심박수들 및/또는 상기 전처리된 결과들을 상기 서버(30)로 전송할 수 있다.The plurality of user terminals 20 may transmit the stored heart rates and/or the preprocessed results to the server 30 .

상기 복수의 사용자 단말기들(20)은 상기 검사일 및 검사결과를 상기 서버(30)로 전송할 수 있다.The plurality of user terminals 20 may transmit the test date and test results to the server 30.

상기 복수의 사용자 단말기들(20)은 상기 서버(30)에 의해 처리된 갑상선기능이상증에 관한 예측결과를 상기 서버(30)로부터 수신할 수 있다.The plurality of user terminals 20 may receive prediction results regarding thyroid dysfunction processed by the server 30 from the server 30 .

(4) 사용자 단말기의 종류(4) Type of user terminal

상기 복수의 사용자 단말기들(20) 스마트폰, 태블릿 PC, 랩탑, 데스크탑 중 적어도 하나일 수 있다.The plurality of user terminals 20 may be at least one of a smartphone, tablet PC, laptop, and desktop.

(5) 서버의 기능(5) Server functions

상기 서버(30)는 각종 네트워크를 통해 상기 복수의 사용자 단말기들(20)에 정보를 전송하고, 또 상기 복수의 사용자 단말기들(20)로부터 정보를 수신한다.The server 30 transmits information to the plurality of user terminals 20 through various networks, and also receives information from the plurality of user terminals 20.

상기 서버(30)는 상기 복수의 사용자 단말기들(20)로부터 측정된 심박수들의전부 또는 일부를 수신할 수 있다. 이때, 상기 서버(30)는 상기 수신된 심박수들을 전처리할 수 있다.The server 30 may receive all or part of the heart rates measured from the plurality of user terminals 20. At this time, the server 30 may preprocess the received heart rates.

또는 상기 서버(30)는 상기 복수의 사용자 단말기들(20)로부터 상기 사용자 단말기(20)에 의해 전처리된 결과를 수신할 수 있다.Alternatively, the server 30 may receive a result preprocessed by the user terminal 20 from the plurality of user terminals 20 .

또한, 상기 서버(30)는 사용자들 각각에 대하여 호르몬 농도 검사일 및 그에 대한 검사결과를 획득할 수 있다. 상기 검사일 및 검사결과는 상기 사용자 단말기(20)로부터 수신되거나 혹은 다른 외부 장치로부터 수신될 수 있다.Additionally, the server 30 can obtain the hormone concentration test date and test results for each user. The test date and test results may be received from the user terminal 20 or from another external device.

상기 서버(30)는 상기 전처리된 결과 등에 기초하여 대상의 갑상선기능이상증에 대한 예측결과를 획득할 수 있다.The server 30 may obtain a prediction result for the target's thyroid dysfunction based on the preprocessed results, etc.

상기 서버(30)는 상기 갑상선기능이상증에 대한 예측결과를 상기 심박수 측정 디바이스(10) 및/또는 상기 복수의 사용자 단말기들(20)로 전송할 수 있다.The server 30 may transmit the prediction result for thyroid dysfunction to the heart rate measurement device 10 and/or the plurality of user terminals 20.

(6) 시스템의 소프트웨어적 구성(6) Software configuration of the system

상기 시스템(1)이 동작하기 위하여, 몇몇 소프트웨어적 구성들이 필요하다.In order for the system 1 to operate, several software configurations are required.

상기 사용자 단말기들(20)과 상기 서버(30) 사이의 통신을 수행하기 위하여, 상기 복수의 사용자 단말기들(20)에 단말기 소프트웨어가 설치될 필요가 있으며, 상기 서버(30)에 서버 소프트웨어가 설치될 필요가 있다. In order to perform communication between the user terminals 20 and the server 30, terminal software needs to be installed on the plurality of user terminals 20, and server software must be installed on the server 30. needs to be

상기 전처리된 결과에 기초하여 갑상선기능이상증을 예측하기 위한 복수의 예측모델들이 사용될 수 있다.A plurality of prediction models can be used to predict thyroid dysfunction based on the preprocessed results.

상기 복수의 예측모델들은 상기 서버(30)에 설치되는 소프트웨어에 의해 구동될 수도 있다. 또는 상기 복수의 예측모델들은 상기 사용자 단말기들(20)에 설치되는 단말기 소프트웨어에 의해 구동될 수 있다. 또는 상기 복수의 학습모델들의 일부는 상기 사용자 단말기들(20)에 의해 실행되고, 나머지 일부는 상기 서버(30)에 의해 실행될 수 있다.The plurality of prediction models may be driven by software installed on the server 30. Alternatively, the plurality of prediction models may be driven by terminal software installed on the user terminals 20. Alternatively, some of the plurality of learning models may be executed by the user terminals 20 and the remaining portions may be executed by the server 30.

(7) 심박수 측정 디바이스의 구성요소(7) Components of heart rate measurement device

도 2는 본 출원에 의해 개시되는 심박수 측정 디바이스의 블록도이다.Figure 2 is a block diagram of a heart rate measurement device disclosed by the present application.

도 2를 참조하면, 본 출원에 의해 개시되는 심박수 측정 디바이스(10)는 출력부(110), 통신부(120), 메모리(130), 심박수 측정부(140) 및 컨트롤러(150)를 포함한다.Referring to FIG. 2, the heart rate measurement device 10 disclosed by the present application includes an output unit 110, a communication unit 120, a memory 130, a heart rate measurement unit 140, and a controller 150.

출력부(110)는 컨트롤러(150)의 제어 명령에 따라 각종 정보를 출력한다. 일 실시예에 따르면, 출력부(110)는 사용자에게 정보를 시각적으로 출력하는 디스플레이(112)를 포함할 수 있다. 또는, 도면에 도시하지는 않았지만, 사용자에게 정보를 청각적으로 출력하는 스피커, 사용자에게 정보를 촉각적으로 출력하는 진동모터를 포함할 수 있다.The output unit 110 outputs various information according to control commands from the controller 150. According to one embodiment, the output unit 110 may include a display 112 that visually outputs information to the user. Alternatively, although not shown in the drawing, it may include a speaker that audibly outputs information to the user and a vibration motor that tactilely outputs information to the user.

통신부(120)는 무선 통신 모듈 및/또는 유선 통신 모듈을 포함할 수 있다. 여기서, 무선 통신 모듈은 와이파이(Wi-Fi) 통신 모듈, 셀룰러 통신 모듈 등을 포함할 수 있다. The communication unit 120 may include a wireless communication module and/or a wired communication module. Here, the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.

메모리(130)는 컨트롤러(150)에 의해 읽힐 수 있는 실행 코드, 처리된 결과값 및 필요한 데이터 등을 저장한다. 상기 메모리(130)는 HDD(Hard Disk Drive), SSD(Solid State Disk), SDD(Silicon Disk Drive), ROM, RAM 등을 포함할 수 있다. 메모리(130)는 전술한 단말기 소프트웨어를 저장할 수 있고, 전술한 다양한 전처리 알고리즘 및/또는 학습모델들을 구현하기 위한 실행코드들을 저장할 수 있다. 나아가, 메모리(130)는 상기 심박수 측정부(140)를 통해 획득되는 대상의 심박수 및 그 심박수의 측정 시각 등을 저장할 수 있다. The memory 130 stores executable codes that can be read by the controller 150, processed result values, and necessary data. The memory 130 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc. The memory 130 may store the above-described terminal software, and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above. Furthermore, the memory 130 may store the heart rate of the target obtained through the heart rate measuring unit 140 and the measurement time of the heart rate.

심박수 측정부(140)는 대상(사용자)의 분당 심박수를 측정할 수 있다. 대상의 분당 심박수를 측정하는 방식은 널리 알려져 있으므로 이에 대한 자세한 설명은 생략한다.The heart rate measurement unit 140 can measure the heart rate per minute of the subject (user). The method of measuring a subject's heart rate per minute is widely known, so a detailed description thereof will be omitted.

컨트롤러(150)는 적어도 하나의 프로세서를 포함할 수 있다. 이때, 각각의 프로세서는 메모리(130)에 저장된 적어도 하나의 명령어를 실행시킴으로써, 소정의 동작을 실행할 수 있다. 구체적으로, 컨트롤러(150)는 상기 심박수 측정 디바이스(10)에서 구동되는 단말기 소프트웨어, 전처리 알고리즘 및/또는 학습모델에 따라 정보를 처리할 수 있다. 한편, 상기 컨트롤러(150)는 심박수 측정 디바이스(10)의 전반적인 동작을 제어한다.Controller 150 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 130. Specifically, the controller 150 may process information according to terminal software running on the heart rate measurement device 10, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 150 controls the overall operation of the heart rate measurement device 10.

도면에 도시하지는 않았지만, 상기 심박수 측정 디바이스(10)는 사용자 입력부를 포함할 수 있다. 상기 심박수 측정 디바이스(10)는 상기 사용자 입력부를 통해 심박수 측정 디바이스(10)의 동작에 필요한 여러 정보를 사용자로부터 입력 받을 수 있다.Although not shown in the drawing, the heart rate measuring device 10 may include a user input unit. The heart rate measurement device 10 may receive various information required for operation of the heart rate measurement device 10 from the user through the user input unit.

(8) 사용자 단말기의 구성요소(8) Components of the user terminal

도 3은 본 출원에 의해 개시되는 사용자 단말기의 블록도이다.Figure 3 is a block diagram of a user terminal disclosed by the present application.

도 3을 참조하면, 본 출원에 의해 개시되는 사용자 단말기(20)는 출력부(210), 통신부(220), 메모리(230) 및 컨트롤러(250)를 포함한다.Referring to FIG. 3, the user terminal 20 disclosed by the present application includes an output unit 210, a communication unit 220, a memory 230, and a controller 250.

출력부(210)는 컨트롤러(250)의 제어 명령에 따라 각종 정보를 출력한다. 일 실시예에 따르면, 출력부(210)는 사용자에게 정보를 시각적으로 출력하는 디스플레이(212)를 포함할 수 있다. 또는, 도면에 도시하지는 않았지만, 사용자에게 정보를 청각적으로 출력하는 스피커, 사용자에게 정보를 촉각적으로 출력하는 진동모터를 포함할 수 있다.The output unit 210 outputs various information according to control commands from the controller 250. According to one embodiment, the output unit 210 may include a display 212 that visually outputs information to the user. Alternatively, although not shown in the drawing, it may include a speaker that audibly outputs information to the user and a vibration motor that tactilely outputs information to the user.

통신부(220)는 무선 통신 모듈 및/또는 유선 통신 모듈을 포함할 수 있다. 여기서, 무선 통신 모듈은 와이파이(Wi-Fi) 통신 모듈, 셀룰러 통신 모듈 등을 포함할 수 있다. The communication unit 220 may include a wireless communication module and/or a wired communication module. Here, the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.

메모리(230)는 컨트롤러(250)에 의해 읽힐 수 있는 실행 코드, 처리된 결과값 및 필요한 데이터 등을 저장한다. 상기 메모리(230)는 HDD(Hard Disk Drive), SSD(Solid State Disk), SDD(Silicon Disk Drive), ROM, RAM 등을 포함할 수 있다. 메모리(230)는 전술한 단말기 소프트웨어를 저장할 수 있고, 전술한 다양한 전처리 알고리즘 및/또는 학습모델들을 구현하기 위한 실행코드들을 저장할 수 있다. The memory 230 stores executable codes that can be read by the controller 250, processed results, and necessary data. The memory 230 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc. The memory 230 may store the above-described terminal software, and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above.

컨트롤러(240)는 적어도 하나의 프로세서를 포함할 수 있다. 이때, 각각의 프로세서는 메모리(230)에 저장된 적어도 하나의 명령어를 실행시킴으로써, 소정의 동작을 실행할 수 있다. 구체적으로, 컨트롤러(240)는 상기 사용자 단말기(20)에서 구동되는 단말기 소프트웨어, 전처리 알고리즘 및/또는 학습모델에 따라 정보를 처리할 수 있다. 한편, 상기 컨트롤러(240)는 상기 사용자 단말기(20)의 전반적인 동작을 제어한다.Controller 240 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 230. Specifically, the controller 240 may process information according to terminal software running on the user terminal 20, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 240 controls the overall operation of the user terminal 20.

도면에 도시하지는 않았지만, 상기 사용자 단말기(20)는 사용자 입력부를 포함할 수 있다. 상기 사용자 단말기(20)는 상기 사용자 입력부를 통해 사용자 단말기(20)의 동작에 필요한 여러 정보를 사용자로부터 입력 받을 수 있다.Although not shown in the drawing, the user terminal 20 may include a user input unit. The user terminal 20 can receive various information required for operation of the user terminal 20 from the user through the user input unit.

(9) 서버의 구성요소(9) Server components

도 4는 본 출원에 의해 개시되는 서버의 블록도이다.Figure 4 is a block diagram of the server disclosed by this application.

도 4를 참조하면, 본 출원에 의해 개시되는 서버(30)는 통신부(310), 메모리(320) 및 컨트롤러(330)를 포함한다.Referring to FIG. 4, the server 30 disclosed by the present application includes a communication unit 310, a memory 320, and a controller 330.

통신부(310)는 무선 통신 모듈 및/또는 유선 통신 모듈을 포함할 수 있다. 여기서, 무선 통신 모듈은 와이파이(Wi-Fi) 통신 모듈, 셀룰러 통신 모듈 등을 포함할 수 있다. The communication unit 310 may include a wireless communication module and/or a wired communication module. Here, the wireless communication module may include a Wi-Fi communication module, a cellular communication module, etc.

메모리(320)는 컨트롤러(330)에 의해 읽힐 수 있는 실행 코드, 처리된 결과값 및 필요한 데이터 등을 저장한다. 상기 메모리(320)는 HDD(Hard Disk Drive), SSD(Solid State Disk), SDD(Silicon Disk Drive), ROM, RAM 등을 포함할 수 있다. 메모리(320)는 전술한 서버 소프트웨어를 저장할 수 있고, 전술한 다양한 전처리 알고리즘 및/또는 학습모델들을 구현하기 위한 실행코드들을 저장할 수 있다. 나아가, 메모리(320)는 상기 사용자 단말기(20)로부터 사용자별 심박수 및/또는 그에 대한 전처리 결과 등을 저장할 수 있다. The memory 320 stores executable codes that can be read by the controller 330, processed results, and necessary data. The memory 320 may include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, etc. The memory 320 may store the above-described server software and may store execution codes for implementing the various preprocessing algorithms and/or learning models described above. Furthermore, the memory 320 may store heart rate and/or pre-processing results for each user from the user terminal 20.

컨트롤러(330)는 적어도 하나의 프로세서를 포함할 수 있다. 이때, 각각의 프로세서는 메모리(320)에 저장된 적어도 하나의 명령어를 실행시킴으로써, 소정의 동작을 실행할 수 있다. 구체적으로, 컨트롤러(330)는 상기 서버(30)에서 구동되는 서버 소프트웨어, 전처리 알고리즘 및/또는 학습모델에 따라 정보를 처리할 수 있다. 한편, 상기 컨트롤러(330)는 상기 서버(30)의 전반적인 동작을 제어한다.The controller 330 may include at least one processor. At this time, each processor can execute a predetermined operation by executing at least one instruction stored in the memory 320. Specifically, the controller 330 may process information according to server software running on the server 30, a preprocessing algorithm, and/or a learning model. Meanwhile, the controller 330 controls the overall operation of the server 30.

4. 갑상선기능이상증의 예측방법4. How to predict thyroid dysfunction

본 출원에 의해 개시되는 갑상선기능이상증(thyroid dysfunction)을 예측하는 방법은 전술한 시스템(1)에 의해 수행될 수 있다.The method for predicting thyroid dysfunction disclosed by the present application can be performed by the above-described system 1.

도 5는 본 출원에 의해 개시되는 갑상선기능이상증을 예측하는 방법을 설명하기 위한 흐름도이다.Figure 5 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.

도 5를 참고하면, 본 출원에 의해 개시되는 대상(subject)에 대한 갑상선기능이상증을 예측하는 방법은 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)을 포함한다.Referring to FIG. 5, the method for predicting thyroid dysfunction for a subject disclosed by the present application acquires a trigger signal (S100), and obtains heart rates in the section corresponding to the target date determined by the trigger signal. Obtaining (S110), preprocessing the section heart rates corresponding to the target date (S120), obtaining the hormone concentration corresponding to the reference date (S130), obtaining preprocessing results for the section heart rates corresponding to the reference date (S140), obtain the difference between the preprocessing results of the section heart rates corresponding to the target date and the preprocessing result of the section heart rates corresponding to the reference date (S150), calculate the obtained difference and the hormone concentration corresponding to the reference date It includes processing using a thyroid dysfunction prediction model and obtaining results on thyroid dysfunction for the subject (S160).

트리거 신호의 획득(S100)Acquisition of trigger signal (S100)

트리거 신호가 획득될 수 있다(S100).A trigger signal may be obtained (S100).

여기서, 트리거 신호라 함은 대상의 갑상선기능이상증에 대한 예측결과를 획득할 것을 요청하는 신호를 의미한다.Here, the trigger signal refers to a signal requesting to obtain a prediction result for the subject's thyroid dysfunction.

상기 트리거 신호는 상기 사용자의 입력에 의해 생성될 수 있다. 예를 들어, 사용자가 사용자 단말기(20)의 사용자 입력부를 통해 갑상선기능이상증에 대한 예측결과를 획득할 것을 요청하고, 상기 사용자의 요청에 응답하여 상기 사용자 단말기(20)는 상기 트리거 신호를 생성할 수 있다. 다른 예를 들어, 사용자가 상기 사용자 단말기(20)와 통신을 수행하고 있는 심박수 측정 디바이스(10)의 사용자 입력부를 통해 갑상선기능이상증에 대한 예측결과를 획득할 것을 요청하고, 상기 사용자의 요청에 응답하여 상기 심박수 측정 디바이스(10) 또는 상기 사용자 단말기(20)는 상기 트리거 신호를 생성할 수 있다.The trigger signal may be generated by the user's input. For example, the user requests to obtain a prediction result for thyroid dysfunction through the user input unit of the user terminal 20, and in response to the user's request, the user terminal 20 generates the trigger signal. You can. For another example, the user requests to obtain a prediction result for thyroid dysfunction through the user input unit of the heart rate measurement device 10 that is communicating with the user terminal 20, and responds to the user's request. Thus, the heart rate measurement device 10 or the user terminal 20 can generate the trigger signal.

또는, 상기 트리거 신호는 상기 시스템(1)에 의해 사용자의 요청과 무관하게 자동으로 생성될 수 있다. 예를 들어, 상기 심박수 측정 디바이스(10), 상기 사용자 단말기(20) 또는 상기 서버(30)에 설치되어 있는 소프트웨어는 미리 정해진 규칙에 따라 상기 트리거 신호를 생성할 수 있다. 만약 상기 심박수 측정 디바이스(10) 또는 상기 서버(30)에 의해 상기 트리거 신호가 생성된 경우, 상기 생성된 트리거 신호는 상기 사용자 단말기(20)로 전송될 수 있다.Alternatively, the trigger signal may be automatically generated by the system 1 regardless of the user's request. For example, software installed on the heart rate measuring device 10, the user terminal 20, or the server 30 may generate the trigger signal according to a predetermined rule. If the trigger signal is generated by the heart rate measurement device 10 or the server 30, the generated trigger signal may be transmitted to the user terminal 20.

트리거 신호에 기초한 대상일의 결정(S110)Determination of target date based on trigger signal (S110)

상기 트리거 신호에 의해 대상일이 결정될 수 있다. 예를 들어, 상기 트리거 신호가 생성된 날이 상기 대상일로 결정될 수 있다. 다른 예를 들어, 상기 트리거 신호가 생성될 때, 상기 사용자의 입력 혹은 상기 시스템(1)에 의해 특정한 날(specific day)이 함께 지정(designate)될 수 있으며, 이 경우 상기 지정된 특정한 날이 상기 대상일로 결정될 수 있다.The target date may be determined by the trigger signal. For example, the day on which the trigger signal was generated may be determined as the target date. For another example, when the trigger signal is generated, a specific day may be designated by the user's input or the system 1, in which case the designated specific day is the target. It can be decided by work.

대상일에 대응되는 구간 심박수들의 획득(S110)Acquisition of section heart rates corresponding to the target date (S110)

미리 정해진 기간에 따라 상기 대상일에 대응되는 구간 심박수들이 획득될 수 있다(S110).Zone heart rates corresponding to the target date may be obtained according to a predetermined period (S110).

상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.The predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23. , 24, 25, 26, 27, 28, 29, and 30 days.

상기 미리 정해진 기간이 10일이고 상기 대상일이 07월 21일인 경우, 상기 구간 심박수들은 07월 21일에 대응되는 휴지기 심박수들, 07월 20일에 대응되는 휴지기 심박수들, 07월 19일에 대응되는 휴지기 심박수들, 07월 18일에 대응되는 휴지기 심박수들, 07월 17일에 대응되는 휴지기 심박수들, 07월 16일에 대응되는 휴지기 심박수들, 07월 15일에 대응되는 휴지기 심박수들, 07월 14일에 대응되는 휴지기 심박수들, 07월 13일에 대응되는 휴지기 심박수들, 07월 12일에 대응되는 휴지기 심박수들이 상기 구간 심박수들로 획득된다.If the predetermined period is 10 days and the target date is July 21st, the zone heart rates correspond to resting heart rates corresponding to July 21st, resting heart rates corresponding to July 20th, and July 19th. Resting heart rates corresponding to July 18, Resting heart rates corresponding to July 17, Resting heart rates corresponding to July 16, Resting heart rates corresponding to July 15, 07 Resting heart rates corresponding to the 14th of the month, resting heart rates corresponding to the 13th of July, and resting heart rates corresponding to the 12th of July are obtained as the zone heart rates.

상기 구간 심박수들은 상기 결정된 대상일 및 상기 미리 정해진 기간에 의해 결정되는 날들에 대응되어 저장된 모든 심박수들 중에서 선택된다. 상기 심박수들에 대응된 측정 시각에 기초하여, 상기 측정 시각이 대상(subject)(즉, 사용자)의 휴지기 내에 있는 심박수들이 휴지기 심박수들로 선택된다.The zone heart rates are selected from all heart rates stored corresponding to days determined by the determined target date and the predetermined period. Based on the measurement times corresponding to the heart rates, heart rates at which the measurement times are within the resting phase of the subject (i.e., the user) are selected as resting heart rates.

상기 휴지기는 다양한 방식에 의해 결정될 수 있다.The resting period can be determined in various ways.

예를 들어, 상기 휴지기는 상기 심박수 측정 디바이스(10) 또는 상기 사용자 단말기(20)에 내장되어 있는 모션 센서를 통해 사용자의 움직임이 기준값 이하인 구간으로 결정될 수 있다.For example, the rest period may be determined as a section in which the user's movement is below a reference value through the heart rate measurement device 10 or a motion sensor built into the user terminal 20.

다른 예를 들어, 상기 휴지기는 상기 심박수 측정 디바이스(10) 또는 상기 사용자 단말기(20)에 내장되어 있는 다양한 센서들을 통해 사용자가 휴식 또는 수면 중이라고 판단된 구간으로 결정될 수 있다.For another example, the rest period may be determined as a section in which the user is determined to be resting or sleeping through various sensors built into the heart rate measurement device 10 or the user terminal 20.

또 다른 예를 들어, 상기 휴지기는 상기 심박수 측정 디바이스(10) 또는 상기 사용자 단말기(20)를 통해 사용자가 직접 입력한 수면 기간(수면 시작 시각 및 수면 종료 시각)에 의해 결정될 수 있다.For another example, the rest period may be determined by the sleep period (sleep start time and sleep end time) directly input by the user through the heart rate measurement device 10 or the user terminal 20.

대상일에 대응되는 구간 심박수들의 전처리(S120)Preprocessing of section heart rates corresponding to the target date (S120)

상기 대상일에 대응되는 구간 심박수들이 전처리될 수 있다(S120).The heart rates of the sections corresponding to the target date may be preprocessed (S120).

상기 대상일에 대응되는 구간 심박수들의 평균이 계산될 수 있다.The average of heart rates in the sections corresponding to the target date may be calculated.

또는, 상기 대상일에 대응되는 구간 심박수들의 조화평균이 계산될 수 있다. Alternatively, the harmonic average of heart rates in the section corresponding to the target date may be calculated.

또는, 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관련된 다양한 지표들이 산출될 수 있다. 예를 들어, 상기 대상일에 대응되는 구간 심박수들의 표준편차(standard deviation)가 계산될 수 있다. 다른 예를 들어, 상기 대상일에 대응되는 구간 심박수들의 왜도(skewness)가 계산될 수 있다. 또 다른 예를 들어, 상기 대상일에 대응되는 구간 심박수들의 첨도(kurtosis)가 계산될 수 있다. 또 다른 예를 들어, 상기 대상일에 대응되는 구간 심박수들의 상대 표준편차(relative standard deviation)가 계산될 수 있다.Alternatively, various indicators related to the distribution of heart rates in the sections corresponding to the target date may be calculated. For example, the standard deviation of heart rates in the section corresponding to the target date may be calculated. For another example, the skewness of heart rates in the sections corresponding to the target date may be calculated. As another example, kurtosis of heart rate intervals corresponding to the target date may be calculated. As another example, the relative standard deviation of heart rates in the zones corresponding to the target date may be calculated.

또는, 상기 대상일에 대응되는 구간 심박수들의 평균, 상기 대상일에 대응되는 구간 심박수들의 조화평균 및 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 중 둘 이상이 계산될 수 있다.Alternatively, two or more of the average of section heart rates corresponding to the target date, the harmonic average of section heart rates corresponding to the target day, and the index regarding the distribution of section heart rates corresponding to the reference date may be calculated.

일 실시예에 따르면, 대상일에 대응되는 구간 심박수들이 전처리될 때에 구간 심박수에 포함된 일부의 심박수값이 노이즈로 제거될 수 있다. 예를 들어, 구간 평균 심박수는 노이즈가 제거된 구간 심박수를 이용해서 계산될 수 있다. 예를 들어, 구간 심박수들의 분포는 노이즈가 제거된 구간 심박수를 이용해서 계산될 수 있다.According to one embodiment, when the zone heart rates corresponding to the target day are preprocessed, some heart rate values included in the zone heart rates may be removed as noise. For example, the section average heart rate can be calculated using the section heart rate from which noise has been removed. For example, the distribution of zone heart rates can be calculated using the zone heart rates from which noise has been removed.

본 명세서에서 언급하는 "노이즈"는 미리 정해진 기준에 따라 사용하지 않기로 결정되는 심박수값을 의미하는 것이지, 실제 심박수를 측정한 값이 아님을 확인하고 제거되는 심박수값은 아니다. “Noise” referred to in this specification refers to a heart rate value that is determined not to be used according to a predetermined standard, and is not a heart rate value that is removed after confirming that it is not a measurement of the actual heart rate.

구간 심박수에서 노이즈를 제거하는 방법은, 미리 설정된 구간 길이의 심박수들의 중앙값을 기준으로 임계치 이상의 차이가 나는 심박수를 이용하지 않는 형태일 수 있다.A method of removing noise from the section heart rate may be in the form of not using heart rates that differ by more than a threshold based on the median of the heart rates of a preset section length.

미리 설정된 구간 길이는 시간에 대응될 수 있다. 예를 들어, 미리 설정된 구간 길이가 10분인 경우, 10분동안 획득된 모든 심박수 값을 기준으로 중앙값을 계산하고, 계산된 중앙값과 각각의 심박수 값을 비교하여 노이즈로 제거될 심박수를 결정할 수 있다. The preset section length may correspond to time. For example, if the preset section length is 10 minutes, the median value is calculated based on all heart rate values obtained for 10 minutes, and the heart rate to be removed as noise can be determined by comparing the calculated median value with each heart rate value.

이 경우, 10분에 실제로 획득되는 심박수값의 개수는 심박수 측정 디바이스의 스펙에 따라 변경될 수 있고, 또는, 심박수의 측정 환경에 따라 변경될 수 있다. In this case, the number of heart rate values actually obtained in 10 minutes may change depending on the specifications of the heart rate measurement device or may change depending on the heart rate measurement environment.

구간 심박수들의 노이즈를 제거함에 있어서, 전체 구간 심박수를 미리 설정된 구간 길이로 균등하게 나누어 분석을 진행할 수도 있으나, 연속적인 변화를 고려하여 심박수의 노이즈를 제거하기 위해 미리 설정된 구간 길이로 나눠진 분석 구간들은 서로 중첩 영역을 가질 수 있다. In removing noise from section heart rates, the analysis may be performed by dividing the entire section heart rate equally into preset section lengths. However, in order to remove heart rate noise by considering continuous changes, the analysis sections divided by preset section lengths are connected to each other. Can have overlapping areas.

구체적인 예를 들어, 1시간의 심박수가 확보되어 있고, 미리 설정된 시간 구간이 10분인 경우, 0분부터 10분까지의 심박수를 제1 분석 구간으로 설정하고, 10분부터 20분까지의 심박수를 제2 분석 구간으로 설정하는 방식으로, 6(60분/10분)개의 분석 구간들로 노이즈 분석을 진행할 수 있지만, 0분부터 10분까지의 심박수를 제1 분석 구간으로 설정하고, 5분부터 15분까지의 심박수를 제2 분석 구간으로 설정하는 방식으로, 11((60분/5분)-1)개의 분석 구간들로 노이즈 분석을 진행할 수도 있다. For a specific example, if the heart rate for 1 hour is secured and the preset time section is 10 minutes, the heart rate from 0 minutes to 10 minutes is set as the first analysis section, and the heart rate from 10 minutes to 20 minutes is set as the first analysis section. By setting it to 2 analysis sections, you can perform noise analysis with 6 (60 minutes/10 minutes) analysis sections, but set the heart rate from 0 to 10 minutes as the first analysis section, and set the heart rate from 0 to 10 minutes as the first analysis section, and set the heart rate from 0 to 10 minutes as the first analysis section. By setting the heart rate up to the minute as the second analysis section, noise analysis can be performed with 11 ((60 minutes/5 minutes)-1) analysis sections.

또는, 구간 심박수에서 노이즈를 제거함에 있어서, 단일 심박수 값의 개수에 기초하여 설정되는 구간의 심박수들의 중앙값을 기준으로 임계치 이상의 차이가 나는 심박수를 이용하지 않는 형태일 수 있다. 예를 들어, 단일 심박수 값의 개수가 10개인 경우, 연속하는 10개의 심박수가 30분 동안 획득된 것인지 10분 동안 획득된 것인지 고려하지 않고, 그를 기준으로 중앙값을 계산하고, 계산된 중앙값과 각각의 심박수 값을 비교하여 노이즈로 제거될 심박수를 결정할 수 있다. Alternatively, in removing noise from the section heart rate, heart rates that differ by more than a threshold based on the median of the heart rates of the section set based on the number of single heart rate values may not be used. For example, if the number of single heart rate values is 10, the median value is calculated based on 10 consecutive heart rates without considering whether they were acquired for 30 minutes or 10 minutes, and the calculated median value and each By comparing heart rate values, you can determine the heart rate to be removed as noise.

구간 심박수들의 노이즈를 제거하는 형태는 서술된 방식이 아닌, 통상적으로 사용하는 방식이 적용될 수도 있다. 또한, 구간 심박수들의 노이즈를 제거하는 형태는 대상일의 구간 심박수들을 전처리할 때, 기준일의 구간 심박수들을 전처리할 때는 물론, 학습 데이터를 생성할 때에도 사용될 수 있다.A commonly used method, rather than the described method, may be applied to remove noise from section heart rates. In addition, a form of removing noise from section heart rates can be used when preprocessing section heart rates of a target day, when preprocessing section heart rates of a reference day, as well as when generating learning data.

상기 대상일에 대응되는 구간 심박수들의 다양한 전처리 결과들 중 후술하는 예측모델의 학습에 사용된 입력값에 관련된 전처리 결과들만 획득될 수 있다. 예를 들어, 상기 예측모델의 학습에 대상일과 기준일에 대한 구간 심박수들의 평균값의 변화율 및 상대 표준편차의 변화량이 사용된 경우, 상기 대상일에 대응되는 구간 심박수들로부터 상기 대상일에 대응되는 구간 심박수들의 평균 및 표준편차만 획득될 수 있다. 다만, 예측모델의 학습에 사용된 입력값과 무관하게 전술한 모든 또는 일부의 전처리 결과들이 획득될 수도 있다.Among the various preprocessing results of the heart rates in the section corresponding to the target date, only preprocessing results related to the input values used for learning the prediction model, which will be described later, can be obtained. For example, when the rate of change and the change in the relative standard deviation of the average value of section heart rates for the target date and the reference date are used in learning the prediction model, the section heart rate corresponding to the target date is calculated from the section heart rates corresponding to the target date. Only the mean and standard deviation of the values can be obtained. However, all or some of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.

기준일의 결정(S130)Determination of base date (S130)

상기 대상에 대한 기준일이 결정될 수 있다(S130).The reference date for the object may be determined (S130).

상기 기준일은 상기 대상이 호르몬 검사를 받은 적어도 하나 이상의 검사일 중에서 선택될 수 있다. 즉, 상기 기준일은 상기 대상의 과거 호르몬 검사일들 중에서 선택된다.The reference date may be selected from at least one test date on which the subject underwent a hormone test. That is, the reference date is selected from among the subject's past hormone test dates.

상기 기준일은 상기 대상의 호르몬 농도가 대응되어 저장된 적어도 하나 이상의 날들 중에서 선택될 수 있다. 즉, 상기 기준일은 상기 대상의 과거 호르몬 농도가 저장되어 있는 날들 중에서 선택될 수 있다.The reference date may be selected from at least one or more days on which the subject's hormone concentration is stored correspondingly. That is, the reference date may be selected from days in which the subject's past hormone concentration is stored.

상기 기준일은 오직 하나만 선택될 수 있으나, 이에 국한되지 않는다. 즉, 상기 기준일은 복수개가 선택될 수 있다.Only one such reference date may be selected, but is not limited thereto. That is, multiple reference dates may be selected.

복수의 검사일 혹은 복수의 호르몬 농도가 저장되어 있는 날들 중에서 오직 하나의 기준일이 선택되는 경우, 상기 기준일은 상기 대상일을 기준으로 가장 가까운 날로 결정될 수 있다.When only one reference date is selected from among multiple test dates or days on which multiple hormone concentrations are stored, the reference date may be determined as the closest day based on the target date.

기준일에 대응되는 호르몬 농도의 획득(S130)Obtaining hormone concentrations corresponding to the reference date (S130)

상기 결정된 기준일에 대응되는 호르몬 농도가 획득될 수 있다.The hormone concentration corresponding to the determined reference date can be obtained.

상기 호르몬 농도는 T4의 농도, free T4의 농도, T3의 농도, free T3의 농도, TSH의 농도 및 TRH의 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수 있다. The hormone concentration may be one or a combination thereof selected from the group consisting of T4 concentration, free T4 concentration, T3 concentration, free T3 concentration, TSH concentration, and TRH concentration.

상기 결정된 기준일에 대응되는 호르몬 농도들 중에서, 후술하는 예측모델의 학습에 사용된 호르몬 농도값들만 선택적으로 선택될 수 있다. 예를 들어, 결정된 기준일에 대응된 호르몬 농도가 T4의 농도, free T4의 농도, T3의 농도, free T3의 농도, TSH의 농도 및 TRH의 농도를 모두 포함하고 있을 때, 상기 예측모델의 학습에 free T4 및 TSH의 농도만 사용된 경우, free T3 및 TSH의 농도만 획득될 수 있다.Among the hormone concentrations corresponding to the determined reference date, only the hormone concentration values used for learning the prediction model described later can be selectively selected. For example, when the hormone concentration corresponding to the determined reference date includes all of the concentration of T4, concentration of free T4, concentration of T3, concentration of free T3, concentration of TSH, and concentration of TRH, learning of the prediction model If only the concentrations of free T4 and TSH are used, only the concentrations of free T3 and TSH can be obtained.

만약, 복수의 기준일이 결정된 경우, 복수의 기준일 각각에 대응되는 호르몬 농도가 획득될 수 있다.If multiple reference dates are determined, hormone concentrations corresponding to each of the multiple reference days can be obtained.

기준일에 대응되는 구간 심박수들의 전처리 결과 획득(S140)Obtain preprocessing results of section heart rates corresponding to the reference date (S140)

상기 기준일에 대응되는 구간 심박수들의 전처리가 이미 수행되었고, 그 전처리 결과가 저장되어 있는 경우, 상기 저장된 전처리 결과가 획득될 수 있다.If preprocessing of the heart rates in the section corresponding to the reference date has already been performed and the preprocessing result is stored, the stored preprocessing result may be obtained.

만약, 상기 기준일에 대응되는 구간 심박수들의 전처리가 아직 수행되지 않은 경우, 상기 기준일에 대응되는 구간 심박수들의 전처리가 수행될 수 있다.If preprocessing of the zone heart rates corresponding to the reference date has not yet been performed, preprocessing of the zone heart rates corresponding to the reference date may be performed.

상기 기준일에 대응되는 구간 심박수들의 획득 및 그에 대한 전처리는 전술한 대상일에 대응되는 구간 심박수들의 획득 및 그에 대한 전처리와 유사하기 때문에 여기서는 자세한 설명을 생략한다.Since the acquisition of zone heart rates corresponding to the reference date and preprocessing thereof are similar to the acquisition of zone heart rates corresponding to the target date and preprocessing thereof, detailed description thereof will be omitted here.

결과적으로, 상기 기준일에 대한 구간 심박수들의 전처리 결과는 상기 기준일에 대응되는 구간 심박수들의 평균, 상기 기준일에 대응되는 구간 심박수들의 조화평균 및 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 중 적어도 하나를 포함할 수 있다. 이때, 상기 기준일에 대응되는 구간 심박수들의 분포에 관한 지표는 상기 대상일에 대응되는 구간 심박수들의 표준편차(standard deviation), 상기 대상일에 대응되는 구간 심박수들의 왜도(skewness), 상기 대상일에 대응되는 구간 심박수들의 첨도(kurtosis) 및 상기 대상일에 대응되는 구간 심박수들의 상대 표준편차(relative standard deviation) 중 적어도 하나를 포함할 수 있다.As a result, the preprocessing result of the section heart rates for the reference date is the average of the section heart rates corresponding to the reference date, the harmonic average of the section heart rates corresponding to the reference date, and the index of the distribution of section heart rates corresponding to the reference date. It can contain at least one. At this time, the indicators regarding the distribution of section heart rates corresponding to the reference date include the standard deviation of section heart rates corresponding to the target date, the skewness of section heart rates corresponding to the target date, and the target date. It may include at least one of kurtosis of corresponding section heart rates and relative standard deviation of section heart rates corresponding to the target date.

일 실시예에 따르면, 기준일에 대응되는 구간 심박수들이 전처리될 때에 구간 심박수에 포함된 일부의 심박수값이 노이즈로 제거될 수 있다. 예를 들어, 구간 평균 심박수는 노이즈가 제거된 구간 심박수를 이용해서 계산될 수 있다. 예를 들어, 구간 심박수들의 분포는 노이즈가 제거된 구간 심박수를 이용해서 계산될 수 있다.According to one embodiment, when the zone heart rates corresponding to the reference date are preprocessed, some heart rate values included in the zone heart rates may be removed as noise. For example, the section average heart rate can be calculated using the section heart rate from which noise has been removed. For example, the distribution of zone heart rates can be calculated using the zone heart rates from which noise has been removed.

노이즈를 제거하는 구체적인 형태는 이미 앞서 설명한 바 있어 생략한다.The specific form of removing noise has already been described previously, so it will be omitted.

만약, 복수의 기준일이 결정된 경우, 복수의 기준일 각각에 대응되는 구간 심박수들의 전처리 결과가 획득될 수 있다.If a plurality of reference days are determined, preprocessing results of section heart rates corresponding to each of the plurality of reference days may be obtained.

한편, 상기 기준일에 대응되는 구간 심박수들의 다양한 전처리 결과들 중 후술하는 예측모델의 학습에 사용된 입력값에 관련된 전처리 결과들만 획득될 수 있다. 예를 들어, 상기 예측모델의 학습에 대상일과 기준일에 대한 구간 심박수들의 평균값의 변화율 및 상대 표준편차의 변화량이 사용된 경우, 상기 기준일에 대응되는 구간 심박수들로부터 상기 기준일에 대응되는 구간 심박수들의 평균 및 표준편차만 획득될 수 있다. 다만, 예측모델의 학습에 사용된 입력값과 무관하게 전술한 모든 또는 일부의 전처리 결과들이 획득될 수도 있다.Meanwhile, among the various preprocessing results of the heart rates in the section corresponding to the reference date, only preprocessing results related to the input value used for learning the prediction model, which will be described later, can be obtained. For example, when the rate of change and the change in the relative standard deviation of the average value of section heart rates for the target date and the reference date are used in learning the prediction model, the average of the section heart rates corresponding to the reference date is calculated from the section heart rates corresponding to the reference date. and only the standard deviation can be obtained. However, all or some of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.

대상일의 구간 심박수들에 대한 전처리 결과와 기준일의 구간 심박수들에 대한 전처리 결과의 차이 산출(S150)Calculate the difference between the preprocessing results for the zone heart rates of the target day and the preprocessing results for the zone heart rates of the reference day (S150)

대상일의 구간 심박수들에 대한 전처리 결과와 기준일의 구간 심박수들에 대한 전처리 결과의 차이는 다음 중 하나 또는 이들의 조합일 수 있다.The difference between the preprocessing result for the zone heart rates of the target day and the preprocessing result for the zone heart rates of the reference day may be one of the following or a combination of these.

1) 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 (대상일의 구간 평균 심박수 - 기준일의 구간 평균 심박수)1) The amount of change in the average of section heart rates corresponding to the target day (segment average heart rate of the target day) to the average of section heart rates corresponding to the reference day (segment average heart rate of the target day) (segment average heart rate of the target day - section average of the reference day) heart rate)

2) 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율 ( (대상일의 구간 평균 심박수 - 기준일의 구간 평균 심박수) / 기준일의 구간 평균 심박수 )2) The rate of change of the average of the section heart rates corresponding to the target day (average heart rate of the section of the target day) to the average of the section heart rates corresponding to the reference day (average heart rate of the section of the target day) ((average heart rate of the section of the target day - section of the reference day) Average heart rate) / Average heart rate of the section on the reference day)

3) 대상일에 대응되는 구간 심박수들의 조화평균(대상일의 구간 조화평균 심박수)의 기준일에 대응되는 구간 심박수들의 조화평균(기준일의 구간 조화평균 심박수)에 대한 변화량 (대상일의 구간 조화평균 심박수 - 기준일의 구간 조화평균 심박수)3) The amount of change in the harmonic average of section heart rates corresponding to the target day (section harmonic average heart rate of the target day) to the harmonic average of section heart rates corresponding to the reference day (section harmonic average heart rate of the target day) - Harmonic average heart rate of the section on the reference day)

4) 대상일에 대응되는 구간 심박수들의 조화평균(대상일의 구간 조화평균 심박수)의 기준일에 대응되는 구간 심박수들의 조화평균(기준일의 구간 조화평균 심박수)에 대한 변화율 ( (대상일의 구간 조화평균 심박수 - 기준일의 구간 조화평균 심박수) / 기준일의 구간 조화평균 심박수 )4) The rate of change of the harmonic average of the section heart rates corresponding to the target day (section harmonic average heart rate of the target day) to the harmonic average of the section heart rates corresponding to the reference day (section harmonic average heart rate of the target day) ((section harmonic average of the target day) Heart rate - Harmonic average heart rate of the section on the reference day) / Harmonic average heart rate of the section on the reference day)

5) 대상일에 대응되는 구간 심박수들의 표준편차의 기준일에 대응되는 구간 심박수들의 표준편차에 대한 변화량 (대상일에 대응되는 구간 심박수들의 표준편차 - 기준일에 대응되는 구간 심박수들의 표준편차)5) The amount of change in the standard deviation of section heart rates corresponding to the target date of the standard deviation of section heart rates corresponding to the reference date (standard deviation of section heart rates corresponding to the target date - standard deviation of section heart rates corresponding to the reference date)

6) 대상일에 대응되는 구간 심박수들의 상대 표준편차의 기준일에 대응되는 구간 심박수들의 상대 표준편차에 대한 변화량 ( (대상일에 대응되는 구간 심박수들의 표준편차 / 대상일의 구간 평균 심박수) - (기준일에 대응되는 구간 심박수들의 표준편차 / 기준일의 구간 평균 심박수) )6) The amount of change in the relative standard deviation of section heart rates corresponding to the reference date of the relative standard deviation of section heart rates corresponding to the target date ((standard deviation of section heart rates corresponding to the target day / average heart rate of the section on the target day) - (base date Standard deviation of the corresponding section heart rates / Average section heart rate on the reference day) )

7) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)7) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

8) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)8) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

9) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence9) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

만약, 복수의 기준일이 결정된 경우, 복수의 기준일 각각에 대한 전처리 결과의 차이들이 획득될 수 있다. 예를 들어, 제1 기준일과 대상일 사이에서 결정되는 제1 전처리 결과의 차이와 제2 기준일과 대상일 사이에서 결정되는 제2 전처리 결과의 차이가 획득될 수 있다.If multiple reference dates are determined, differences in preprocessing results for each of the multiple reference dates may be obtained. For example, the difference between the first preprocessing result determined between the first reference date and the target date and the difference between the second preprocessing result determined between the second reference date and the target date may be obtained.

한편, 상기 전처리 결과의 차이들 중 후술하는 예측모델의 학습에 사용된 입력값에 관련된 전처리 결과의 차이들만 획득될 수 있다. 예를 들어, 상기 예측모델의 학습에 대상일과 기준일에 대한 구간 심박수들의 평균값의 변화율 및 상대 표준편차의 변화량이 사용된 경우, 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율 및 대상일에 대응되는 구간 심박수들의 표준편차의 기준일에 대응되는 구간 심박수들의 표준편차에 대한 변화량만 획득될 수 있다. 다만, 예측모델의 학습에 사용된 입력값과 무관하게 전술한 모든 또는 일부의 전처리 결과의 차이들이 획득될 수도 있다.Meanwhile, among the differences in the preprocessing results, only the differences in the preprocessing results related to the input values used for learning the prediction model described later can be obtained. For example, when the rate of change and the change in the relative standard deviation of the average value of section heart rates for the target date and reference day are used in learning the prediction model, the average of section heart rates corresponding to the target date (average heart rate of the section on the target day) Only the rate of change with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of the reference date) and the change in the standard deviation of section heart rates corresponding to the reference date of the standard deviation of section heart rates corresponding to the target date can be obtained. However, differences in all or part of the above-mentioned preprocessing results may be obtained regardless of the input values used for learning the prediction model.

조합값의 산출Calculation of combination values

본 출원에 의해 개시되는 몇몇 실시예들에 따르면, 변수들의 조합값이 산출될 수 있다.According to some embodiments disclosed by the present application, combination values of variables may be calculated.

제1 기준일에 제1 구간 평균 심박수, 제1 표준편차, 제1 왜도, 제1 첨도, 제1 TSH 호르몬 농도, 제1 free T4 농도, 제1 T4 농도, 제1 free T3 농도, 제1 T3 농도, 제1 TRH 농도 및 제1 기준일에 대한 갑상선기능이상증에 대한 진단결과가 대응되어 있고, 제2 기준일에 제2 구간 평균 심박수, 제2 표준편차, 제2 왜도, 제2 첨도, 제2 TSH 호르몬 농도, 제2 free T4 농도, 제2 T4 농도, 제2 free T3 농도, 제2 T3 농도, 제2 TRH 농도 및 제2 기준일에 대한 갑상선기능이상증에 대한 진단결과가 대응되어 있을 때, 상기 변수들은 다음과 같이 산출된다.On the first reference day, the first section average heart rate, first standard deviation, first skewness, first kurtosis, first TSH hormone concentration, first free T4 concentration, first T4 concentration, first free T3 concentration, first T3 Concentration, first TRH concentration, and diagnosis results for dysthyroidism for the first reference date are corresponded, and on the second reference date, the second section average heart rate, second standard deviation, second skewness, second kurtosis, and second When the diagnosis results for dysthyroidism for the TSH hormone concentration, second free T4 concentration, second T4 concentration, second free T3 concentration, second T3 concentration, second TRH concentration and second reference date correspond to the above The variables are calculated as follows.

(1) 구간 평균 심박수 변화량 = 제2 기준일의 구간 평균 심박수 - 제1 기준일의 구간 평균 심박수,(1) Change in section average heart rate = section average heart rate on the second reference day - section average heart rate on the first reference day,

(2) 구간 평균 심박수 변화율 = (제2 기준일의 구간 평균 심박수 - 제1 기준일의 구간 평균 심박수) / 제2 기준일의 구간 평균 심박수,(2) Zone average heart rate change rate = (section average heart rate on the second reference day - section average heart rate on the first reference day) / section average heart rate on the second reference day,

(3) 구간 조화평균 심박수 변화량 = 제2 기준일의 구간 조화평균 심박수 - 제1 기준일의 구간 조화평균 심박수,(3) Change in section harmonic average heart rate = section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day,

(4) 구간 조화평균 심박수 변화율 = (제2 기준일의 구간 조화평균 심박수 - 제1 기준일의 구간 조화평균 심박수) / 제2 기준일의 구간 조화평균 심박수,(4) Section harmonic average heart rate change = (section harmonic average heart rate on the second reference day - section harmonic average heart rate on the first reference day) / section harmonic average heart rate on the second reference day,

(5) 표준편차 변화량 = 제2 기준일의 구간 심박수들의 표준편차 - 제1 기준일의 구간심박수들의 표준편차,(5) Standard deviation change = standard deviation of zone heart rates on the second reference day - standard deviation of zone heart rates on the first reference day,

(6) 상대 표준편차 변화량 = (제2 기준일의 구간 심박수들의 표준편차 / 제2 기준일의 구간 평균 심박수) - (제1 기준일의 구간심박수들의 표준편차 / 제1 기준일의 구간 평균 심박수),(6) Relative standard deviation change = (standard deviation of section heart rates on the second reference day / section average heart rate on the second reference day) - (standard deviation of section heart rates on the first reference day / section average heart rate on the first reference day),

(7) 구간 심박수의 왜도 변화량 = 제2 기준일의 구간 심박수들의 왜도 - 제1 기준일의 구간 심박수들의 왜도,(7) Skewness change in section heart rates = skewness of section heart rates on the second reference day - skewness of section heart rates on the first reference day,

(8) 구간 심박수의 첨도 변화량 = 제2 기준일의 구간 심박수들의 첨도 - 제1 기준일의 구간 심박수들의 왜도,(8) Change in kurtosis of section heart rates = kurtosis of section heart rates on the second reference day - skewness of section heart rates on the first reference day,

(9) JS Divergence : 제2 기준일의 구간 심박수들과 제1 기준일의 구간 심박수들 사이에서 계산되는 JS Divergence,(9) JS Divergence: JS Divergence calculated between the zone heart rates of the second reference day and the zone heart rates of the first reference day,

(10) 제1 TSH 호르몬 농도,(10) first TSH hormone concentration,

(11) 제1 free T4 농도,(11) first free T4 concentration,

(12) 제1 T4 농도, (12) first T4 concentration,

(13) 제1 free T3 농도, (13) first free T3 concentration,

(14) 제1 T3 농도, (14) first T3 concentration,

(15) 제1 TRH 농도,(15) first TRH concentration,

(16) 제2 TSH 호르몬 농도, (16) secondary TSH hormone concentration,

(17) 제2 free T4 농도, (17) second free T4 concentration,

(18) 제2 T4 농도, (18) second T4 concentration,

(19) 제2 free T3 농도, (19) second free T3 concentration,

(20) 제2 T3 농도, 및 (20) a second T3 concentration, and

(21) 제2 TRH 농도(21) Second TRH concentration

이때, 상기 변수들간의 조합값은 상기 변수들 중에서 선택되는 하나의 값(이하, 제1 값이라 함)과 상기 상기 변수들 중에서 선택되는 다른 하나의 값(이하, 제2 값이라 함)을 서로 곱한 값, 제1 값을 제2 값으로 나눈 값 및 제2 값을 제1 값으로 나눈 값 중 하나를 의미한다. 예를 들어, 상기 조합값은 제2 기준일의 free T4의 농도값을 JS Divergence로 나눈값일 수 있다.At this time, the combination value between the variables is one value selected from among the variables (hereinafter referred to as the first value) and another value selected from the variables (hereinafter referred to as the second value). It means one of a multiplied value, a value obtained by dividing a first value by a second value, and a value obtained by dividing a second value by a first value. For example, the combination value may be the concentration value of free T4 on the second reference date divided by JS Divergence.

대상일와 기준일의 차이(day gap)를 산출Calculate the difference (day gap) between the target date and the reference date

본 출원에 의해 개시되는 몇몇 실시예들에 따르면, 상기 대상일과 상기 기준일 사이의 차이(day gap)가 획득될 수 있다.According to some embodiments disclosed by the present application, a day gap between the target date and the reference date may be obtained.

사용자의 개인 정보 획득Obtaining your personal information

본 출원에 의해 개시되는 몇몇 실시예들에 따르면, 대상(subject)의 개인 정보, 즉 사용자의 개인 정보가 획득될 수 있다.According to some embodiments disclosed by the present application, the subject's personal information, that is, the user's personal information, may be obtained.

추정 호르몬 수치의 획득Obtaining estimated hormone levels

본 출원에 의해 개시되는 몇몇 실시예에 따르면, 대상일에 대응되는 갑상선 호르몬 농도에 관한 값이 획득될 수 있다. 일 예로, 갑상선 호르몬 예측모델의 출력값 또는 갑상선 호르몬 예측모델의 출력값을 기초로 획득된 갑상선 호르몬 농도에 관한 값이 획득될 수 있다. According to some embodiments disclosed by the present application, values regarding thyroid hormone concentration corresponding to the target date can be obtained. As an example, a value related to the thyroid hormone concentration obtained based on the output value of the thyroid hormone prediction model or the output value of the thyroid hormone prediction model may be obtained.

갑상선기능이상 예측모델을 통한 갑상선기능이상증에 대한 예측결과 획득(S160)Obtaining prediction results for thyroid dysfunction through thyroid dysfunction prediction model (S160)

갑상선기능항진증 예측모델을 통한 갑상선기능항진증에 대한 예측결과 획득Obtaining prediction results for hyperthyroidism through hyperthyroidism prediction model

학습방법 실시예#1에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량이 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.Learning method As described in Example #1, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The amount of change in the average of the heart rates in each section (average heart rate in the section on the reference day) can be entered, and accordingly, a result can be obtained as to whether the subject has hyperthyroidism or is not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

학습방법 실시예#2에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 상기 대상일에 대응되는 구간 평균 심박수가 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.Learning method As described in Example #2, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day). The amount of change in the average of the heart rates in each section (average heart rate in the section on the reference day) and the average heart rate in the section corresponding to the target date can be entered, and a result is obtained as to whether the subject has hyperthyroidism or not hyperthyroidism. It can be. In other words, a prediction result regarding hyperthyroidism can be obtained.

학습방법 실시예#3에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이가 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. 이때, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이는 다음에서 선택되는 하나 또는 이들의 조합일 수 있다.Learning method As described in Example #3, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The amount of change in the average of section heart rates (average section heart rate of the reference day) and the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date The difference can be entered, and a result can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained. At this time, the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.

1) 대상일에 대응되는 구간 심박수들의 표준편차의 기준일에 대응되는 구간 심박수들의 표준편차에 대한 변화량 (대상일에 대응되는 구간 심박수들의 표준편차 - 기준일에 대응되는 구간 심박수들의 표준편차)1) The amount of change in the standard deviation of section heart rates corresponding to the target date of the standard deviation of section heart rates corresponding to the target date (standard deviation of section heart rates corresponding to the target date - standard deviation of section heart rates corresponding to the reference date)

2) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)2) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

3) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)3) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

4) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence4) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

학습방법 실시예#4에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 심박수들의 평균, 및 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이가 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.Learning method As described in Example #4, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The amount of change in the average of the section heart rates (the average heart rate of the section on the reference day), the average of the section heart rates corresponding to the target date, and the index of the distribution of section heart rates corresponding to the target date and the index corresponding to the reference date The difference between the indices regarding the distribution of zone heart rates can be input, and accordingly a result can be obtained as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

학습방법 실시예#5에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율이 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.Learning method As described in Example #5, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The rate of change for the average of the heart rates in the zone (average heart rate in the zone on the reference day) can be input, and accordingly, a result as to whether the subject has hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

학습방법 실시예#6에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율 및 상기 대상일에 대응되는 구간 심박수들의 평균이 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. Learning method As described in Example #6, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day). The rate of change for the average of the heart rates in the zone (average heart rate in the zone on the reference day) and the average of the heart rates in the zone corresponding to the target date can be input, and accordingly, the result of whether the subject is hyperthyroid or not hyperthyroid is given. can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

학습방법 실시예#7에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 및 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이가 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. 이때, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이는 다음에서 선택되는 하나 또는 이들의 조합일 수 있다.Learning method As described in Example #7, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (segment average heart rate of the target day). The rate of change in the average of the zone heart rates (average heart rate of the zone on the reference day), and the difference between the index of the distribution of zone heart rates corresponding to the target day and the index of the distribution of zone heart rates corresponding to the reference day. can be entered, and accordingly, a result can be obtained as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained. At this time, the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.

1) 대상일에 대응되는 구간 심박수들의 상대 표준편차의 기준일에 대응되는 구간 심박수들의 상대 표준편차에 대한 변화량 ( (대상일에 대응되는 구간 심박수들의 표준편차 / 대상일의 구간 평균 심박수) - (기준일에 대응되는 구간 심박수들의 표준편차 / 기준일의 구간 평균 심박수) )1) The amount of change in the relative standard deviation of section heart rates corresponding to the target date of the relative standard deviation of section heart rates corresponding to the target date ((standard deviation of section heart rates corresponding to the target day / average heart rate of the section on the target day) - (base date Standard deviation of the corresponding section heart rates / Average section heart rate on the reference day) )

2) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)2) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

3) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)3) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

4) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence4) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

학습방법 실시예#8에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 상기 대상일에 대응되는 구간 심박수들의 평균 및 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이가 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. 이때, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이는 다음에서 선택되는 하나 또는 이들의 조합일 수 있다.Learning method As described in Example #8, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The rate of change for the average of the section heart rates (the average heart rate of the section on the reference day), the average of the section heart rates corresponding to the target date and the distribution of the section heart rates corresponding to the target day, and the index of the section heart rates corresponding to the reference date Differences between indicators regarding distribution can be entered, and results can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained. At this time, the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.

1) 대상일에 대응되는 구간 심박수들의 상대 표준편차의 기준일에 대응되는 구간 심박수들의 상대 표준편차에 대한 변화량 ( (대상일에 대응되는 구간 심박수들의 표준편차 / 대상일의 구간 평균 심박수) - (기준일에 대응되는 구간 심박수들의 표준편차 / 기준일의 구간 평균 심박수) )1) The amount of change in the relative standard deviation of section heart rates corresponding to the target date of the relative standard deviation of section heart rates corresponding to the target date ((standard deviation of section heart rates corresponding to the target day / average heart rate of the section on the target day) - (base date Standard deviation of the corresponding section heart rates / Average section heart rate on the reference day) )

2) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)2) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

3) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)3) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

4) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence4) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

학습방법 실시예#9에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상일과 상기 기준일 사이의 차이(day gap)이 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. 이때, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이는 다음에서 선택되는 하나 또는 이들의 조합일 수 있다.Learning method As described in Example #9, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day). The rate of change (or amount of change) of the average of the zone heart rates (average heart rate of the zone on the base day), an indicator of the distribution of the zone heart rates corresponding to the target date, and an indicator of the distribution of the zone heart rates corresponding to the base day. The difference between and the difference (day gap) between the target date and the reference date can be entered, and accordingly, a result as to whether the target is hyperthyroid or not hyperthyroid can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained. At this time, the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.

1) 대상일에 대응되는 구간 심박수들의 표준편차의 기준일에 대응되는 구간 심박수들의 표준편차에 대한 변화량 (대상일에 대응되는 구간 심박수들의 표준편차 - 기준일에 대응되는 구간 심박수들의 표준편차)1) The amount of change in the standard deviation of section heart rates corresponding to the target date of the standard deviation of section heart rates corresponding to the target date (standard deviation of section heart rates corresponding to the target date - standard deviation of section heart rates corresponding to the reference date)

2) 대상일에 대응되는 구간 심박수들의 상대 표준편차의 기준일에 대응되는 구간 심박수들의 상대 표준편차에 대한 변화량 ( (대상일에 대응되는 구간 심박수들의 표준편차 / 대상일의 구간 평균 심박수) - (기준일에 대응되는 구간 심박수들의 표준편차 / 기준일의 구간 평균 심박수) )2) The amount of change in the relative standard deviation of section heart rates corresponding to the reference date of the relative standard deviation of section heart rates corresponding to the target date ((standard deviation of section heart rates corresponding to the target day / average heart rate of the section on the target day) - (base date Standard deviation of the corresponding section heart rates / Average section heart rate on the reference day) )

3) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)3) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

4) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)4) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

5) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence5) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

학습방법 실시예#10에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 변수들 간의 조합값이 입력될 수 있으며, 그에 따라 대상이 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다. Learning method As described in Example #10, the learned hyperthyroidism prediction model corresponds to the reference date of the hormone concentration of the obtained reference day and the average of the section heart rates corresponding to the obtained target date (average heart rate of the section of the target day) The rate of change (or amount of change) of the average of the zone heart rates (average heart rate of the zone on the base day), an indicator of the distribution of the zone heart rates corresponding to the target date, and an indicator of the distribution of the zone heart rates corresponding to the base day. Differences between variables and combination values between variables can be input, and results can be obtained accordingly as to whether the subject is hyperthyroid or not hyperthyroid. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

이때, 상기 대상일에 대응되는 구간 심박수들의 분포에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이는 다음에서 선택되는 하나 또는 이들의 조합일 수 있다.At this time, the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date may be one selected from the following or a combination thereof.

1) 대상일에 대응되는 구간 심박수들의 표준편차의 기준일에 대응되는 구간 심박수들의 표준편차에 대한 변화량 (대상일에 대응되는 구간 심박수들의 표준편차 - 기준일에 대응되는 구간 심박수들의 표준편차)1) The amount of change in the standard deviation of section heart rates corresponding to the target date of the standard deviation of section heart rates corresponding to the target date (standard deviation of section heart rates corresponding to the target date - standard deviation of section heart rates corresponding to the reference date)

2) 대상일에 대응되는 구간 심박수들의 상대 표준편차의 기준일에 대응되는 구간 심박수들의 상대 표준편차에 대한 변화량 ( (대상일에 대응되는 구간 심박수들의 표준편차 / 대상일의 구간 평균 심박수) - (기준일에 대응되는 구간 심박수들의 표준편차 / 기준일의 구간 평균 심박수) )2) The amount of change in the relative standard deviation of section heart rates corresponding to the reference date of the relative standard deviation of section heart rates corresponding to the target date ((standard deviation of section heart rates corresponding to the target day / average heart rate of the section on the target day) - (base date Standard deviation of the corresponding section heart rates / Average section heart rate on the reference day) )

3) 대상일에 대응되는 구간 심박수들의 왜도의 기준일에 대응되는 구간 심박수들의 왜도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 왜도 - 기준일에 대응되는 구간 심박수들의 왜도)3) The amount of change in the skewness of section heart rates corresponding to the reference date of the skewness of section heart rates corresponding to the target date (skewness of section heart rates corresponding to the target date - skewness of section heart rates corresponding to the reference date)

4) 대상일에 대응되는 구간 심박수들의 첨도의 기준일에 대응되는 구간 심박수들의 첨도에 대한 변화량 (대상일에 대응되는 구간 심박수들의 첨도 - 기준일에 대응되는 구간 심박수들의 첨도)4) The amount of change in the kurtosis of section heart rates corresponding to the reference date of the kurtosis of section heart rates corresponding to the target date (kurtosis of section heart rates corresponding to the target date - kurtosis of section heart rates corresponding to the reference date)

5) 대상일에 대응되는 구간 심박수들의 분포(distribution)과 기준일에 대응되는 구간 심박수들의 분포(distribution) 사이의 JS divergence5) JS divergence between the distribution of zone heart rates corresponding to the target day and the distribution of zone heart rates corresponding to the reference day

한편, 변수들 간의 조합값은 학습방법 실시예#10에서 자세히 설명한 바 있으므로, 여기서는 자세한 설명을 생략한다.Meanwhile, since the combination values between variables have been explained in detail in Learning Method Example #10, detailed explanation is omitted here.

학습방법 실시예#11에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 다음과 같은 값들이 입력될 수 있으며, 그에 따라 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.As described in Learning Method Example #11, the following values can be input into the learned hyperthyroidism prediction model, and accordingly, a result of whether hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

(1) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 상기 대상의 나이 및/또는 성별(1) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) and the age and/or gender of said subject.

(2) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 평균 심박수 및 상기 대상의 나이 및/또는 성별(2) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the average heart rate of the section corresponding to the target date, and the age and/or gender of the target

(3) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상의 나이 및/또는 성별(3) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date, and the age and/or gender of the subject.

(4) 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 심박수들의 평균, 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상의 나이 및/또는 성별(4) The amount of change in the hormone concentration of the obtained reference day and the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the average of the section heart rates corresponding to the target date, the difference between the index regarding the distribution of section heart rates corresponding to the target day and the index regarding the distribution of section heart rates corresponding to the reference date, and the target age and/or gender

(5) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율 및 상기 대상의 나이 및/또는 성별(5) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) and the age and/or gender of said subject.

(6) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 상기 대상일에 대응되는 구간 심박수들의 평균 및 상기 대상의 나이 및/또는 성별(6) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference day (segment average heart rate of reference day) , the average of the interval heart rates corresponding to the target date and the age and/or gender of the target

(7) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상의 나이 및/또는 성별(7) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date, and the age and/or gender of the subject.

(8) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 상기 대상일에 대응되는 구간 심박수들의 평균, 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상의 나이 및/또는 성별(8) The rate of change of the hormone concentration of the obtained reference date, the average of section heart rates corresponding to the obtained target date (average section heart rate of the target day) with respect to the average of section heart rates corresponding to the reference day (average section heart rate of the reference day) , the average of the section heart rates corresponding to the target date, the difference between the index of the distribution of section heart rates corresponding to the target day and the index of the distribution of section heart rates corresponding to the reference date, and the age of the target. and/or gender

(9) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이, 상기 대상일과 상기 기준일 사이의 차이(day gap) 및 상기 대상의 나이 및/또는 성별(9) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) (or amount of change), the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference day, the difference between the target date and the reference date (day gap) ) and the age and/or gender of said subject.

(9) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이, 변수들 간의 조합값 및 상기 대상의 나이 및/또는 성별(9) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) (or amount of change), the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference date, the combination value between variables and the age of the subject, and /or gender

학습방법 실시예#12에서 설명한 바와 같이 학습된 갑상선기능항진증 예측모델에 다음과 같은 값들이 입력될 수 있으며, 그에 따라 갑상선기능항진증 인지 혹은 갑상선기능항진증이 아닌지에 대한 결과가 획득될 수 있다. 즉, 갑상선기능항진 여부에 대한 예측결과가 획득될 수 있다.As described in Learning Method Example #12, the following values can be input into the learned hyperthyroidism prediction model, and accordingly, a result of whether hyperthyroidism or not hyperthyroidism can be obtained. In other words, a prediction result regarding whether hyperthyroidism is present can be obtained.

(1) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 상기 대상일의 추정 호르몬 농도에 관한 값(1) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) and values related to the estimated hormone concentration on the target date.

(2) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 평균 심박수 및 상기 대상일의 추정 호르몬 농도에 관한 값(2) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the average heart rate of the section corresponding to the target date and the value related to the estimated hormone concentration on the target date

(3) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상일의 추정 호르몬 농도에 관한 값(3) The amount of change in the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the difference between an index regarding the distribution of section heart rates corresponding to the target date and an index related to the distribution of section heart rates corresponding to the reference date, and a value related to the estimated hormone concentration of the target date

(4) 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량, 상기 대상일에 대응되는 구간 심박수들의 평균, 상기 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 상기 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상일의 추정 호르몬 농도에 관한 값(4) The amount of change in the hormone concentration of the obtained reference day and the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the average of the section heart rates corresponding to the target date, the difference between the index regarding the distribution of section heart rates corresponding to the target day and the index regarding the distribution of section heart rates corresponding to the reference date, and the target Values for estimated hormone concentrations per day

(5) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율 및 상기 대상일의 추정 호르몬 농도에 관한 값(5) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) and values related to the estimated hormone concentration on the target date.

(6) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 상기 대상일에 대응되는 구간 심박수들의 평균 및 상기 대상일의 추정 호르몬 농도에 관한 값(6) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference day (segment average heart rate of reference day) , the average of section heart rates corresponding to the target date and the value related to the estimated hormone concentration on the target date

(7) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상일의 추정 호르몬 농도에 관한 값(7) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) , the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference day, and the value related to the estimated hormone concentration on the target date.

(8) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율, 상기 대상일에 대응되는 구간 심박수들의 평균, 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이 및 상기 대상일의 추정 호르몬 농도에 관한 값(8) The rate of change of the hormone concentration of the obtained reference date, the average of section heart rates corresponding to the obtained target date (average section heart rate of the target day) with respect to the average of section heart rates corresponding to the reference day (average section heart rate of the reference day) , the average of the section heart rates corresponding to the target date, the difference between the index regarding the distribution of section heart rates corresponding to the target day and the index regarding the distribution of section heart rates corresponding to the reference day, and the target date Values for estimated hormone concentrations

(9) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이, 상기 대상일과 상기 기준일 사이의 차이(day gap) 및 상기 대상일의 추정 호르몬 농도에 관한 값(9) The rate of change of the hormone concentration of the obtained reference day, the average of section heart rates corresponding to the obtained target date (segment average heart rate of target day) with respect to the average of section heart rates corresponding to the reference date (segment average heart rate of reference day) (or amount of change), the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference day, the difference between the target date and the reference date (day gap) ) and values related to the estimated hormone concentration on the target date

(10) 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화율(또는 변화량), 대상일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 기준일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표 사이의 차이, 변수들 간의 조합값 및 상기 대상일의 추정 호르몬 농도에 관한 값(10) The rate of change of the hormone concentration of the obtained reference day, the average of the section heart rates corresponding to the obtained target date (segment average heart rate of the target day) with respect to the average of section heart rates corresponding to the reference day (segment average heart rate of the reference day) (or amount of change), the difference between the index regarding the distribution of section heart rates corresponding to the target date and the index regarding the distribution of section heart rates corresponding to the reference day, the combination value between variables, and the estimate of the target date Values related to hormone concentration

갑상선기능저하증 예측모델을 통한 갑상선기능저하증에 대한 예측결과 획득Obtaining prediction results for hypothyroidism through hypothyroidism prediction model

갑상선기능항진증 예측모델을 통한 갑상선기능항진증에 대한 예측결과 획득에 설명한 바와 유사하게, 갑상선기능저하증 예측모델을 통해 갑상선기능저항증에 대한 예측결과를 획득할 수 있다. 이에 대한 자세한 설명은 생략한다.Similar to what was described in obtaining prediction results for hyperthyroidism through a hyperthyroidism prediction model, prediction results for hypothyroidism can be obtained through a hypothyroidism prediction model. A detailed description of this will be omitted.

갑상선기능이상증에 대한 예측결과의 획득Obtaining prediction results for thyroid dysfunction

상기 갑상선기능항진증에 대한 예측결과와 상기 갑상성기능저하증에 대한 예측결과를 종합하여 대상의 갑상선기능이상증에 대한 예측 결과가 획득된다.By combining the prediction results for hyperthyroidism and the prediction results for hypothyroidism, a prediction result for the subject's dysthyroidism is obtained.

갑상선기능항진증에 대한 예측결과(A)와 갑상선기능저하증에 대한 예측결과(B)를 고려한 갑상선기능이상증에 대한 예측 결과는 다음의 [표2]와 같이 획득될 수 있다.The prediction results for dysthyroidism considering the prediction results for hyperthyroidism (A) and the prediction results for hypothyroidism (B) can be obtained as shown in [Table 2] below.

AA BB 갑상선기능이상증 예측 결과Dysthyroidism prediction results 갑상선기능항진증hyperthyroidism 갑상성기능저하증hypothyroidism 갑상선기능항진증 또는 갑상선기능저하증Hyperthyroidism or hypothyroidism 갑상선기능항진증아님Not hyperthyroidism 갑상성기능저하증아님Not hypothyroidism 정상normal 갑상선기능항진증hyperthyroidism 갑상성기능저하증아님Not hypothyroidism 갑상선기능항진증hyperthyroidism 갑상선기능항진증아님Not hyperthyroidism 갑상성기능저하증hypothyroidism 갑상성기능저하증hypothyroidism

한편, 갑상선기능항진증에 대한 예측결과가 갑상선기능항진증으로 나오고, 갑상선기능저하증에 대한 예측결과도 갑상선기능저하증으로 나온 경우, 갑상선기능항진증 예측모델의 출력값(확률값)과 갑상선기능저하증 예측모델의 출력값(확률값)을 비교하여, 보다 더 높은 확률값을 출력한 결과가 최종적인 갑상선기능이상증에 대한 결과로 채택될 수 있다.전술한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 모두 상기 심박수 측정 디바이스(10)에 의해 수행될 수 있다.On the other hand, if the prediction result for hyperthyroidism comes out as hyperthyroidism and the prediction result for hypothyroidism also comes out as hypothyroidism, the output value (probability value) of the hyperthyroidism prediction model and the output value (probability value) of the hypothyroidism prediction model ( probability value), the result outputting a higher probability value can be adopted as the final result for thyroid dysfunction. Obtaining the above-mentioned trigger signal (S100), on the target date determined by the trigger signal Acquire corresponding section heart rates (S110), preprocess the section heart rates corresponding to the target date (S120), obtain hormone concentrations corresponding to the reference date (S130), and obtain the section heart rates corresponding to the reference date (S130). Obtaining preprocessing results (S140), obtaining the difference between the preprocessing results of the zone heart rates corresponding to the target date and the preprocessing results of the zone heart rates corresponding to the reference date (S150), the obtained difference and the reference date Processing the corresponding hormone concentration using a thyroid dysfunction prediction model and obtaining results on thyroid dysfunction for the subject (S160) can all be performed by the heart rate measurement device 10.

전술한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 모두 상기 사용자 단말기(20)에 의해 수행될 수 있다.Obtaining the above-described trigger signal (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), preprocessing the section heart rates corresponding to the target date (S120), on the reference date Obtaining the corresponding hormone concentration (S130), obtaining preprocessing results for section heart rates corresponding to the reference date (S140), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Obtaining the difference in pre-processing results (S150), processing the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, and obtaining the results for thyroid dysfunction for the subject (S150). S160) can all be performed by the user terminal 20.

전술한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 모두 상기 서버(30)에 의해 수행될 수 있다.Obtaining the above-described trigger signal (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), preprocessing the section heart rates corresponding to the target date (S120), on the reference date Obtaining the corresponding hormone concentration (S130), obtaining preprocessing results for section heart rates corresponding to the reference date (S140), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Obtaining the difference in pre-processing results (S150), processing the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, and obtaining the results for thyroid dysfunction for the subject (S150). S160) can all be performed by the server 30.

전술한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 상기 심박수 측정 디바이스(10), 상기 사용자 단말기(20) 및 상기 서버(30)에 의해 적절히 분산되어 수행될 수 있다.Obtaining the above-described trigger signal (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), preprocessing the section heart rates corresponding to the target date (S120), on the reference date Obtaining the corresponding hormone concentration (S130), obtaining preprocessing results for section heart rates corresponding to the reference date (S140), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Obtaining the difference in pre-processing results (S150), processing the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, and obtaining the results for thyroid dysfunction for the subject (S150). S160) may be appropriately distributed and performed by the heart rate measurement device 10, the user terminal 20, and the server 30.

예를 들어, 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120)은 상기 사용자 단말기(20)에 의해 수행되고, 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 상기 서버(30)에 의해 수행될 수 있다.For example, obtaining a trigger signal (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), and preprocessing the section heart rates corresponding to the target date (S120). Performed by the user terminal 20, the hormone concentration corresponding to the reference date is obtained (S130), the preprocessing results for the heart rates of the section corresponding to the reference date are obtained (S140), the section corresponding to the target date Obtain the difference between the preprocessing result of the heart rates and the preprocessing result of the heart rates in the section corresponding to the reference date (S150), process the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, and Obtaining results on thyroid dysfunction for the subject (S160) may be performed by the server 30.

다른 예를 들어, 트리거 신호를 획득함(S100)은 상기 사용자 단말기(20)에 의해 수행되고, 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)은 상기 서버(30)에 의해 수행될 수 있다.For another example, acquiring a trigger signal (S100) is performed by the user terminal 20, and obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), on the target date Preprocess the corresponding section heart rates (S120), obtain the hormone concentration corresponding to the reference date (S130), obtain preprocessing results for the section heart rates corresponding to the reference date (S140), and obtain the preprocessing results for the section heart rates corresponding to the target date (S140). Obtaining the difference between the preprocessing results of the section heart rates and the preprocessing results of the section heart rates corresponding to the reference date (S150), processing the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model, Obtaining results on thyroid dysfunction for the subject (S160) may be performed by the server 30.

다만, 전술한 분산 수행의 형태에 국한되지 않고, 더 다양한 형태로 분산 수행될 수 있다.However, it is not limited to the form of distributed execution described above, and distributed execution may be performed in more diverse forms.

5. 실험예5. Experimental example

이하에서는, 본 출원에 의해 개시되는 갑상선기능이상증 예측모델을 학습시킨 실험예들과 그에 대한 정확도 등을 분석한 결과를 설명한다.Below, experimental examples of learning the thyroid dysfunction prediction model disclosed by this application and the results of analyzing the accuracy thereof will be described.

이하에서 설명하는 실험예들 및 비교예들에 사용된 임상 데이터는 전술한 방법에 의해 수집을 하였으며, 총 297명의 환자를 대상으로 수집한 총 1,027개의 임상 데이터 세트들이 활용되었고, 이 중 호르몬 농도에 대한 진단 결과 " 갑상선기능항진증"으로 분류된 임상 데이터 세트들은 총 168개, "정상"으로 분류된 임상 데이터 세트들은 총 801개, "갑상선기능저하증"으로 분류된 임상 데이터 세트들은 총 58개였다.The clinical data used in the experimental examples and comparative examples described below were collected by the method described above, and a total of 1,027 clinical data sets collected from a total of 297 patients were used, of which hormone concentration As a result of the diagnosis, there were a total of 168 clinical data sets classified as “hyperthyroidism,” a total of 801 clinical data sets classified as “normal,” and a total of 58 clinical data sets classified as “hypothyroidism.”

(1) 비교예#1 - 정상 기준일에 대한 심박수 차이 - 라벨링(1) Comparative Example #1 - Heart rate difference compared to normal reference day - Labeling

학습 데이터 세트들의 준비Preparation of training data sets

본 출원에 의해 개시되는 다양한 실시예들의 실험 결과와 비교하기 위한 비교예에 따른 예측모델을 학습시키기 위하여, 다음과 같은 방식으로 학습 데이터 세트가 준비되었다.In order to learn a prediction model according to a comparative example for comparison with the experimental results of various embodiments disclosed by this application, a learning data set was prepared in the following manner.

환자들 각각에 대하여, 진단결과가 정상이 아닌 검사일에 대응되는 구간 평균 심박수에서 진단결과가 정상인 검사일에 대응되는 구간 평균 심박수를 뺀 값이 입력값(예를 들어, 구간 평균 심박수의 변화량)으로 사용되었고, 상기 진단결과가 정상이 아닌 검사일에 대응되어 있는 진단 결과(항진 또는 저하)가 라벨링값으로 사용되었다. 그리고, 진단결과가 정상인 검사일이 둘 이상인 경우, 하나의 검사일(정상인 검사일)에 대응되는 구간 평균 심박수에서 다른 하나의 검사일(정상인 다른 검사일)에 대응되는 구간 평균 심박수를 뺀 값이 입력값으로 사용되었고, 항진아님 또는 저하아님이 라벨링값으로 사용되었다.For each patient, the value obtained by subtracting the section average heart rate corresponding to the test day when the diagnosis result was normal from the section average heart rate corresponding to the test day when the diagnosis result was not normal is used as an input value (e.g., the amount of change in section average heart rate). and the diagnostic result (increased or decreased) corresponding to the test date when the diagnostic result was not normal was used as the labeling value. In addition, when there were two or more test days with normal diagnostic results, the value obtained by subtracting the section average heart rate corresponding to one test day (the other normal test day) from the section average heart rate corresponding to one test day (the normal test day) was used as the input value. , not high or not low were used as labeling values.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 157명의 환자로부터 확보한 495개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,199개였다.At this time, the clinical data used for learning was 495 data obtained from 157 patients, and the total number of learning data sets created from this was 1,199.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 121명의 환자로부터 확보된 437개의 임상 데이터들로부터 생성되었으며, 총 1,138개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 437 clinical data obtained from a total of 121 patients, and a total of 1,138 test data sets were created.

비교예#1에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표3]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Comparative Example #1 are shown in [Table 3] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV 비교예#1Comparative example #1 71.6271.62 60.6660.66 73.4373.43 27.0127.01 92.0092.00

(2) 실험예#1 - 심박수 차이 + 호르몬수치 (vs 비교예 #1)(2) Experimental example #1 - Heart rate difference + hormone level (vs comparative example #1)

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#1에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #1. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day] Training data sets in the same format were created, and these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4의 농도만 사용하거나 또는 free T4 및 TSH 농도 모두 사용하는 방식을 함께 사용하였다.Additionally, the hormone levels used were either free T4 alone or both free T4 and TSH.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#1에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.Testing the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #1 was conducted using the above-described test sets.

실험예#1에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표4]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #1 are shown in [Table 4] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4free T4 82.3382.33 78.3878.38 83.0883.08 46.7746.77 95.2995.29 free T4 + TSHfree T4+TSH 82.5482.54 86.4986.49 81.7981.79 47.4147.41 96.9696.96

본 출원의 발명자들은, 비교예#1과 같이 호르몬 수치가 "정상"인 때를 기준일로 결정하고, 해당 기준일과 분석 대상일 사이의 구간 평균 심박수 간의 차이를 활용하지 않고, "정상"인지 "갑상선기능항진증"인지 여부와 무관하게 기준일을 결정하고, 해당 기준일에 대응되는 호르몬 농도와 함께 기준일과 분석 대상일 사이의 구간 평균 심박수 간의 차이를 활용하여도 "갑상선기능항진증"인지 여부를 충분히 예측할 수 있다는 것을 알 수 있었으며, 오히려 정확도와 민감도 등의 결과가 훨씬 더 높아지는 것을 알 수 있었다.이로써, 동일한 수준의 임상 데이터로부터 보다 더 많은 학습 데이터 세트를 얻을 수 있게 되었을 뿐만 아니라, 보다 더 정확하게 갑상선기능이상증을 예측할 수 있는 예측모델을 확보할 수 있게 되었다.The inventors of the present application determined the time when the hormone level was "normal" as the reference date, as in Comparative Example #1, and did not use the difference between the average heart rate of the section between the reference date and the date of analysis, and determined whether it was "normal" or "thyroid". It is possible to sufficiently predict whether a person has “hyperthyroidism” by determining the reference date regardless of whether it is “hyperthyroidism” and by using the difference between the average heart rate of the section between the reference date and the analysis target date along with the hormone concentration corresponding to the reference date. In fact, it was found that the results such as accuracy and sensitivity were much higher. This not only made it possible to obtain more learning data sets from the same level of clinical data, but also more accurately diagnosed thyroid dysfunction. It has become possible to secure a predictive model that can make predictions.

(3) 실험예#2 - 심박수 차이 + 현재 심박수 + 호르몬수치 (3) Experimental example #2 - Heart rate difference + current heart rate + hormone level

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#2에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 구간 평균 심박수, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #2. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), section average heart rate corresponding to the first test day, thyroid hormone concentration corresponding to the second test day, corresponding to the first test day. Training data sets in the same format as diagnostic results (hyperactivity or non-hyperactivity disorder) were created, and these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4의 농도만 사용하거나 또는 free T4 및 TSH 농도 모두 사용하는 방식을 함께 사용하였다.Additionally, the hormone levels used were either free T4 alone or both free T4 and TSH.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#2에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.Testing the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #2 was conducted using the test sets described above.

실험예#2에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표5]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #2 are shown in [Table 5] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4free T4 83.6283.62 71.6271.62 85.9085.90 49.0749.07 94.1094.10 free T4 + TSHfree T4+TSH 85.3485.34 74.3274.32 87.4487.44 52.8852.88 94.7294.72

본 출원의 발명자들은, 갑상선기능항진증을 예측함에 있어서, 구간 평균 심박수의 변화량에 더하여 대상일의 구간 평균 심박수(현재의 구간 평균 심박수)를 입력값으로 더 활용하는 경우, 정확도가 조금 더 상승하는 것을 알 수 있었다.The inventors of the present application have found that when predicting hyperthyroidism, the accuracy increases slightly when the section average heart rate (current section average heart rate) of the target day is used as an input value in addition to the change in section average heart rate. Could know.

(4) 실험예#3 - 심박수 차이 + 호르몬수치 + 표준편차 변화(4) Experimental Example #3 - Heart rate difference + hormone level + standard deviation change

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#3에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in Learning Method Example #3 described above. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day) , thyroid hormone concentration corresponding to the second examination day, diagnosis result (hyperactivity or not hyperactivity) corresponding to the first examination day] were created, and these learning data sets were used to train the model. .

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도를 모두 입력값으로 사용하였다.Additionally, the hormone levels used included both free T4 and TSH concentrations as input values.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#3에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.Testing the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #3 was conducted using the test sets described above.

실험예#3에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표6]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #3 are shown in [Table 6] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 81.0381.03 79.7379.73 81.2881.28 44.7044.70 95.4895.48

본 출원의 발명자들은, 갑상선기능항진증을 예측함에 있어서, 구간 평균 심박수의 변화량에 더하여 구간 심박수의 표준편차의 변화량을 입력값으로 더 활용하는 경우, 정확도가 조금 더 상승하는 것을 알 수 있었다.The inventors of the present application found that in predicting hyperthyroidism, when the change in the standard deviation of the section heart rate was used as an input value in addition to the change in the section average heart rate, the accuracy increased slightly.

(5) 실험예#4 - 심박수 차이 + 호르몬수치 + 표준편차 변화 + 왜도 변화(5) Experimental example #4 - Heart rate difference + hormone level + standard deviation change + skewness change

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#4에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 다만, 표준편차 및 왜도를 분포와 관련된 지표로 사용하였다. 즉, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), (제1 검사일에 대응되는 구간 심박수들의 왜도 - 제2 검사일에 대응되는 구간 심박수들의 왜도), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #4. However, standard deviation and skewness were used as indicators related to distribution. That is, [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of section heart rates corresponding to the second test day) , (skewness of segment heart rates corresponding to the first examination day - skewness of segment heart rates corresponding to the second examination day), thyroid hormone concentration corresponding to the second examination day, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first examination day )], training data sets in the same format were created, and these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도를 모두 입력값으로 사용하였다.Additionally, the hormone levels used included both free T4 and TSH concentrations as input values.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#4에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #4 was conducted using the test sets described above.

실험예#4에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표7]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #4 are shown in [Table 7] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 83.6283.62 81.0881.08 84.1084.10 49.1849.18 95.9195.91

(6) 실험예#5 - 심박수 차이 + 호르몬수치 + 표준편차 변화 + 첨도 변화(6) Experimental example #5 - Heart rate difference + hormone level + standard deviation change + kurtosis change

학습 데이터 세트들의 준비Preparation of training data sets

실험예#5는, 실험예#4와 비교할 때, 분포와 관련된 지표로 표준편차 및 왜도를 사용하는 대신 표준편차 및 첨도를 사용하였다는 것을 제외하면 동일하게 학습 데이터 세트들 및 테스트 데이터 세트들이 생성되었다.In Experimental Example #5, compared to Experimental Example #4, the training data sets and test data sets were the same except that standard deviation and kurtosis were used instead of standard deviation and skewness as indices related to distribution. was created.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

실험예#5에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.Testing the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #5 was conducted using the test sets described above.

실험예#5에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표8]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #5 are shown in [Table 8] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 83.1983.19 83.7883.78 83.0883.08 48.4448.44 96.4396.43

(7) 실험예#6 - 심박수 차이 + 호르몬수치 + 표준편차 변화 + JS Divergence(7) Experimental Example #6 - Heart rate difference + hormone level + standard deviation change + JS Divergence

학습 데이터 세트들의 준비Preparation of training data sets

실험예#6은, 실험예#4와 비교할 때, 분포와 관련된 지표의 차이로 표준편차 및 왜도의 차이를 사용하는 대신 표준편차 및 JS Divergence를 사용하였다는 것을 제외하면 동일하게 학습 데이터 세트들 및 테스트 데이터 세트들이 생성되었다. Experimental Example #6, compared to Experimental Example #4, used the same learning data sets except that standard deviation and JS Divergence were used instead of using the difference in standard deviation and skewness due to the difference in indicators related to the distribution. and test data sets were created.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

실험예#6에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #6 was conducted using the test sets described above.

실험예#6에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표9]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #6 are shown in [Table 9] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 83.1983.19 78.3878.38 84.1084.10 48.3348.33 95.3595.35

본 출원의 발명자들은, 실험예#4,5,6을 통해 알 수 있듯이, 갑상선기능항진증을 예측함에 있어서, 표준편차의 변화량에 더하여, 구간 심박수들의 분포의 변화를 관찰하기 위한 추가적인 변수들이 함께 입력값으로 활용될 때 조금씩 정확도가 더 상승하는 것을 알 수 있었다.As can be seen through Experimental Examples #4, 5, and 6, the inventors of the present application input additional variables to observe changes in the distribution of heart rates in addition to the change in standard deviation when predicting hyperthyroidism. It was found that the accuracy gradually increased when used as a value.

(8) 실험예#7 - 심박수 변화율 + 호르몬 수치 + 상대표준편차 변화(8) Experimental Example #7 - Heart rate change rate + hormone level + relative standard deviation change

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#5에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #5. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Learning data sets in the following formats were created: (relative standard deviation of heart rates in the interval corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, and diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test day. , these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도 모두 사용하는 방식이 사용되었다.Additionally, the hormone levels used were based on both free T4 and TSH concentrations.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#7에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #7 was conducted using the test sets described above.

실험예#7에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표10]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #7 are shown in [Table 10] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 83.4183.41 85.1485.14 83.0883.08 48.8448.84 96.7296.72

실험예#7은, 실험예#3과 비교할 때, 구간 평균 심박수의 변화량 및 표준편차의 변화량을 사용하는 대신, 구간 평균 심박수의 변화율 및 상대 표준편차의 변화량을 사용하였다는 점을 제외하면, 동일한 조건으로 진행된 실험이다.본 출원의 발명자들은, 구간 평균 심박수의 변화율 및 상대 표준편차의 변화량이 입력값으로 활용되어도 무방하다는 것을 알 수 있었다.Experimental Example #7 is the same as Experimental Example #3, except that instead of using the change amount of the section average heart rate and the change amount of the standard deviation, the rate of change of the section average heart rate and the change amount of the relative standard deviation were used. This is an experiment conducted under the following conditions. The inventors of the present application found that the rate of change in section average heart rate and the change in relative standard deviation could be used as input values.

(9) 실험예#8 - 심박수 변화율 + 호르몬 수치 + 상대표준편차 변화 + day gap (9) Experimental Example #8 - Heart rate change rate + hormone level + relative standard deviation change + day gap

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#9에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), 제2 검사일과 제1 검사일 사이의 차이(day gap), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in Learning Method Example #9 described above. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Relative standard deviation of zone heart rates corresponding to the second test day), difference between the second test day and the first test day (day gap), thyroid hormone concentration corresponding to the second test day, diagnostic result (hyperactivity disorder) corresponding to the first test day or not) were created and these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도 모두 사용하는 방식이 사용되었다.Additionally, the hormone levels used were based on both free T4 and TSH concentrations.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#8에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the hyperthyroidism prediction model learned according to Experimental Example #8 was tested using the test sets described above.

실험예#8에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표11]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #8 are shown in [Table 11] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV free T4 + TSHfree T4+TSH 83.8483.84 82.4382.43 84.1084.10 49.5949.59 96.1996.19

(10) 실험예#9 - 심박수 변화율 + 호르몬 수치 + 상대표준편차 변화 + 조합값(10) Experimental Example #9 - Heart rate change rate + hormone level + relative standard deviation change + combination value

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#10에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. 즉, [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), 제2 검사일과 제1 검사일 사이의 차이(day gap), 제2 검사일에 대응되는 갑상선 호르몬 농도, 변수들의 조합값, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #10. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Relative standard deviation of interval heart rates corresponding to the second test day), difference between the second test day and the first test day (day gap), thyroid hormone concentration corresponding to the second test day, combination value of variables, corresponding to the first test day Training data sets in the same format as diagnostic results (hyperactivity or non-hyperactivity disorder) were created, and these training data sets were used to train the model.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도 모두 사용하는 방식이 사용되었다.Additionally, the hormone levels used were based on both free T4 and TSH concentrations.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#9에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #9 was conducted using the test sets described above.

실험예#9에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표12] 및 [표 13]과 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #9 are shown in [Table 12] and [Table 13].

제1 값first value 제2 값second value 정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV 1One 심박수 평균 변화율Heart rate average rate of change ÷÷ 제2 기준일 TSHSecond reference date TSH 82.9782.97 87.8487.84 82.0582.05 48.1548.15 97.2697.26 22 상대 표준편차 변화량relative standard deviation change xx 심박수 왜도 변화량Heart rate skewness variation 82.5482.54 85.1485.14 82.0582.05 47.3747.37 96.6896.68 33 제2 기준일 TSHSecond reference date TSH xx 상대 표준편차 변화량relative standard deviation change 84.4884.48 87.8487.84 83.8583.85 50.7850.78 97.3297.32 44 심박수 평균 변화율Heart rate average rate of change xx JS DivergenceJS Divergence 84.4884.48 89.1989.19 83.5983.59 50.7750.77 97.6097.60 55 제2 기준일 free T4Second reference date free T4 ÷÷ 제2 기준일 TSHSecond reference date TSH 83.8483.84 89.1989.19 82.8282.82 49.6249.62 97.5897.58 66 JS DivergenceJS Divergence ÷÷ 제2 기준일 TSHSecond reference date TSH 83.4183.41 87.8487.84 82.5682.56 48.8748.87 97.2897.28 77 심박수 첨도 변화량Heart rate kurtosis change ÷÷ 상대 표준편차 변화량relative standard deviation change 83.4183.41 89.1989.19 82.3182.31 48.8948.89 97.5797.57 88 심박수 평균 변화율Heart rate average rate of change ÷÷ 심박수 왜도 변화량Heart rate skewness variation 84.0584.05 89.1989.19 83.0883.08 50.0050.00 97.5997.59 99 심박수 평균 변화율Heart rate average rate of change ÷÷ JS DivergenceJS Divergence 84.2784.27 87.8487.84 83.5983.59 50.3950.39 97.3197.31 1010 제2 기준일 free T4Second reference date free T4 ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 84.4884.48 89.1989.19 83.5983.59 50.7750.77 97.6097.60 1111 상대 표준편차 변화량relative standard deviation change ÷÷ 제2 기준일 free T4Second reference date free T4 84.2784.27 87.8487.84 83.5983.59 50.3950.39 97.3197.31 1212 제2 기준일 TSHSecond reference date TSH ÷÷ 제2 기준일 free T4Second reference date free T4 84.2784.27 87.8487.84 83.5983.59 50.3950.39 97.3197.31 1313 상대 표준편차 변화량relative standard deviation change ÷÷ 심박수 평균 변화율Heart rate average rate of change 83.4183.41 89.1989.19 82.3182.31 48.8948.89 97.5797.57 1414 심박수 왜도 변화량Heart rate skewness variation ÷÷ 제2 기준일 free T4Second reference date free T4 83.4183.41 87.8487.84 82.5682.56 48.8748.87 97.2897.28 1515 제2 기준일 TSHSecond reference date TSH ÷÷ 심박수 왜도 변화량Heart rate skewness variation 83.1983.19 86.4986.49 82.5682.56 48.4848.48 96.9996.99 1616 제2 기준일 TSHSecond reference date TSH ÷÷ 상대 표준편차 변화량relative standard deviation change 84.0584.05 87.8487.84 83.3383.33 50.0050.00 97.3197.31 1717 심박수 평균 변화율Heart rate average rate of change ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 83.6283.62 89.1989.19 82.5682.56 49.2549.25 97.5897.58 1818 심박수 첨도 변화량Heart rate kurtosis change xx JS DivergenceJS Divergence 84.4884.48 87.8487.84 83.8583.85 50.7850.78 97.3297.32 1919 심박수 왜도 변화량Heart rate skewness variation ÷÷ 제2 기준일 TSHSecond reference date TSH 83.6283.62 89.1989.19 82.5682.56 49.2549.25 97.5897.58 2020 제2 기준일 TSHSecond reference date TSH xx 심박수 왜도 변화량Heart rate skewness variation 84.2784.27 90.5490.54 83.0883.08 50.3850.38 97.8997.89 2121 제2 기준일 free T4Second reference date free T4 xx 심박수 첨도 변화량Heart rate kurtosis change 83.8483.84 87.8487.84 83.0883.08 49.6249.62 97.3097.30 2222 상대 표준편차 변화량relative standard deviation change ÷÷ JS DivergenceJS Divergence 83.8483.84 89.1989.19 82.8282.82 49.6249.62 97.5897.58 2323 상대 표준편차 변화량relative standard deviation change xx JS DivergenceJS Divergence 84.0584.05 87.8487.84 83.3383.33 50.0050.00 97.3197.31 2424 제2 기준일 free T4Second reference date free T4 xx 심박수 왜도 변화량Heart rate skewness variation 83.4183.41 86.4986.49 82.8282.82 48.8548.85 97.0097.00 2525 제2 기준일 free T4Second reference date free T4 ÷÷ 심박수 평균 변화율Heart rate average rate of change 84.2784.27 89.1989.19 83.3383.33 50.3850.38 97.6097.60 2626 심박수 첨도 변화량Heart rate kurtosis change ÷÷ 심박수 평균 변화율Heart rate average rate of change 83.1983.19 86.4986.49 82.5682.56 48.4848.48 96.9996.99 2727 JS DivergenceJS Divergence ÷÷ 심박수 왜도 변화량Heart rate skewness variation 83.8483.84 89.1989.19 82.8282.82 49.6249.62 97.5897.58 2828 심박수 왜도 변화량Heart rate skewness variation ÷÷ 상대 표준편차 변화량relative standard deviation change 83.8483.84 87.8487.84 83.0883.08 49.6249.62 97.3097.30 2929 제2 기준일 free T4Second reference date free T4 xx JS DivergenceJS Divergence 84.0584.05 89.1989.19 83.0883.08 50.0050.00 97.5997.59 3030 심박수 평균 변화율Heart rate average rate of change xx 제2 기준일 free T4Second reference date free T4 84.0584.05 87.8487.84 83.3383.33 50.0050.00 97.3197.31 3131 심박수 평균 변화율Heart rate average rate of change xx 상대 표준편차 변화량relative standard deviation change 83.6283.62 89.1989.19 82.5682.56 49.2549.25 97.5897.58 3232 심박수 왜도 변화량Heart rate skewness variation xx 심박수 첨도 변화량Heart rate kurtosis change 83.6283.62 86.4986.49 83.0883.08 49.2349.23 97.0197.01 3333 제2 기준일 TSHSecond reference date TSH ÷÷ JS DivergenceJS Divergence 83.1983.19 89.1989.19 82.0582.05 48.5348.53 97.5697.56 3434 상대 표준편차 변화량relative standard deviation change ÷÷ 심박수 왜도 변화량Heart rate skewness variation 83.6283.62 89.1989.19 82.5682.56 49.2549.25 97.5897.58 3535 제2 기준일 free T4Second reference date free T4 xx 제2 기준일 TSHSecond reference date TSH 83.6283.62 87.8487.84 82.8282.82 49.2449.24 97.2997.29 3636 심박수 평균 변화율Heart rate average rate of change ÷÷ 상대 표준편차 변화량relative standard deviation change 84.2784.27 89.1989.19 83.3383.33 50.3850.38 97.6097.60 3737 심박수 첨도 변화량Heart rate kurtosis change ÷÷ 심박수 왜도 변화량Heart rate skewness variation 83.8483.84 89.1989.19 82.8282.82 49.6249.62 97.5897.58 3838 제2 기준일 TSHSecond reference date TSH ÷÷ 심박수 평균 변화율Heart rate average rate of change 84.4884.48 89.1989.19 83.5983.59 50.7750.77 97.6097.60 3939 JS DivergenceJS Divergence ÷÷ 심박수 평균 변화율Heart rate average rate of change 84.4884.48 89.1989.19 83.5983.59 50.7750.77 97.6097.60

제1 값first value 제2 값second value 정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV 4040 제2 기준일 TSHSecond reference date TSH xx 심박수 첨도 변화량Heart rate kurtosis change 83.8483.84 87.8487.84 83.0883.08 49.6249.62 97.3097.30 4141 심박수 평균 변화율Heart rate average rate of change xx 심박수 왜도 변화량Heart rate skewness variation 84.4884.48 89.1989.19 83.5983.59 50.7750.77 97.6097.60 4242 심박수 평균 변화율Heart rate average rate of change xx 심박수 첨도 변화량Heart rate kurtosis change 84.7084.70 89.1989.19 83.8583.85 51.1651.16 97.6197.61 4343 JS DivergenceJS Divergence ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 84.0584.05 86.4986.49 83.5983.59 50.0050.00 97.0297.02 4444 제2 기준일 free T4Second reference date free T4 ÷÷ JS DivergenceJS Divergence 85.3485.34 90.5490.54 84.3684.36 52.3452.34 97.9297.92 4545 심박수 첨도 변화량Heart rate kurtosis change ÷÷ 제2 기준일 free T4Second reference date free T4 83.6283.62 87.8487.84 82.8282.82 49.2449.24 97.2997.29 4646 상대 표준편차 변화량relative standard deviation change ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 83.4183.41 89.1989.19 82.3182.31 48.8948.89 97.5797.57 4747 심박수 평균 변화율Heart rate average rate of change xx 제2 기준일 TSHSecond reference date TSH 83.6283.62 87.8487.84 82.8282.82 49.2449.24 97.2997.29 4848 심박수 왜도 변화량Heart rate skewness variation ÷÷ JS DivergenceJS Divergence 84.4884.48 87.8487.84 83.8583.85 50.7850.78 97.3297.32 4949 제2 기준일 free T4Second reference date free T4 xx 상대 표준편차 변화량relative standard deviation change 84.2784.27 90.5490.54 83.0883.08 50.3850.38 97.8997.89 5050 심박수 왜도 변화량Heart rate skewness variation xx JS DivergenceJS Divergence 84.2784.27 87.8487.84 83.5983.59 50.3950.39 97.3197.31 5151 JS DivergenceJS Divergence ÷÷ 상대 표준편차 변화량relative standard deviation change 84.7084.70 86.4986.49 84.3684.36 51.2051.20 97.0597.05 5252 상대 표준편차 변화량relative standard deviation change xx 심박수 첨도 변화량Heart rate kurtosis change 83.6283.62 87.8487.84 82.8282.82 49.2449.24 97.2997.29 5353 상대 표준편차 변화량relative standard deviation change ÷÷ 제2 기준일 TSHSecond reference date TSH 83.6283.62 87.8487.84 82.8282.82 49.2449.24 97.2997.29 5454 제2 기준일 free T4Second reference date free T4 ÷÷ 심박수 왜도 변화량Heart rate skewness variation 83.1983.19 87.8487.84 82.3182.31 48.5148.51 97.2797.27 5555 JS DivergenceJS Divergence ÷÷ 제2 기준일 free T4Second reference date free T4 84.0584.05 87.8487.84 83.3383.33 50.0050.00 97.3197.31 5656 심박수 첨도 변화량Heart rate kurtosis change ÷÷ JS DivergenceJS Divergence 84.0584.05 86.4986.49 83.5983.59 50.0050.00 97.0297.02 5757 제2 기준일 TSHSecond reference date TSH xx JS DivergenceJS Divergence 84.0584.05 87.8487.84 83.3383.33 50.0050.00 97.3197.31 5858 심박수 왜도 변화량Heart rate skewness variation ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 83.8483.84 87.8487.84 83.0883.08 49.6249.62 97.3097.30 5959 심박수 왜도 변화량Heart rate skewness variation ÷÷ 심박수 평균 변화율Heart rate average rate of change 84.7084.70 89.1989.19 83.8583.85 51.1651.16 97.6197.61 6060 제2 기준일 TSHSecond reference date TSH ÷÷ 심박수 첨도 변화량Heart rate kurtosis change 84.0584.05 86.4986.49 83.5983.59 50.0050.00 97.0297.02 6161 제2 기준일 free T4Second reference date free T4 ÷÷ 상대 표준편차 변화량relative standard deviation change 83.8483.84 90.5490.54 82.5682.56 49.6349.63 97.8797.87 6262 심박수 첨도 변화량Heart rate kurtosis change ÷÷ 제2 기준일 TSHSecond reference date TSH 83.6283.62 89.1989.19 82.5682.56 49.2549.25 97.5897.58 6363 심박수 평균 변화율Heart rate average rate of change ÷÷ 제2 기준일 free T4Second reference date free T4 84.7084.70 89.1989.19 83.8583.85 51.1651.16 97.6197.61

(11) 실험예#10 - 대상의 개인 정보 활용(11) Experimental Example #10 - Utilization of the subject’s personal information

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#11에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #11.

특히, (1) [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 free T4의 농도, 대상의 나이, 대상의 성별, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]과 같은 학습 데이터 세트들을 이용하여 학습이 진행되었으며, 또한 (2) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 free T4의 농도, 대상의 나이, 대상의 성별, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]과 같은 학습 데이터 세트들을 이용하여 학습이 진행되었고, 나아가 (3) [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 대상의 나이, 대상의 성별, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]과 같은 학습데이터 세트들을 이용하여 학습이 진행되었으며, 또한 (4) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 대상의 나이, 대상의 성별, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]과 같은 학습데이터 세트들을 이용하여 학습이 진행되었고, 마지막으로 (5) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 대상의 나이, 대상의 성별, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 학습 데이터 세트들을 이용하여 학습이 진행되었다.In particular, (1) [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), concentration of free T4 corresponding to the second test day, subject's age, subject's gender, first test day Learning was conducted using learning data sets such as (2) [((((section average heart rate corresponding to the first test day) - section average corresponding to the second test day). heart rate) / section average heart rate corresponding to the second test day), concentration of free T4 corresponding to the second test day, age of the subject, gender of the subject, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first test day] Learning was carried out using learning data sets, and further (3) [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), concentration of free T4 corresponding to the second test day, 2 Learning was conducted using learning data sets such as TSH concentration corresponding to the test date, subject's age, subject's gender, and diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date, and (4) [ ((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), concentration of free T4 corresponding to the second test day, corresponding to the second test day Learning was carried out using learning data sets such as TSH concentration, subject's age, subject's gender, and diagnosis result (hyperactivity or non-hyperactivity) corresponding to the first test date, and finally (5) [((( 1 Zone average heart rate corresponding to the second inspection day - Section average heart rate corresponding to the second inspection day) / Section average heart rate corresponding to the second inspection day), (Relative standard deviation of section heart rates corresponding to the first inspection day - Corresponding to the second inspection day Relative standard deviation of section heart rates corresponding to the second test day), (skewness of the section heart rates corresponding to the second test day - skewness of the section heart rates corresponding to the first test day), (kurtosis of the section heart rates corresponding to the second test day - first test day kurtosis of the interval heart rates corresponding to), (JS Divergence between the interval heart rates corresponding to the second test day and the interval heart rates corresponding to the first test day), the concentration of free T4 corresponding to the second test day, the second test day Learning was conducted using the following learning data sets: corresponding TSH concentration, subject's age, subject's gender, and diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test date.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#10에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #10 was conducted using the test sets described above.

실험예#10에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표14]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #10 are shown in [Table 14] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV (1)(One) 84.2784.27 53.3353.33 85.3085.30 10.8110.81 98.2198.21 (2)(2) 83.4183.41 53.3353.33 84.4184.41 10.2610.26 98.1998.19 (3)(3) 83.4183.41 53.3353.33 84.4184.41 10.2610.26 98.1998.19 (4)(4) 84.4884.48 53.3353.33 85.5285.52 10.9610.96 98.2198.21 (5)(5) 85.9985.99 60.0060.00 86.8686.86 13.2413.24 98.4898.48

(12) 실험예#11 - 미리 정해진 구간의 길이에 대한 실험(12) Experimental Example #11 - Experiment on the length of a predetermined section

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#10에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #10.

특히, 분포의 변화를 확인하기 위한 지표들로 상대 표준편차의 변화량, 왜도의 변화량, 첨도의 변화량, JS Divergence를 사용하였으며, 나아가, 변수들의 조합값으로 (제2 기준일의 free T4 농도 / JS divergence)를 사용하였다. 즉, [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일 및 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일과 제1 검사일 사이의 차이(day gap), 제2 검사일에 대응되는 갑상선 호르몬 농도, (제2 검사일 free T4의 농도 / 제2 검사일 및 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷의 학습 데이터 세트들이 생성되었고, 이 학습 데이터 세트들이 모델을 학습시키는 데에 사용되었다.In particular, the change in relative standard deviation, change in skewness, change in kurtosis, and JS Divergence were used as indicators to check changes in distribution. Furthermore, as a combination of variables (free T4 concentration on the second reference date / JS divergence) was used. That is, [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of the section heart rates corresponding to the first test day) - Relative standard deviation of section heart rates corresponding to the second test day), (skewness of section heart rates corresponding to the second test day - Skewness of section heart rates corresponding to the first test day), (zone heart rate corresponding to the second test day kurtosis of - kurtosis of the interval heart rates corresponding to the first test day), (JS Divergence between the interval heart rates corresponding to the second test day and the first test day), difference between the second test day and the first test day (day gap), Thyroid hormone concentration corresponding to the second test day, (concentration of free T4 on the second test day / JS Divergence between the zone heart rates corresponding to the second test day and the first test day), diagnostic result (hyperactivity or hyperactivity) corresponding to the first test day (not)], training data sets in the same format were created, and these training data sets were used to train the model.

본 실험예#11에서는 상기 검사일에 대응되는 구간 평균 심박수를 산출하기 위한 미리 정해진 기간을 1, 5, 10, 15, 25 및 30일로 변경하면서 정확도에 미치는 영향을 확인하였다. In this Experimental Example #11, the predetermined period for calculating the section average heart rate corresponding to the test date was changed to 1, 5, 10, 15, 25, and 30 days to confirm the effect on accuracy.

그리고, 사용된 호르몬 수치는 free T4 및 TSH 농도 모두 사용하는 방식이 사용되었다.And, the hormone levels used were based on both free T4 and TSH concentrations.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#11에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the hyperthyroidism prediction model learned according to Experimental Example #11 was tested using the test sets described above.

실험예#11에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표15]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #11 are shown in [Table 15] below.

미리 정해진 기간
(days)
a predetermined period of time
(days)
정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV
1One 80.6080.60 71.6271.62 82.3182.31 43.4443.44 93.8693.86 55 81.9081.90 74.3274.32 83.3383.33 45.8345.83 94.4894.48 1010 83.4183.41 79.7379.73 84.1084.10 48.7648.76 95.6395.63 1515 86.2186.21 81.0881.08 87.1887.18 54.5554.55 96.0596.05 2020 84.4884.48 85.1485.14 84.3684.36 50.8150.81 96.7696.76 2525 82.5482.54 81.0881.08 82.8282.82 47.2447.24 95.8595.85 3030 82.9782.97 79.7379.73 83.5983.59 47.9747.97 95.6095.60

본 출원의 발명자들은, 갑상선기능항진증을 예측함에 있어서, 구간 심박수들을 산출하기 위한 미리 정해진 기간은 15일을 사용하였을 때 가장 높은 정확도가 나오는 것을 알 수 있었다.The inventors of the present application found that, in predicting hyperthyroidism, the highest accuracy was obtained when 15 days was used as a predetermined period for calculating zone heart rates.

(13) 실험예#12 - 심박수 변화율 + 호르몬수치 + 추정 호르몬 수치(13) Experimental Example #12 - Heart rate change rate + hormone level + estimated hormone level

학습 데이터 세트들의 준비Preparation of training data sets

총 126명의 환자를 대상으로 수집한 총 464개의 임상 데이터 세트들이 사용되었다.A total of 464 clinical data sets collected from a total of 126 patients were used.

전술한 학습방법 실시예#12에 기재한 갑상선기능항진증 예측 모델의 학습 방법에 따라 모델이 학습되었다. The model was learned according to the learning method of the hyperthyroidism prediction model described in the above-described learning method example #12.

특히, (1) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), (제2 검사일에 대응되는 free T4의 농도/제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)] (2) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), (제2 검사일에 대응되는 free T4의 농도/제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제1 검사일에 대응되는 free T4의 추정 호르몬 농도에 관한 값, 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 학습 데이터 세트들을 이용하여 학습이 진행되었다.In particular, (1) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (segment heart rate corresponding to the first test day) Relative standard deviation - relative standard deviation of the interval heart rates corresponding to the second test day), (skewness of the interval heart rates corresponding to the second test day - skewness of the interval heart rates corresponding to the first test day), (skewness of the interval heart rates corresponding to the second test day) kurtosis of the zone heart rates corresponding to the first test day), (JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day), (corresponding to the second test day) concentration of free T4/JS Divergence between zone heart rates corresponding to the second test day and zone heart rates corresponding to the first test day), concentration of free T4 corresponding to the second test day, TSH corresponding to the second test day Concentration, diagnostic result corresponding to the first test day (hyperactivity or not hyperactivity)] (2) [(((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / corresponding to the second test day interval average heart rate), (relative standard deviation of interval heart rates corresponding to the first test day - relative standard deviation of interval heart rates corresponding to the second test day), (skewness of interval heart rates corresponding to the second test day - first test day Skewness of the corresponding interval heart rates), (Kurtosis of the interval heart rates corresponding to the second test day - Kurtosis of the interval heart rates corresponding to the first test day), (Kurtosis of the interval heart rates corresponding to the second test day and corresponding to the first test day (JS Divergence between zone heart rates), (Concentration of free T4 corresponding to the second test day/JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day), on the first test day Values for the corresponding estimated hormone concentration of free T4, concentration of free T4 corresponding to the second test day, concentration of TSH corresponding to the second test day, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first test day] Learning was conducted using training data sets.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 126명의 환자로부터 확보한 464개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,416개였다. 이들 중 일부가 테스트 데이터 세트로 이용되었고, 실험예#12에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 데이터 세트들을 이용하여 진행되었다.At this time, the clinical data used for learning was 464 data obtained from 126 patients, and the total number of learning data sets created from this was 1,416. Some of these were used as test data sets, and the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #12 was tested using the test data sets described above.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

실험예#12에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표16]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #12 are shown in [Table 16] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV (1)(One) 83.4183.41 79.7379.73 84.1084.10 48.7648.76 95.6395.63 (2)(2) 87.0787.07 89.1989.19 86.6786.67 55.9355.93 97.6997.69

본 출원의 발명자들은, 갑상선기능항진증을 예측함에 있어서, 대상일을 추정 갑상선 호르몬 수치에 관한 값을 함께 사용하였을 때 정확도, 민감도, 특이도, PPV 및 NPV 모두의 성능 향상이 있음을 확인할 수 있었다. 또한, 갑상선기능항진증에서 확인되는 성능 향상은 갑상선기능저하증을 예측함에 있어서도 유사하게 발현될 것으로 보인다.The inventors of the present application were able to confirm that when predicting hyperthyroidism, the performance of both accuracy, sensitivity, specificity, PPV, and NPV was improved when the target date was used together with the value of the estimated thyroid hormone level. In addition, the performance improvement observed in hyperthyroidism is expected to be similar in predicting hypothyroidism.

이상, 갑상선기능이상증 예측모델의 학습방법, 이를 이용한 갑상선기능이상증 예측방법 및 이를 위한 시스템에 대하여 설명하였다.Above, the learning method of the dysthyroidism prediction model, the dysthyroidism prediction method using it, and the system for this were explained.

이하에서는, 갑상선 호르몬 수치를 예측하는 모델의 학습방법 및 이를 이용한 갑상선기능이상증 예측방법 및 이를 위한 시스템에 대하여 간단히 설명한다.Below, we will briefly describe the learning method of a model for predicting thyroid hormone levels, the method for predicting thyroid dysfunction using the same, and the system for the same.

6. 갑상선 호르몬 예측모델의 학습방법6. Learning method of thyroid hormone prediction model

학습방법 실시예#13. 기준일 심박수와의 차이값과 기준일의 호르몬값Learning method example #13. Difference value from heart rate on reference day and hormone value on reference day

본 출원에 의해 개시되는 제13 실시예에 따르면, 학습 모델을 학습시키기 위하여, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값 및 제2 검사일의 호르몬 농도를 입력값으로 사용하고, 상기 제1 검사일에 대응되어 있는 호르몬 농도를 라벨링값으로 사용할 수 있다. According to the 13th embodiment disclosed by the present application, in order to learn a learning model, the value obtained by subtracting the section average heart rate corresponding to the second test day from the section average heart rate corresponding to the first test day and the hormone concentration of the second test day It can be used as an input value, and the hormone concentration corresponding to the first test date can be used as a labeling value.

이때, 상기 학습 모델은 regression을 할 수 있는 머신러닝 모델이 사용될 수 있다. 예를 들어, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등이 사용될 수 있다.At this time, a machine learning model capable of regression may be used as the learning model. For example, the learning model may be light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc.

상기 갑상선 호르몬 농도 예측모델(thyroid hormone concentration prediction model)은 대상일의 TRH 농도 예측모델, TSH 농도 예측모델, T4 농도 예측모델, free T4 농도 예측모델, T3 농도 예측모델 및 free T3 농도 예측모델로 이루어진 그룹에서 선택된 적어도 하나를 포함한다. The thyroid hormone concentration prediction model consists of a TRH concentration prediction model, a TSH concentration prediction model, a T4 concentration prediction model, a free T4 concentration prediction model, a T3 concentration prediction model, and a free T3 concentration prediction model on the target day. Contains at least one selected from the group.

구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, It can be one of the following days: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30.

갑상선 호르몬 농도 예측모델의 종류에 따라서, 상기 라벨링 값은 상기 제1 검사일에 대응되어 있는 복수의 호르몬 농도들 중 상기 종류에 대응되는 농도값이 사용될 수 있다. 예를 들어, free T4 농도 예측모델을 학습시키고자 할 때는, 상기 라벨링값으로 상기 제1 검사일에 대응되어 있는 free T4의 농도값이 사용된다. 다른 예를 들어, TSH 농도 예측모델을 학습시키고자 할 때는, 상기 라벨링값으로 상기 제1 검사일에 대응되어 있는 TSH의 농도값이 사용된다.Depending on the type of thyroid hormone concentration prediction model, the labeling value may be a concentration value corresponding to the type among a plurality of hormone concentrations corresponding to the first test date. For example, when learning a free T4 concentration prediction model, the free T4 concentration value corresponding to the first inspection date is used as the labeling value. For another example, when learning a TSH concentration prediction model, the TSH concentration value corresponding to the first test date is used as the labeling value.

이하에서는, 설명의 편의를 위하여, free T4 농도 예측모델을 학습시키는 것을 기준으로 설명한다.Below, for convenience of explanation, the description is based on learning a free T4 concentration prediction model.

제13 실시예에 따라, free T4 농도 예측 모델을 학습시키기 위하여, 확보된 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 free T4 농도]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다. 이때, 상기 제2 검사일에 대응되는 갑상선 호르몬 농도는 TRH농도, TSH 농도, T4 농도, free T4 농도, T3 농도 및 free T3 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수 있다.According to the 13th embodiment, in order to learn a free T4 concentration prediction model, the obtained clinical data were used to calculate [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), second inspection day. Learning data sets in the following format can be created: thyroid hormone concentration corresponding to , free T4 concentration corresponding to the first test date. At this time, the thyroid hormone concentration corresponding to the second test day may be one selected from the group consisting of TRH concentration, TSH concentration, T4 concentration, free T4 concentration, T3 concentration, and free T3 concentration, or a combination thereof.

예를 들어, 제1 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사 시, 확인된 free T4 호르몬 농도는 제1 free T4 농도였고 확인된 TSH의 농도는 제1 TSH 농도이고, 두번째 검사 시, 확인된 free T4 호르몬 농도는 제2 free T4 농도였고 확인된 TSH의 농도는 제2 TSH 농도이며, 세번째 검사 시, 확인된 free T4 호르몬 농도는 제3 free T4 농도였고 확인된 TSH의 농도는 제3 TSH 농도라고 가정한다면, 해당 환자의 임상 데이터로부터 확보될 수 있는 학습 데이터 세트들은 다음과 같을 수 있다.For example, there are a total of three data sets obtained from the first patient, and at the first test, the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration, and at the second test, The confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration. At the third test, the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the third TSH concentration. Assuming the TSH concentration, the learning data sets that can be obtained from the patient's clinical data may be as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 free T4 농도, 제2 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), second free T4 concentration, second TSH concentration, first free T4 concentration],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 free T4 농도, 제3 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), third free T4 concentration, third TSH concentration, first free T4 concentration],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 free T4 농도, 제1 TSH 농도, 제2 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), first free T4 concentration, first TSH concentration, second free T4 concentration],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 free T4 농도, 제3 TSH 농도, 제3 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), third free T4 concentration, third TSH concentration, third free T4 concentration],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 free T4 농도, 제1 TSH 농도, 제3 free T4 농도], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), first free T4 concentration, first TSH concentration, third free T4 concentration], and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 free T4 농도, 제2 TSH 농도, 제3 free T4 농도][(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), second free T4 concentration, second TSH concentration, third free T4 concentration]

다른 예를 들어, 제2 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사 시, 확인된 free T4 호르몬 농도는 제1 free T4 농도였고 확인된 TSH의 농도는 제1 TSH 농도이고, 두번째 검사 시, 확인된 free T4 호르몬 농도는 제2 free T4 농도였고 확인된 TSH의 농도는 제2 TSH 농도이며, 세번째 검사 시, 확인된 free T4 호르몬 농도는 제3 free T4 농도였고 확인된 TSH의 농도는 제3 TSH 농도이고, 네번째 검사 시, 확인된 free T4 호르몬 농도는 제4 free T4 농도였고 확인된 TSH의 농도는 제4 TSH 농도라고 가정한다면, 해당 환자의 임상 데이터로부터 확보될 수 있는 학습 데이터 세트들은 다음과 같을 수 있다.As another example, there are a total of 4 data sets obtained from the second patient, and at the first test, the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration, and at the second test , the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration. At the third test, the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the second TSH concentration. 3 TSH concentration, and at the time of the fourth test, assuming that the confirmed free T4 hormone concentration was the 4th free T4 concentration and the confirmed TSH concentration was the 4th TSH concentration, the learning data sets that can be obtained from the patient's clinical data are It could be like this:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 free T4 농도, 제2 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), second free T4 concentration, second TSH concentration, first free T4 concentration],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 free T4 농도, 제3 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), third free T4 concentration, third TSH concentration, first free T4 concentration],

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 free T4 농도, 제4 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), fourth free T4 concentration, fourth TSH concentration, first free T4 concentration],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 free T4 농도, 제1 TSH 농도, 제2 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), first free T4 concentration, first TSH concentration, second free T4 concentration],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 free T4 농도, 제3 TSH 농도, 제2 free T4 농도],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), third free T4 concentration, third TSH concentration, second free T4 concentration],

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 free T4 농도, 제4 TSH 농도, 제2 free T4 농도],[(section average heart rate corresponding to the 2nd test day - section average heart rate corresponding to the 4th test day), 4th free T4 concentration, 4th TSH concentration, 2nd free T4 concentration],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 free T4 농도, 제1 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), first free T4 concentration, first TSH concentration, third free T4 concentration],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 free T4 농도, 제2 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), second free T4 concentration, second TSH concentration, third free T4 concentration],

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제4 free T4 농도, 제4 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the 3rd test day - section average heart rate corresponding to the 4th test day), 4th free T4 concentration, 4th TSH concentration, 3rd free T4 concentration],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제1 free T4 농도, 제1 TSH 농도, 제4 free T4 농도],[(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the first test day), first free T4 concentration, first TSH concentration, fourth free T4 concentration],

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 free T4 농도, 제2 TSH 농도, 제4 free T4 농도], 및[(section average heart rate corresponding to the fourth test day - section average heart rate corresponding to the second test day), second free T4 concentration, second TSH concentration, fourth free T4 concentration], and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제3 free T4 농도, 제3 TSH 농도, 제4 free T4 농도][(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), 3rd free T4 concentration, 3rd TSH concentration, 4th free T4 concentration]

만약, free T4의 농도를 예측하기 위한 모델이 아니라, TSH의 농도를 예측하기 위한 모델을 학습시키고자 하는 경우, 전술한 학습 데이터 세트들에서 입력값은 그대로 유지하고, 라벨링값만 free T4의 농도를 사용하는 대신 TSH의 농도를 사용하면 된다.If you want to train a model for predicting the concentration of TSH rather than a model for predicting the concentration of free T4, keep the input values in the above-mentioned training data sets as is and only the labeling value is the concentration of free T4. Instead of using , you can use the concentration of TSH.

학습방법 실시예#14. 기준일 심박수와의 변화율과 기준일의 호르몬값Learning method example #14. Rate of change from heart rate on the reference day and hormone values on the reference date

본 출원에 의해 개시되는 제14 실시예에 따르면, 전술한 제13 실시예와 비교할 때, 제1 검사일에 대응되는 구간 평균 심박수에서 제2 검사일에 대응되는 구간 평균 심박수를 뺀 값 대신 제1 검사일에 대한 제2 검사일의 구간 평균 심박수의 변화율을 입력값으로 사용하며, 나머지는 제13 실시예와 동일하다.According to the fourteenth embodiment disclosed by the present application, compared to the above-described thirteenth embodiment, on the first examination day instead of the interval average heart rate corresponding to the first examination day minus the interval average heart rate corresponding to the second examination day The change rate of the section average heart rate on the second test day is used as an input value, and the rest is the same as in the 13th embodiment.

학습방법 실시예#15. 기준일 심박수와의 변화율, 상대 표준편차 변화량과 기준일의 호르몬값Learning method example #15. Rate of change from heart rate on the reference day, relative standard deviation change and hormone value on the reference day

본 출원에 의해 개시되는 제15 실시예에 따르면, 전술한 제14 실시예와 비교할 때, 제1 검사일에 대응되는 구간 심박수들의 상대 표준편차에서 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차를 뺀 값을 입력값으로 더 사용하며, 나머지는 제14 실시예와 동일하다.According to the 15th embodiment disclosed by the present application, compared to the 14th embodiment described above, the relative standard deviation of the zone heart rates corresponding to the first inspection day is subtracted from the relative standard deviation of the zone heart rates corresponding to the second inspection day. The value is further used as an input value, and the rest is the same as the 14th embodiment.

학습방법 실시예#16. 기준일 심박수와의 변화율, 상대 표준편차 변화량, 왜도변화량, 첨도 변화량, JS Divergence와 기준일의 호르몬값Learning method example #16. Rate of change from heart rate on the reference day, relative standard deviation change, skewness change, kurtosis change, JS Divergence and hormone value on the reference day

본 출원에 의해 개시되는 제16 실시예에 따르면, 전술한 제15 실시예와 비교할 때, 제1 검사일에 대응되는 구간 심박수들의 분포(distribution)에 관한 지표와 제2 검사일에 대응되는 구간 심박수들의 분포에 관한 지표 사이의 차이에 관한 값을 입력값으로 더 사용하며, 나머지는 제15 실시예와 동일하다.According to the sixteenth embodiment disclosed by the present application, compared to the above-described fifteenth embodiment, an index regarding the distribution of zone heart rates corresponding to the first inspection day and the distribution of zone heart rates corresponding to the second inspection day The value of the difference between the indicators is further used as an input value, and the rest is the same as the 15th embodiment.

학습방법 실시예#17. 기준일 심박수와의 변화율, 상대 표준편차 변화량, 왜도변화량, 첨도 변화량, JS Divergence와 기준일의 호르몬값 및 Day GapLearning method example #17. Rate of change from heart rate on the reference day, relative standard deviation change, skewness change, kurtosis change, JS Divergence, hormone value on the reference date, and Day Gap

본 출원에 의해 개시되는 제17 실시예에 따르면, 전술한 제16 실시예와 비교할 때, 제1 검사일과 제2 검사일 사이의 차이(day gaps)에 관한 값을 입력값으로 더 사용하며, 나머지는 제16 실시예와 동일하다.According to the 17th embodiment disclosed by the present application, compared to the above-described 16th embodiment, the value related to the difference (day gaps) between the first inspection date and the second inspection date is further used as an input value, and the remaining Same as Example 16.

학습방법 실시예#18. 기준일 심박수와의 차이값, 기준일의 호르몬값 및 추정 갑상선 상태Learning method example #18. Difference value from heart rate on reference date, hormone value on reference date and estimated thyroid status

본 출원에 의해 개시되는 제18 실시예에 따르면, 학습 모델을 학습시키기 위하여, 전술한 제13 실시예 내지 제17 실시예에서 설명하는 입력값에 더하여, 제1 검사일에 대응되는 갑상선 기능에 관한 값이 입력값으로 사용할 수 있다. According to the 18th embodiment disclosed by the present application, in order to learn a learning model, in addition to the input values described in the 13th to 17th embodiments described above, a value related to thyroid function corresponding to the first examination date This can be used as input value.

제1 검사일에 대응되는 갑상선 기능에 관한 값은 갑상선 기능 분류모델의 출력값을 기초로 획득될 수 있다. 일 예로, 갑상선 기능에 관한 값은 갑상선 기능 분류모델의 출력값을 기초로 추정된 대상일의 갑상선 기능을 나타낸 값일 수 있다. 다른 예로, 갑상선 기능에 관한 값은 갑상선 기능 분류모델의 출력값일 수 있다.Values related to thyroid function corresponding to the first test date can be obtained based on the output value of the thyroid function classification model. As an example, the value regarding thyroid function may be a value representing the thyroid function on the target date estimated based on the output value of the thyroid function classification model. As another example, the value related to thyroid function may be the output value of a thyroid function classification model.

제1 검사일에 대응되는 갑상선 기능에 관한 값은 하나 이상의 값일 수 있다. 예를 들어, 갑상선 항진증에 특화된 갑상선 기능 분류 모델의 출력값을 기초로 생성된 제1 값 및 갑상선 저하증에 특화된 갑상선 기능 분류 모델의 출력값을 기초로 생성된 제2 값을 모두 이용하여 학습 모델을 학습시킬 수 있으나, 이에 한정되지 않는다.The value regarding thyroid function corresponding to the first test date may be one or more values. For example, a learning model can be trained using both the first value generated based on the output value of a thyroid function classification model specialized for hyperthyroidism and the second value generated based on the output value of a thyroid function classification model specialized for hypothyroidism. However, it is not limited to this.

이때, 상기 학습 모델은 regression을 할 수 있는 머신러닝 모델이 사용될수있다. 예를 들어, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등이 사용될 수 있다.At this time, a machine learning model capable of regression may be used as the learning model. For example, the learning model may be light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc.

상기 갑상선 호르몬 농도 예측모델(thyroid hormone concentration prediction model)은 대상일의 TRH 농도 예측모델, TSH 농도 예측모델, T4 농도 예측모델, free T4 농도 예측모델, T3 농도 예측모델 및 free T3 농도 예측모델로 이루어진 그룹에서 선택된 적어도 하나를 포함한다. The thyroid hormone concentration prediction model consists of a TRH concentration prediction model, a TSH concentration prediction model, a T4 concentration prediction model, a free T4 concentration prediction model, a T3 concentration prediction model, and a free T3 concentration prediction model on the target day. Contains at least one selected from the group.

구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, It can be one of the following days: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30.

갑상선 호르몬 농도 예측모델의 종류에 따라서, 상기 라벨링 값은 상기 제1 검사일에 대응되어 있는 복수의 호르몬 농도들 중 상기 종류에 대응되는 농도값이 사용될 수 있다. 예를 들어, free T4 농도 예측모델을 학습시키고자 할 때는, 상기 라벨링값으로 상기 제1 검사일에 대응되어 있는 free T4의 농도값이 사용된다. 다른 예를 들어, TSH 농도 예측모델을 학습시키고자 할 때는, 상기 라벨링값으로 상기 제1 검사일에 대응되어 있는 TSH의 농도값이 사용된다.Depending on the type of thyroid hormone concentration prediction model, the labeling value may be a concentration value corresponding to the type among a plurality of hormone concentrations corresponding to the first test date. For example, when learning a free T4 concentration prediction model, the free T4 concentration value corresponding to the first inspection date is used as the labeling value. For another example, when learning a TSH concentration prediction model, the TSH concentration value corresponding to the first test date is used as the labeling value.

이하에서는, 설명의 편의를 위하여, free T4 농도 예측모델을 학습시키는 것을 기준으로 설명한다.Below, for convenience of explanation, the description is based on learning a free T4 concentration prediction model.

제18 실시예에 따라, free T4 농도 예측 모델을 학습시키기 위하여, 확보된 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 갑상선 기능에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 free T4 농도]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다. 이때, 상기 제2 검사일에 대응되는 갑상선 호르몬 농도는 TRH농도, TSH 농도, T4 농도, free T4 농도, T3 농도 및 free T3 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합일 수 있다.According to the 18th embodiment, in order to learn a free T4 concentration prediction model, the obtained clinical data were used to calculate [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), first inspection day. Learning data sets in the following format can be created: values for thyroid function corresponding to , thyroid hormone concentration corresponding to the second examination date, and free T4 concentration corresponding to the first examination day. At this time, the thyroid hormone concentration corresponding to the second test day may be one selected from the group consisting of TRH concentration, TSH concentration, T4 concentration, free T4 concentration, T3 concentration, and free T3 concentration, or a combination thereof.

예를 들어, 제1 환자로부터 획득한 데이터 세트들은 총 3개이고, 첫번째 검사 시, 확인된 free T4 호르몬 농도는 제1 free T4 농도였고 확인된 TSH의 농도는 제1 TSH 농도이고, 두번째 검사 시, 확인된 free T4 호르몬 농도는 제2 free T4 농도였고 확인된 TSH의 농도는 제2 TSH 농도이며, 세번째 검사 시, 확인된 free T4 호르몬 농도는 제3 free T4 농도였고 확인된 TSH의 농도는 제3 TSH 농도라고 가정한다면, 해당 환자의 임상 데이터로부터 확보될 수 있는 학습 데이터 세트들은 다음과 같을 수 있다.For example, there are a total of three data sets obtained from the first patient, and at the first test, the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration, and at the second test, The confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration. At the third test, the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the third TSH concentration. Assuming the TSH concentration, the learning data sets that can be obtained from the patient's clinical data may be as follows.

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일의 갑상선 기능에 관한 추정값, 제2 free T4 농도, 제2 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), estimated value of thyroid function on the first test day, second free T4 concentration, second TSH concentration, first free T4 concentration] ,

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일의 갑상선 기능에 관한 추정값, 제3 free T4 농도, 제3 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), estimated value of thyroid function on the first test day, third free T4 concentration, third TSH concentration, first free T4 concentration] ,

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일의 갑상선 기능에 관한 추정값, 제1 free T4 농도, 제1 TSH 농도, 제2 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), estimated value of thyroid function on the second test day, first free T4 concentration, first TSH concentration, second free T4 concentration] ,

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일의 갑상선 기능에 관한 추정값, 제3 free T4 농도, 제3 TSH 농도, 제3 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), estimated value of thyroid function on the second test day, third free T4 concentration, third TSH concentration, third free T4 concentration] ,

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일의 갑상선 기능에 관한 추정값, 제1 free T4 농도, 제1 TSH 농도, 제3 free T4 농도], 및[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), estimated value of thyroid function on the third test day, first free T4 concentration, first TSH concentration, third free T4 concentration] , and

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일의 갑상선 기능에 관한 추정값, 제2 free T4 농도, 제2 TSH 농도, 제3 free T4 농도][(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), estimated value of thyroid function on the third test day, second free T4 concentration, second TSH concentration, third free T4 concentration]

다른 예를 들어, 제2 환자로부터 획득한 데이터 세트들은 총 4개이고, 첫번째 검사 시, 확인된 free T4 호르몬 농도는 제1 free T4 농도였고 확인된 TSH의 농도는 제1 TSH 농도이고, 두번째 검사 시, 확인된 free T4 호르몬 농도는 제2 free T4 농도였고 확인된 TSH의 농도는 제2 TSH 농도이며, 세번째 검사 시, 확인된 free T4 호르몬 농도는 제3 free T4 농도였고 확인된 TSH의 농도는 제3 TSH 농도이고, 네번째 검사 시, 확인된 free T4 호르몬 농도는 제4 free T4 농도였고 확인된 TSH의 농도는 제4 TSH 농도라고 가정한다면, 해당 환자의 임상 데이터로부터 확보될 수 있는 학습 데이터 세트들은 다음과 같을 수 있다.As another example, there are a total of 4 data sets obtained from the second patient, and at the first test, the confirmed free T4 hormone concentration was the first free T4 concentration and the confirmed TSH concentration was the first TSH concentration, and at the second test , the confirmed free T4 hormone concentration was the second free T4 concentration and the confirmed TSH concentration was the second TSH concentration. At the third test, the confirmed free T4 hormone concentration was the third free T4 concentration and the confirmed TSH concentration was the second TSH concentration. 3 TSH concentration, and at the time of the fourth test, assuming that the confirmed free T4 hormone concentration was the 4th free T4 concentration and the confirmed TSH concentration was the 4th TSH concentration, the learning data sets that can be obtained from the patient's clinical data are It could be like this:

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일의 갑상선 기능에 관한 추정값, 제2 free T4 농도, 제2 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), estimated value of thyroid function on the first test day, second free T4 concentration, second TSH concentration, first free T4 concentration] ,

[(제1 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제1 검사일의 갑상선 기능에 관한 추정값, 제3 free T4 농도, 제3 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the third test day), estimated value of thyroid function on the first test day, third free T4 concentration, third TSH concentration, first free T4 concentration] ,

[(제1 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제1 검사일의 갑상선 기능에 관한 추정값, 제4 free T4 농도, 제4 TSH 농도, 제1 free T4 농도], [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the fourth test day), estimated value of thyroid function on the first test day, fourth free T4 concentration, fourth TSH concentration, first free T4 concentration] ,

[(제2 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제2 검사일의 갑상선 기능에 관한 추정값, 제1 free T4 농도, 제1 TSH 농도, 제2 free T4 농도], [(section average heart rate corresponding to the second test day - section average heart rate corresponding to the first test day), estimated value of thyroid function on the second test day, first free T4 concentration, first TSH concentration, second free T4 concentration] ,

[(제2 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제2 검사일의 갑상선 기능에 관한 추정값, 제3 free T4 농도, 제3 TSH 농도, 제2 free T4 농도],[(section average heart rate corresponding to the second test day - section average heart rate corresponding to the third test day), estimated value of thyroid function on the second test day, third free T4 concentration, third TSH concentration, second free T4 concentration] ,

[(제2 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제2 검사일의 갑상선 기능에 관한 추정값, 제4 free T4 농도, 제4 TSH 농도, 제2 free T4 농도],[(section average heart rate corresponding to the 2nd test day - section average heart rate corresponding to the 4th test day), estimated value of thyroid function on the 2nd test day, 4th free T4 concentration, 4th TSH concentration, 2nd free T4 concentration] ,

[(제3 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제3 검사일의 갑상선 기능에 관한 추정값, 제1 free T4 농도, 제1 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the first test day), estimated value of thyroid function on the third test day, first free T4 concentration, first TSH concentration, third free T4 concentration] ,

[(제3 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제3 검사일의 갑상선 기능에 관한 추정값, 제2 free T4 농도, 제2 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the third test day - section average heart rate corresponding to the second test day), estimated value of thyroid function on the third test day, second free T4 concentration, second TSH concentration, third free T4 concentration] ,

[(제3 검사일에 대응되는 구간 평균 심박수 - 제4 검사일에 대응되는 구간 평균 심박수), 제3 검사일의 갑상선 기능에 관한 추정값, 제4 free T4 농도, 제4 TSH 농도, 제3 free T4 농도],[(section average heart rate corresponding to the 3rd test day - section average heart rate corresponding to the 4th test day), estimated value of thyroid function on the 3rd test day, 4th free T4 concentration, 4th TSH concentration, 3rd free T4 concentration] ,

[(제4 검사일에 대응되는 구간 평균 심박수 - 제1 검사일에 대응되는 구간 평균 심박수), 제4 검사일의 갑상선 기능에 관한 추정값, 제1 free T4 농도, 제1 TSH 농도, 제4 free T4 농도],[(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 1st test day), estimated value of thyroid function on the 4th test day, 1st free T4 concentration, 1st TSH concentration, 4th free T4 concentration] ,

[(제4 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제4 검사일의 갑상선 기능에 관한 추정값, 제2 free T4 농도, 제2 TSH 농도, 제4 free T4 농도], 및[(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 2nd test day), estimated value of thyroid function on the 4th test day, 2nd free T4 concentration, 2nd TSH concentration, 4th free T4 concentration] , and

[(제4 검사일에 대응되는 구간 평균 심박수 - 제3 검사일에 대응되는 구간 평균 심박수), 제4 검사일의 갑상선 기능에 관한 추정값, 제3 free T4 농도, 제3 TSH 농도, 제4 free T4 농도][(section average heart rate corresponding to the 4th test day - section average heart rate corresponding to the 3rd test day), estimated value of thyroid function on the 4th test day, 3rd free T4 concentration, 3rd TSH concentration, 4th free T4 concentration]

만약, free T4의 농도를 예측하기 위한 모델이 아니라, TSH의 농도를 예측하기 위한 모델을 학습시키고자 하는 경우, 전술한 학습 데이터 세트들에서 입력값은 그대로 유지하고, 라벨링값만 free T4의 농도를 사용하는 대신 TSH의 농도를 사용하면 된다.If you want to train a model for predicting the concentration of TSH rather than a model for predicting the concentration of free T4, keep the input values in the above-mentioned training data sets as is and only the labeling value is the concentration of free T4. Instead of using , you can use the concentration of TSH.

제18 실시예를 개시함에 있어서, 설명의 편의를 위해,In disclosing the 18th embodiment, for convenience of explanation,

[(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 추정 갑상선 기능에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도]를 학습 데이터의 입력으로 사용하면서, 그에 대한 라벨링값으로 [제1 검사일에 대응되는 free T4 농도]를 사용하는 경우를 상정하여 설명하였으나,[(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), estimated thyroid function value corresponding to the first inspection day, thyroid hormone concentration corresponding to the second inspection day] of the learning data. When using it as an input, the explanation was made assuming that [free T4 concentration corresponding to the first test date] was used as the labeling value.

제18 실시예에서의 개시하는 핵심적인 사상은 "제1 검사일에 대응되는 갑상선 기능에 관한 값 "을 입력으로 이용한다는 점이며, 학습모델에 입력되는 그 외의 정보는 추가되거나 생략될 수 있다. 예를 들어, 제13 내지 제17 실시예에서 개시된 입력 데이터에 제1 검사일에 대응되는 추정 갑상선 기능에 관한 값을 입력으로 더 이용하는 형태로 학습 모델이 구현될 수 있다.The core idea disclosed in the 18th embodiment is that “values related to thyroid function corresponding to the first test date” are used as input, and other information input to the learning model can be added or omitted. For example, the learning model may be implemented in a form that further uses a value related to estimated thyroid function corresponding to the first examination date as input to the input data disclosed in the 13th to 17th embodiments.

실시예 14를 빗대어 설명하면, 학습모델을 학습시키기 위하여, 항진증 환자들의 임상 데이터들을 이용하여 [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / (제2 검사일에 대응되는 구간 평균 심박수)), 제1 검사일에 대응되는 추정 갑상선 기능에 관한 값, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 free T4농도]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다.To explain Example 14, in order to learn a learning model, clinical data of patients with hyperactivity were used [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / (second test day) Training data set in the following format: interval average heart rate corresponding to the test day)), estimated thyroid function value corresponding to the first test day, thyroid hormone concentration corresponding to the second test day, free T4 concentration corresponding to the first test day] can be created.

학습방법 실시예#19. 기준일의 호르몬값, 변수들의 조합Learning method example #19. Hormone values on the reference date, combination of variables

본 출원에 의해 개시되는 제19 실시예에 따르면, 갑상선 호르몬 농도 예측모델(thyroid hormone concentration prediction model)을 학습시키기 위하여, 제2 검사일의 호르몬 농도와 다음의 [표 17] 및 [표 18]에 개시된 변수들 중 적어도 하나를 입력값으로 사용하고, 제1 검사일에 대응되어 있는 호르몬 농도를 라벨링값으로 사용할 수 있다.According to the 19th example disclosed by the present application, in order to learn a thyroid hormone concentration prediction model, the hormone concentration on the second test day and the hormone concentration disclosed in the following [Table 17] and [Table 18] At least one of the variables can be used as an input value, and the hormone concentration corresponding to the first test date can be used as a labeling value.

본 출원에 의해 개시되는 제19 실시예에 따르면, 갑상선 호르몬 농도 예측모델(thyroid hormone concentration prediction model)을 학습시키기 위하여, 제2 검사일의 호르몬 농도와 다음의 [표 17]에 개시된 변수 중 적어도 하나가 입력값으로 사용되고, 제1 검사일에 대응되어 있는 호르몬 농도가 라벨링값으로 사용될 수 있다. 또는, 갑상선 호르몬 농도 예측모델을 학습시키기 위하여, 제2 검사일의 호르몬 농도와 다음의 [표 18]에 개시된 변수들의 조합값(combination value) 중 적어도 하나가 입력값으로 사용되고, 제1 검사일에 대응되어 있는 호르몬 농도가 라벨링값으로 사용될 수 있다. 또는, 갑상선 호르몬 농도 예측모델을 학습시키기 위하여, 제2 검사일의 호르몬 농도, 다음의 [표 17]에 개시된 변수 중 적어도 하나 및 다음의 [표 18]에 개시된 변수들의 조합값 중 적어도 하나가 입력값으로 사용되고, 제1 검사일에 대응되어 있는 호르몬 농도가 라벨링값으로 사용될 수 있다.According to the 19th embodiment disclosed by the present application, in order to learn a thyroid hormone concentration prediction model, the hormone concentration on the second test day and at least one of the variables disclosed in the following [Table 17] It is used as an input value, and the hormone concentration corresponding to the first test date can be used as a labeling value. Alternatively, in order to learn a thyroid hormone concentration prediction model, at least one of the hormone concentration on the second test day and the combination value of the variables disclosed in the following [Table 18] is used as an input value, and corresponds to the first test day. The hormone concentration present can be used as a labeling value. Alternatively, in order to learn a thyroid hormone concentration prediction model, the hormone concentration on the second test day, at least one of the variables disclosed in the following [Table 17], and at least one of the combination values of the variables disclosed in the following [Table 18] are input values. , and the hormone concentration corresponding to the first test date can be used as the labeling value.

이때, 변수들의 조합값은 변수들 중에서 선택되는 하나의 값(이하, 1 값이라 함)과 상기 상기 변수들 중에서 선택되는 다른 하나의 값(이하, 제2 값이라 함)을 서로 곱한 값, 제1 값을 제2 값으로 나눈 값 및 제2 값을 제1 값으로 나눈 값 중 하나를 의미한다.At this time, the combination value of the variables is the product of one value selected from among the variables (hereinafter referred to as the 1 value) and another value selected from the variables (hereinafter referred to as the second value). It means one of the value obtained by dividing the 1 value by the second value and the value divided by the second value by the first value.

No.No. 입력 변수input variable 1One 구간 평균 심박수 변화량Average heart rate change for each section 22 구간 평균 심박수 변화율Zone average heart rate change rate 33 구간 조화평균 심박수 변화량Section harmonic average heart rate change 44 구간 조화평균 심박수 변화율Section harmonic average heart rate change rate 55 구간 심박수의 첨도 변화량Change in kurtosis of interval heart rate 66 구간 심박수의 왜도 변화량Skewness change in zone heart rate 77 구간 심박수의 상대표준편차 변화량Relative standard deviation change in zone heart rate 88 구간 심박수의 JS DivergenceJS Divergence in Zone Heart Rate 99 제2 검사일의 구간 조화평균 심박수Interval harmonic average heart rate on the second test day 1010 제2 검사일의 구간 심박수의 표준편차Standard deviation of zone heart rate on test day 2 1111 제2 검사일의 구간 심박수의 상대표준편차Relative standard deviation of zone heart rate on the second test day 1212 제2 검사일의 구간 심박수의 왜도Skewness of zone heart rate on the second test day 1313 제2 검사일의 구간 깊은 수면 비율Interval deep sleep percentage on the second test day 1414 제2 검사일의 구간 REM 수면 비율Interval REM sleep rate on the second test day 1515 제2 검사일의 구간 얕은 수면 비율Percentage of light sleep in the interval on the second test day 1616 제1 검사일의 구간 평균 심박수Zone average heart rate on test day 1 1717 제1 검사일의 구간 심박수의 첨도Kurtosis of interval heart rate on first test day 1818 제1 검사일의 구간 심박수의 왜도Skewness of zone heart rate on first test day 1919 제1 검사일의 구간 얕은 수면 비율Percentage of light sleep in the interval on the first test day 2020 제1 검사일의 구간 REM 수면 비율Interval REM sleep rate on first test day 2121 성별gender

No.No. 입력 변수input variable 1One 구간 평균 심박수 변화량Average heart rate change for each section 22 구간 평균 심박수 변화율Zone average heart rate change rate 33 구간 조화평균 심박수 변화량Section harmonic average heart rate change 44 구간 조화평균 심박수 변화율Section harmonic average heart rate change rate 55 구간 심박수의 첨도 변화량Change in kurtosis of interval heart rate 66 구간 심박수의 왜도 변화량Skewness change in zone heart rate 77 구간 심박수의 상대표준편차 변화량Relative standard deviation change in zone heart rate 88 구간 심박수의 JS DivergenceJS Divergence in Zone Heart Rate 99 제1 검사일의 구간 평균 심박수Zone average heart rate on test day 1 1010 제1 검사일의 구간 심박수의 왜도Skewness of zone heart rate on first test day 1111 제1 검사일의 구간 심박수의 첨도Kurtosis of interval heart rate on first test day 1212 제1 검사일의 구간 얕은 수면 비율Percentage of light sleep in the interval on the first test day 1313 제1 검사일의 REM 수면 비율REM sleep percentage on first test day 1414 제2 검사일의 free T4 농도Free T4 concentration on the second test day 1515 제2 검사일의 TSH 농도TSH concentration on second test day 1616 제2 검사일의 구간 조화평균 심박수Interval harmonic average heart rate on the second test day 1717 제2 검사일의 구간 심박수의 표준편차Standard deviation of zone heart rate on test day 2 1818 제2 검사일의 구간 심박수의 상대표준편차Relative standard deviation of zone heart rate on the second test day 1919 제2 검사일의 구간 심박수의 왜도Skewness of zone heart rate on the second test day 2020 제2 검사일의 구간 깊은 수면 비율Interval deep sleep percentage on the second test day 2121 제2 검사일의 구간 얕은 수면 비율Percentage of interval light sleep on the second test day 2222 제2 검사일의 구간 REM 수면 비율Interval REM sleep rate on the second test day

구간 평균 심박수 변화량, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화량, 구간 조화평균 심박수 변화율, 구간 심박수의 첨도 변화량, 구간 심박수의 왜도 변화량, 구간 심박수의 상대표준편차 변화량 및 구간 심박수의 JS Divergence의 산출 방법은 실시예#10에서 상술한 바 있으므로, 중복 설명은 생략한다.Calculation of section average heart rate change, section average heart rate change, section harmonic average heart rate change, section harmonic average heart rate change, section kurtosis change of section heart rate, skewness change of section heart rate, relative standard deviation change of section heart rate, and JS Divergence of section heart rate. Since the method has been described above in Example #10, redundant description will be omitted.

상기 갑상선 호르몬 농도 예측모델은 회귀분석(regression)을 할 수 있는 머신러닝 모델이 사용될 수 있다. 예를 들어, 상기 학습 모델은 light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost 등이 사용될 수 있다.The thyroid hormone concentration prediction model may be a machine learning model capable of regression analysis. For example, the learning model may be light gradient boosting machine, support vector machine, random forest, extra trees, ada boost, extreme gradient boosting, CatBoost, etc.

상기 갑상선 호르몬 농도 예측모델은 대상일의 free T4 농도를 예측하는 모델을 포함할 수 있다. 이 경우, 상기 라벨링값으로 상기 제1 검사일에 대응되어 있는 복수의 호르몬 농도들 중 free T4의 농도값이 사용될 수 있다.The thyroid hormone concentration prediction model may include a model that predicts the free T4 concentration on the target day. In this case, the concentration value of free T4 among the plurality of hormone concentrations corresponding to the first test date may be used as the labeling value.

구간 평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9, It can be one of the following days: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30.

구간 조화평균 심박수를 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 휴지기 심박수들을 활용할 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section harmonic average heart rate, resting heart rates during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4, 5, 6, 7, 8, 9. , 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, and 30 days.

구간 깊은 수면 비율을 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 총 수면 시간에서 총 깊은 수면 시간이 차지하는 비율이 활용될 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section deep sleep ratio, the ratio of the total deep sleep time to the total sleep time during a predetermined period based on a specific day may be used, where the predetermined period is 1, 2, 3, 4. , 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30 days.

구간 얕은 수면 비율을 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 총 수면 시간에서 총 얕은 수면 시간이 차지하는 비율이 활용될 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the interval light sleep ratio, the ratio of the total light sleep time to the total sleep time during a predetermined period based on a specific day may be used, where the predetermined period is 1, 2, 3, 4. , 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30 days.

구간 REM 수면 비율을 산출함에 있어서, 특정일을 기준으로 미리 정해진 기간 동안의 총 수면 시간에서 총 REM 수면 시간이 차지하는 비율이 활용될 수 있으며, 이 때 상기 미리 정해진 기간은 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 및 30일 중 하나일 수 있다.In calculating the section REM sleep ratio, the ratio of the total REM sleep time to the total sleep time during a predetermined period based on a specific day can be used, where the predetermined period is 1, 2, 3, 4. , 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29 and 30 days.

제19 실시예에 따라, 갑상선 호르몬 농도 예측모델을 학습시키기 위하여, 확보된 임상 데이터들을 이용하여 [제2 검사일에 대응되는 갑상선 호르몬 농도, 전술한 입력 변수 중 적어도 하나, 제1 검사일에 대응되는 free T4 농도]와 같은 포맷의 학습 데이터 세트들이 생성될 수 있다. 이때, 상기 제2 검사일에 대응되는 갑상선 호르몬 농도는 TSH 농도 및 free T4 농도일 수 있다.According to the 19th embodiment, in order to learn a thyroid hormone concentration prediction model, the obtained clinical data are used [thyroid hormone concentration corresponding to the second examination date, at least one of the above-mentioned input variables, and free corresponding to the first examination day. Learning data sets in a format such as [T4 concentration] can be created. At this time, the thyroid hormone concentration corresponding to the second test date may be the TSH concentration and the free T4 concentration.

예를 들어, 제19 실시예에 따라 갑상선 호르몬 농도 예측모델을 학습시킬 때, 입력값은 다음과 같을 수 있다.For example, when learning a thyroid hormone concentration prediction model according to the 19th embodiment, the input values may be as follows.

(1) 제1 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량](1) First set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change]

(2) 제2 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율](2) Second set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate]

(3) 제3 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화량](3) Third set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic average heart rate change]

(4) 제4 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화율](4) Fourth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic mean heart rate change rate]

(5) 제5 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율](5) Fifth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day]

(6) 제6 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 조화평균 심박수 변화율](6) Sixth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval harmonic average heart rate change rate]

(7) 제7 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 얕은 수면 비율](7) Seventh set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the first test day]

(8) 제8 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 평균 심박수 변화율](8) Eighth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval average heart rate change rate]

(9) 제9 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 심박수의 첨도](9) Ninth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, kurtosis of interval heart rate on the first test day]

(10) 제10 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도](10) Tenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. kurtosis]

(11) 제11 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 REM 수면 비율](11) Eleventh set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval REM sleep of the second test day ratio]

(12) 제12 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 구간 조화평균 심박수 변화율](12) Twelfth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio on the second test day, interval harmonic average heart rate change rate]

(13) 제13 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 심박수의 상대표준편차](13) Thirteenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the second test day relative standard deviation]

(14) 제14 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 심박수의 표준편차](14) Fourteenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the second test day Standard Deviation]

(15) 제15 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 구간 평균 심박수 변화율](15) Fifteenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio on the second test day, interval average heart rate change rate]

(16) 제15 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 얕은 수면 비율](16) Fifteenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval light sleep of the first test day ratio]

(17) 제17 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차](17) Seventeenth set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of interval heart rate on the second test day]

(18) 제18 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 구간 조화평균 심박수 변화율](18) 18th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, interval harmonic average heart rate change rate]

(19) 제19 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 표준편차](19) 19th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, standard deviation of interval heart rate on the second test day]

(20) 제20 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 구간 평균 심박수 변화율](20) The 20th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, interval average heart rate change rate]

(21) 제21 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율](21) 21st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate]

(22) 제22 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화율](22) 22nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of interval heart rate on the second test day, interval harmonic mean heart rate change rate]

(23) 제23 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화량](23) The 23rd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of interval heart rate on the second test day, interval harmonic average heart rate change]

(24) 제24 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율](24) The 24th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate]

(25) 제25 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 깊은 수면 비율](25) The 25th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, interval deep sleep ratio on the second test day]

(26) 제26 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화량](26) 26th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change]

(27) 제27 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 심박수의 왜도](27) The 27th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, skewness of the interval heart rate on the second test day]

(28) 제28 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 대상의 성별](28) 28th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, gender of subject]

(29) 제29 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 심박수의 JS Divergence](29) The 29th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, JS Divergence of interval heart rate]

(30) 제30 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 REM 수면 비율](30) The 30th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, interval REM sleep ratio on the second test day]

(31) 제31 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 심박수의 첨도의 변화량](31) 31st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, change in kurtosis of the interval heart rate]

(32) 제32 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 심박수의 표준편차](32) A 32nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, standard deviation of the interval heart rate on the second test day]

(33) 제33 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도)](33) 33rd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day)]

(34) 제34 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 / 제2 검사일의 구간 얕은 수면 비율)](34) 34th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, interval harmonic mean heart rate change rate, (free T4 concentration on the second test day / interval light sleep ratio on the second test day)]

(35) 제35 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제2 검사일의 구간 심박수의 상대표준편차)](35) 35th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x relative standard deviation of section heart rate on the second test day)]

(36) 제36 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도)](36) 36th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (section average heart rate change / free T4 concentration on the second test day)]

(37) 제37 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도)](37) 37th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of interval heart rate on the second test day, interval average heart rate change rate, interval harmonic mean heart rate change rate, (interval light sleep ratio on the second test day / free T4 concentration on the second test day)]

(38) 제38 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차)](38) The 38th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (zone light sleep ratio on the second test day x relative standard deviation of section heart rate on the second test day)]

(39) 제39 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화량 / 제2 검사일의 구간 심박수의 상대표준편차)](39) The 39th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of section heart rate on the first test day), (section average heart rate change / second 2 Relative standard deviation of zone heart rate on the test day)]

(40) 제40 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 조화평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차)](40) The 40th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day. Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (interval harmonic average heart rate change rate / Relative standard deviation of zone heart rate on the second test day)]

(41) 제41 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도)](41) 41st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, interval harmonic mean heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (interval shallow on the second test day sleep ratio/free T4 concentration on the second test day)]

(42) 제42 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도)](42) The 42nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of section heart rate on the first test day), (section average heart rate change / second 2 Free T4 concentration on test day)]

(43) 제43 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차)](43) The 43rd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (interval average heart rate change rate / second 2 Relative standard deviation of zone heart rate on the test day)]

(44) 제44 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차)](44) The 44th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval heart rate of the first test day Kurtosis, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, interval harmonic mean heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (interval shallow on the second test day sleep ratio x relative standard deviation of zone heart rate on the second test day)]

(45) 제45 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율](45) The 45th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day]

(46) 제46 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제2 검사일의 구간 조화평균 심박수](46) 46th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate of the second test day]

(47) 제47 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화량](47) 47th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate change]

(48) 제48 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율](48) The 48th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate change rate]

(49) 제49 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별](49) The 49th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject]

(50) 제50 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 구간 심박수의 상대표준편차 변화량](50) The 50th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, relative standard deviation change in interval heart rate ]

(51) 제51 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 깊은 수면 비율](51) 51st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, interval deep sleep of the second test day ratio]

(52) 제52 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 심박수의 표준편차](52) The 52nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, interval heart rate of the second test day Standard Deviation]

(53) 제53 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 구간 심박수의 왜도 변화량](53) The 53rd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, skewness change in interval heart rate]

(54) 제54 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제1 검사일의 구간 평균 심박수](54) The 54th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, interval average heart rate of the first test day ]

(55) 제55 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량](55) The 55th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change]

(56) 제56 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 제2 검사일의 구간 깊은 수면 비율](56) The 56th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, second test day [Interval deep sleep ratio]

(57) 제57 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율](57) The 57th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, section deep sleep ratio on the second test day]

(58) 제58 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 심박수의 왜도](58) The 58th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, skewness of heart rate in the section on the second test day]

(59) 제59 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차](59) The 59th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of interval heart rate on the second test day]

(60) 제60 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 구간 조화평균 심박수의 변화율](60) The 60th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, section deep sleep rate on the second test day, change rate of section harmonic average heart rate]

(61) 제61 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 얕은 수면 비율](61) The 61st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, section deep sleep rate on the second test day, section light sleep rate on the second test day]

(62) 제62 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도](62) The 62nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day]

(63) 제63 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제2 검사일의 구간 얕은 수면 비율](63) The 63rd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, section deep sleep rate on the second test day, standard deviation of section heart rate on the second test day, section light sleep rate on the second test day]

(64) 제64 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 제2 검사일의 구간 깊은 수면 비율)](64) The 64th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x interval deep sleep ratio on the second test day )]

(65) 제65 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량)](65) The 65th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate)]

(66) 제66 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 TSH 농도 x 제2 검사일의 구간 깊은 수면 비율)](66) The 66th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (TSH concentration on the second test day x interval deep sleep ratio on the second test day) ]

(67) 제67 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도)](67) The 67th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (Interval average heart rate change rate / free T4 concentration on the second test day)]

(68) 제68 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제2 검사일의 free T4 농도 / 제2 검사일의 TSH 농도)](68) The 68th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (free T4 concentration on the second test day / TSH concentration on the second test day)]

(69) 제69 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 심박수의 왜도의 변화량 x 제2 검사일의 구간 깊은 수면 비율)](69) The 69th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (change in skewness of segment heart rate x segment deep sleep ratio on the second test day)]

(70) 제70 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화율 / 제1 검사일의 구간 심박수의 왜도)](70) The 70th set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (section average heart rate change rate / free T4 concentration on the second test day), (section average heart rate change rate / skewness of section heart rate on the first test day)]

(71) 제71 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 깊은 수면 비율 / 제1 검사일의 구간 심박수의 왜도)](71) The 71st set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average heart rate change rate / free T4 concentration on the second test day), (interval deep sleep ratio on the second test day / skewness of interval heart rate on the first test day)]

(72) 제72 입력 변수 세트: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제1 검사일의 구간 REM 수면 비율 / 제1 검사일의 구간 심박수의 왜도)](72) The 72nd set of input variables: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio on the first test day, gender of the subject, interval heart rate Skewness change, interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average heart rate change rate / free T4 concentration on the second test day), (interval REM sleep ratio on the first test day / skewness of interval heart rate on the first test day)]

한편, 제19 실시예에 따라 갑상선 호르몬 농도 예측모델을 학습시킬 때, 라벨링값은 제1 검사일의 free T4 농도일 수 있다.Meanwhile, when learning the thyroid hormone concentration prediction model according to the 19th embodiment, the labeling value may be the free T4 concentration on the first test day.

한편, 전술한 제13 내지 제19 실시예에 있어서, 갑상선 호르몬 농도 예측모델을 학습시킴에 있어서 유의미한 정보들을 개시한 것이며, 만약, 변화량(차이값) 또는 변화율이 유의미한 정보로 특정된 경우 갑상선 호르몬 농도 예측 모델을 학습시키는 데이터로 변화량(차이값) 또는 변화율을 대신하여 제1 검사일에 대응되는 값과 제2 검사일에 대응되는 값을 각각 활용하는 것은 당연히 가능하다.Meanwhile, in the above-described 13th to 19th examples, meaningful information is disclosed in learning the thyroid hormone concentration prediction model, and if the change amount (difference value) or change rate is specified as meaningful information, the thyroid hormone concentration It is naturally possible to use the value corresponding to the first inspection date and the value corresponding to the second inspection date instead of the change amount (difference value) or change rate as data for training the prediction model.

7. 갑상선 호르몬 예측모델에 기초한 갑상선기능이상증의 예측방법7. Prediction method for thyroid dysfunction based on thyroid hormone prediction model

본 출원에 의해 개시되는 갑상선기능이상증(thyroid dysfunction)을 예측하는 방법은 전술한 시스템(1)에 의해 수행될 수 있다.The method for predicting thyroid dysfunction disclosed by the present application can be performed by the above-described system 1.

도 6은 본 출원에 의해 개시되는 갑상선기능이상증을 예측하는 방법을 설명하기 위한 흐름도이다.Figure 6 is a flowchart for explaining the method for predicting thyroid dysfunction disclosed by the present application.

도 6을 참고하면, 본 출원에 의해 개시되는 대상(subject)에 대한 갑상선기능이상증을 예측하는 방법은 트리거 신호를 획득함(S200), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S210), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S220), 기준일에 대응되는 호르몬 농도를 획득함(S230), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S240), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S250), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선 호르몬 농도 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선 호르몬 농도에 대한 예측 결과를 획득함(S260), 상기 호르몬 농도에 대한 예측 결과에 기초하여 상기 대상의 갑상선기능이상증에 대한 결과를 획득함(S270)을 포함한다.Referring to FIG. 6, the method for predicting thyroid dysfunction for a subject disclosed by the present application acquires a trigger signal (S200), and obtains heart rates in the section corresponding to the target date determined by the trigger signal. Obtaining (S210), preprocessing the section heart rates corresponding to the target date (S220), obtaining the hormone concentration corresponding to the reference date (S230), obtaining preprocessing results for the section heart rates corresponding to the reference date (S240), obtain the difference between the preprocessing results of the section heart rates corresponding to the target date and the preprocessing result of the section heart rates corresponding to the reference date (S250), calculate the obtained difference and the hormone concentration corresponding to the reference date Processing using a thyroid hormone concentration prediction model and obtaining a prediction result for the thyroid hormone concentration for the subject (S260), obtaining a result for thyroid dysfunction of the subject based on the prediction result for the hormone concentration Includes (S270).

이때, 트리거 신호를 획득함(S200), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S210), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S220), 기준일에 대응되는 호르몬 농도를 획득함(S230), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S240), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S250)은 각각 도 5를 참조하여 설명한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150)과 각각 동일하거나 매우 유사하기 때문에 여기서는 자세한 설명을 생략한다.At this time, a trigger signal is acquired (S200), section heart rates corresponding to the target date determined by the trigger signal are acquired (S210), the section heart rates corresponding to the target date are preprocessed (S220), and on the reference date Obtaining the corresponding hormone concentration (S230), obtaining preprocessing results for section heart rates corresponding to the reference date (S240), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Obtaining the difference between the preprocessing results (S250), obtaining the trigger signal described with reference to FIG. 5 (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), Preprocess the section heart rates corresponding to the target date (S120), obtain the hormone concentration corresponding to the reference date (S130), obtain preprocessing results for the section heart rates corresponding to the reference date (S140), and the target Since the difference between the preprocessing results of the section heart rates corresponding to the day and the preprocessing result of the section heart rates corresponding to the reference day are obtained (S150) are the same or very similar, a detailed description is omitted here.

일 예로, 학습 방법 실시예 # 18에 따르면, 대상일에 대응되는 갑상선 기능에 관한 값이 획득될 수 있다. 일 예로, 갑상선 기능 예측모델의 출력값 또는 갑상선 기능 예측모델의 출력값을 기초로 획득된 갑상선 기능에 관한 값이 획득될 수 있다.As an example, according to learning method example #18, a value related to thyroid function corresponding to the target date can be obtained. For example, an output value of a thyroid function prediction model or a value related to thyroid function obtained based on an output value of a thyroid function prediction model may be obtained.

갑상선 호르몬 농도 예측모델을 통한 갑상선 호르몬 농도에 대한 예측결과 획득(S260)Obtain prediction results for thyroid hormone concentration through thyroid hormone concentration prediction model (S260)

제1 갑상선 호르몬 농도 예측모델을 통한 제1 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for first thyroid hormone concentration through first thyroid hormone concentration prediction model

제1 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제1 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제1 갑상선 호르몬은 TRH일 수 있다.The first thyroid hormone concentration for the subject may be obtained through the first thyroid hormone concentration prediction model. For example, the first thyroid hormone may be TRH.

제2 갑상선 호르몬 농도 예측모델을 통한 제2 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for secondary thyroid hormone concentration through secondary thyroid hormone concentration prediction model

제2 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제2 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제2 갑상선 호르몬은 TSH일 수 있다.The second thyroid hormone concentration for the subject can be obtained through the second thyroid hormone concentration prediction model. For example, the second thyroid hormone may be TSH.

제3 갑상선 호르몬 농도 예측모델을 통한 제3 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for third thyroid hormone concentration through third thyroid hormone concentration prediction model

제3 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제3 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제3 갑상선 호르몬은 T4일 수 있다.The third thyroid hormone concentration for the subject can be obtained through the third thyroid hormone concentration prediction model. For example, the third thyroid hormone may be T4.

제4 갑상선 호르몬 농도 예측모델을 통한 제4 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for 4th thyroid hormone concentration through 4th thyroid hormone concentration prediction model

제4 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제4 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제4 갑상선 호르몬은 free T4일 수 있다.The fourth thyroid hormone concentration for the subject can be obtained through the fourth thyroid hormone concentration prediction model. For example, the fourth thyroid hormone may be free T4.

제5 갑상선 호르몬 농도 예측모델을 통한 제5 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for fifth thyroid hormone concentration through the fifth thyroid hormone concentration prediction model

제5 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제5 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제5 갑상선 호르몬은 T3일 수 있다.The fifth thyroid hormone concentration for the subject can be obtained through the fifth thyroid hormone concentration prediction model. For example, the fifth thyroid hormone may be T3.

제6 갑상선 호르몬 농도 예측모델을 통한 제6 갑상선 호르몬 농도에 대한 예측결과 획득Obtaining prediction results for 6th thyroid hormone concentration through 6th thyroid hormone concentration prediction model

제6 갑상선 호르몬 농도 예측모델을 통해 상기 대상에 대한 제6 갑상선 호르몬 농도가 획득될 수 있다. 예를 들어, 상기 제6 갑상선 호르몬은 free T3일 수 있다.The sixth thyroid hormone concentration for the subject can be obtained through the sixth thyroid hormone concentration prediction model. For example, the sixth thyroid hormone may be free T3.

전술한 바와 같이, 상기 제1 갑상선 호르몬 농도 내지 상기 제6 갑상선 호르몬 농도로 이루어진 그룹에서 선택된 하나 또는 이들의 조합이 획득될 수 있다.As described above, one or a combination thereof selected from the group consisting of the first to sixth thyroid hormone concentrations may be obtained.

상기 호르몬 농도에 대한 예측 결과에 기초하여 상기 대상의 갑상선기능이상증에 대한 결과를 획득함(S270)Obtain results for thyroid dysfunction of the subject based on the predicted results for the hormone concentration (S270)

상기 획득된 호르몬 농도에 대한 예측값들에 기초하여, 대상이 갑상선기능항진증 상태인지, 갑상선기능저하증 상태인지 혹은 정상 상태인지가 판단될 수 있다.Based on the obtained predicted values for hormone concentration, it can be determined whether the subject is in a hyperthyroid state, hypothyroidism state, or normal state.

각 갑상선 호르몬 농도들에 대해서 '정상', '항진증' 및 '저하증'에 대응되는 범위들이 미리 결정되어 있기 때문에, 이러한 미리 결정된 범위들이 사용될 수 있을 것이다.Since the ranges corresponding to 'normal', 'hyperactivity' and 'hyporegulation' are predetermined for each thyroid hormone concentration, these predetermined ranges may be used.

전술한 트리거 신호를 획득함(S200), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S210), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S220), 기준일에 대응되는 호르몬 농도를 획득함(S230), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S240), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S250), 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선 호르몬 농도 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선 호르몬 농도에 대한 예측 결과를 획득함(S260), 상기 호르몬 농도에 대한 예측 결과에 기초하여 상기 대상의 갑상선기능이상증에 대한 결과를 획득함(S270)은 상기 심박수 측정 디바이스(10), 상기 사용자 단말기(20) 및 상기 서버(30)에 의해 적절히 분산되어 수행될 수 있다.Obtaining the above-described trigger signal (S200), obtaining section heart rates corresponding to the target date determined by the trigger signal (S210), preprocessing the section heart rates corresponding to the target date (S220), on the reference date Obtaining the corresponding hormone concentration (S230), obtaining preprocessing results for section heart rates corresponding to the reference date (S240), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Acquire the difference between the preprocessing results (S250), process the obtained difference and the hormone concentration corresponding to the reference date using a thyroid hormone concentration prediction model, and obtain a prediction result for the thyroid hormone concentration for the subject. (S260), obtaining a result for thyroid dysfunction of the subject based on the prediction result for the hormone concentration (S270), the heart rate measurement device 10, the user terminal 20, and the server 30 It can be carried out appropriately distributed.

8. 실험예8. Experimental example

실험예#13.Experimental Example #13.

학습 데이터 세트들의 준비Preparation of training data sets

총 171명의 환자를 대상으로 수집한 총 563개의 임상 데이터 세트들이 사용되었다.A total of 563 clinical data sets collected from a total of 171 patients were used.

전술한 학습방법 실시예#13에 기재한 갑상선 호르몬 농도 예측 모델의 학습 방법에 따라 모델이 학습되었다. 특히, free T4의 농도를 예측하기 위한 예측모델이 학습되었다.The model was learned according to the learning method of the thyroid hormone concentration prediction model described in Learning Method Example #13 described above. In particular, a prediction model was learned to predict the concentration of free T4.

한편, 각 검사일에 대응되는 호르몬 농도는 free T4의 농도와 TSH의 농도가 사용되었으며, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다.Meanwhile, the concentration of free T4 and TSH were used as the hormone concentration corresponding to each test day, and the average heart rate of the section corresponding to the test day was obtained by obtaining resting heart rates for 10 days based on the test date, and the obtained resting heart rate The average was calculated.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 171명의 환자로부터 확보한 563개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,542개였다.At this time, the clinical data used for learning was 563 data obtained from 171 patients, and the total number of learning data sets created from this was 1,542.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

extra trees을 이용하여, 전술한 학습 데이터 세트들을 이용하여 free T4의농도 예측모델이 학습되었다.Using extra trees, a free T4 concentration prediction model was learned using the aforementioned training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 126명의 환자로부터 확보된 464개의 임상 데이터들로부터 생성되었으며, 총 1,416개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 464 clinical data obtained from a total of 126 patients, and a total of 1,416 test data sets were created.

실험예#13에 따라 학습된 free T4 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.Testing the accuracy of the free T4 concentration prediction model learned according to Experimental Example #13 was conducted using the test sets described above.

실험예#13에 따르는 free T4 농도 예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표19]와 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the free T4 concentration prediction model according to Experimental Example #13 are shown in [Table 19] below.

MAEM.A.E. MSEMSE MAPEMAPE free T4free T4 0.22720.2272 0.12810.1281 0.16240.1624

실험예#14.실험예#14에서 사용된 학습 데이터 세트들과 테스트 데이터 세트들 및 그 생성방법은 실험예#13에서 설명한 바와 같다. Experimental Example #14. The learning data sets and test data sets used in Experimental Example #14 and their generation method are the same as described in Experimental Example #13.

다만, 실험예#14은, 실험예#13와 달리, 전술한 학습방법 실시예#13에 따라 예측모델을 학습시킨 대신, 학습방법 실시예#14에 따라 예측모델이 학습되었다.However, in Experimental Example #14, unlike Experimental Example #13, instead of learning the prediction model according to the above-described learning method example #13, the prediction model was learned according to learning method example #14.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

실험예#14에 따라 학습된 free T4 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the free T4 concentration prediction model learned according to Experimental Example #14 was conducted using the test sets described above.

실험예#14에 따르는 free T4 농도 예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 20]과 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the free T4 concentration prediction model according to Experimental Example #14 are shown in [Table 20] below.

MAEM.A.E. MSEMSE MAPEMAPE free T4free T4 0.22190.2219 0.12030.1203 0.16260.1626

실험예#15.실험예#15에서 사용된 학습 데이터 세트들과 테스트 데이터 세트들 및 그 생성방법은 실험예#13에서 설명한 바와 같다. Experimental Example #15. The learning data sets and test data sets used in Experimental Example #15 and their generation method are the same as described in Experimental Example #13.

다만, 실험예#15는, 실험예#13와 달리, 전술한 학습방법 실시예#13에 따라 예측모델을 학습시킨 대신, 학습방법 실시예#15에 따라 예측모델이 학습되었다.However, in Experimental Example #15, unlike Experimental Example #13, instead of learning the prediction model according to the above-described learning method example #13, the prediction model was learned according to learning method example #15.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

실험예#15에 따라 학습된 free T4 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the free T4 concentration prediction model learned according to Experimental Example #15 was conducted using the test sets described above.

실험예#15에 따르는 free T4 농도 예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 21]과 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the free T4 concentration prediction model according to Experimental Example #15 are shown in [Table 21] below.

MAEM.A.E. MSEMSE MAPEMAPE free T4free T4 0.21770.2177 0.11040.1104 0.15960.1596

실험예#16.Experimental Example #16.

실험예#16에서 사용된 학습 데이터 세트들과 테스트 데이터 세트들 및 그 생성방법은 실험예#13에서 설명한 바와 같다.The learning data sets and test data sets used in Experimental Example #16 and their generation method are the same as described in Experimental Example #13.

다만, 실험예#16는, 실험예#13과 달리, 전술한 학습방법 실시예#13에 따라 예측모델을 학습시킨 대신, 학습방법 실시예#16에 따라 예측모델이 학습되었다.However, in Experimental Example #16, unlike Experimental Example #13, instead of learning the prediction model according to the above-described learning method example #13, the prediction model was learned according to learning method example #16.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

실험예#16에 따라 학습된 free T4 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the free T4 concentration prediction model learned according to Experimental Example #16 was conducted using the test sets described above.

실험예#16에 따르는 free T4 농도 예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 22]와 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the free T4 concentration prediction model according to Experimental Example #16 are shown in [Table 22] below.

MAEM.A.E. MSEMSE MAPEMAPE free T4free T4 0.21590.2159 0.10790.1079 0.15840.1584

실험예#17.Experimental example #17.

학습 데이터 세트들의 준비Preparation of training data sets

총 146명의 환자를 대상으로 수집한 총 542개의 임상 데이터 세트들이 사용되었다.A total of 542 clinical data sets collected from a total of 146 patients were used.

전술한 학습방법 실시예#18에 기재한 갑상선호르몬농도 예측 모델의 학습 방법에 따라 모델이 학습되었다. The model was learned according to the learning method of the thyroid hormone concentration prediction model described in the above-described learning method example #18.

특히, (1) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 free T4의 농도] In particular, (1) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (segment heart rate corresponding to the first test day) Relative standard deviation - relative standard deviation of the interval heart rates corresponding to the second test day), (skewness of the interval heart rates corresponding to the second test day - skewness of the interval heart rates corresponding to the first test day), (skewness of the interval heart rates corresponding to the second test day) kurtosis of the interval heart rates corresponding to the first test date - kurtosis of the interval heart rates corresponding to the first test day), (JS Divergence between the interval heart rates corresponding to the second test day and the interval heart rates corresponding to the first test day), Concentration of free T4, concentration of TSH corresponding to the second test day, concentration of free T4 corresponding to the first test day]

(2) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제1 검사일에 대응되는 추정 갑상선 기능에 관한 값(항진 또는 항진아님), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 free T4의 농도](2) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard of section heart rates corresponding to the first test day Deviation - relative standard deviation of section heart rates corresponding to the second test day), (skewness of section heart rates corresponding to the second test day - skewness of section heart rates corresponding to the first test day), (distortion of section heart rates corresponding to the second test day) Kurtosis of the heart rates - kurtosis of the interval heart rates corresponding to the first examination day), (JS Divergence between the interval heart rates corresponding to the second examination day and the interval heart rates corresponding to the first examination day), estimated thyroid corresponding to the first examination day Values related to function (hyperactivity or not hyperactivity), concentration of free T4 corresponding to the second examination day, concentration of TSH corresponding to the second examination day, concentration of free T4 corresponding to the first examination day]

(3) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제1 검사일에 대응되는 추정 갑상선 기능에 관한 제1 값(항진 또는 항진아님), 제1 검사일에 대응되는 추정 갑상선 기능에 관한 제2 값(저하 또는 저하아님), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 free T4의 농도]와 같은 학습 데이터 세트들을 이용하여 학습이 진행되었다. (3) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard of section heart rates corresponding to the first test day Deviation - relative standard deviation of section heart rates corresponding to the second test day), (skewness of section heart rates corresponding to the second test day - skewness of section heart rates corresponding to the first test day), (distortion of section heart rates corresponding to the second test day) Kurtosis of the heart rates - kurtosis of the interval heart rates corresponding to the first examination day), (JS Divergence between the interval heart rates corresponding to the second examination day and the interval heart rates corresponding to the first examination day), estimated thyroid corresponding to the first examination day First value for function (hyper or not hyperactive), estimated corresponding to the first test date. Second value for thyroid function (hypo or not hypo), corresponding to the second test date. Concentration of free T4, corresponding to the second test date. Learning was conducted using learning data sets such as the concentration of TSH and the concentration of free T4 corresponding to the first test day.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 146명의 환자로부터 확보한 542개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,728개였으며, 이들 중 일부가 테스트 데이터 세트로 사용되었다.At this time, the clinical data used for learning was 542 data obtained from 146 patients, and the total number of learning data sets created from this was 1,728, some of which were used as test data sets.

갑상선호르몬농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

extra trees을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선호르몬농도 예측모델이 학습되었다.Using extra trees, a thyroid hormone concentration prediction model was learned using the aforementioned training data sets.

실험예#17에 따르는 free T4 농도 예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 23]과 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the free T4 concentration prediction model according to Experimental Example #17 are shown in [Table 23] below.

MAEM.A.E. MSEMSE MAPEMAPE (1)(One) 0.24210.2421 0.18300.1830 0.17220.1722 (2) (2) 0.24040.2404 0.17800.1780 0.17120.1712 (3)(3) 0.24030.2403 0.17700.1770 0.17200.1720

본 출원의 발명자들은, 갑상선호르몬농도를 예측함에 있어서, 대상일의 추정 갑상선 기능에 관한 값을 함께 사용하였을 때 MAE, MSE 및 MAPE의 성능 향상이 있음을 확인할 수 있었다. The inventors of the present application were able to confirm that when predicting thyroid hormone concentration, there was an improvement in the performance of MAE, MSE, and MAPE when values related to the estimated thyroid function of the target date were used together.

실험예#18.Experimental Example #18.

학습 데이터 세트들의 준비Preparation of training data sets

총 360명의 환자를 대상으로 수집한 총 1,504개의 임상 데이터 세트들이 사용되었다.A total of 1,504 clinical data sets collected from a total of 360 patients were used.

전술한 학습방법 실시예#19에 기재한 갑상선 호르몬 농도 예측 모델의 학습 방법에 따라 모델이 학습되었다. 특히, free T4의 농도를 예측하기 위한 예측모델이 학습되었다.The model was learned according to the learning method of the thyroid hormone concentration prediction model described in Learning Method Example #19 described above. In particular, a prediction model was learned to predict the concentration of free T4.

구체적으로, 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.Specifically, each thyroid hormone concentration prediction model was trained using the following training data set.

(1) 모델 1: [제2 검사일에 대응되는 free T4 농도, 제2 검사일에 대응되는 TSH 농도]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(1) Model 1: Learning data using [free T4 concentration corresponding to the second test day, TSH concentration corresponding to the second test day] as input values, and free T4 concentration corresponding to the first test day as the labeling value. Trained as a set.

(2) 모델 2: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(2) Model 2: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained using a training data set.

(3) 모델 3: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(3) Model 3: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change rate] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained using a training data set.

(4) 모델 4: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화율]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(4) Model 4: Use [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic average heart rate change rate] as input values, and label the free T4 concentration corresponding to the first test day. Trained with a training data set used as values.

(5) 모델 5: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(5) Model 5: Use [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic average heart rate change] as input values, and label the free T4 concentration corresponding to the first test day. Trained with a training data set used as values.

(6) 모델 6: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 제2 기준일의 구간 심박수의 표준편차]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(6) Model 6: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, standard deviation of the zone heart rate of the second reference day] are used as input values, and free corresponding to the first test day Trained with a training data set using T4 concentration as the labeling value.

(7) 모델 7: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 심박수의 왜도의 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(7) Model 7: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, change in skewness of interval heart rate] as input values, and free T4 concentration corresponding to the first test day Trained with a training data set using as labeling values.

(8) 모델 8: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 심박수의 첨도의 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(8) Model 8: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, change in kurtosis of interval heart rate] are used as input values, and free T4 concentration corresponding to the first test day is used as input. Trained with a training data set used as labeling values.

모델 2 내지 모델 8에서 이용된 입력값은 모델 1에서 이용된 입력값에 각각 구간 평균 심박수 변화량, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, 구간 조화평균 심박수 변화량, 제2 기준일의 구간 심박수의 표준편차, 구간 심박수의 왜도의 변화량, 구간 심박수의 첨도의 변화량을 더 포함한다.The input values used in Models 2 to 8 are the input values used in Model 1, respectively, the section average heart rate change, section average heart rate change rate, section harmonic average heart rate change, section harmonic average heart rate change, and the standard of section heart rate on the second reference day. It further includes deviation, change in skewness of the interval heart rate, and change in kurtosis of the interval heart rate.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 24]와 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 24] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 1model 1 0.20140.2014 0.43820.4382 모델 2model 2 0.47760.4776 0.35160.3516 모델 3model 3 0.47150.4715 0.35360.3536 모델 4model 4 0.46270.4627 0.35250.3525 모델 5model 5 0.45490.4549 0.35160.3516 모델 6model 6 0.19880.1988 0.43490.4349 모델 7model 7 0.19410.1941 0.43490.4349 모델 8model 8 0.18360.1836 0.43260.4326

본 출원의 발명자들은 모델 1을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 6 내지 8을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 2 내지 5를 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도와 함께 이용되는 구간 평균 심박수 변화량, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율 또는 구간 조화평균 심박수 변화량은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using Model 1, the Pearson Correlation was lowered when predicting thyroid hormone concentration using Models 6 to 8, while the thyroid hormone concentration was predicted using Models 2 to 5. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date It was confirmed that the section average heart rate change, section average heart rate change rate, section harmonic average heart rate change, or section harmonic average heart rate change used together are significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(9) 모델 9: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(9) Model 9: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day] as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(10) 모델 10: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 조화평균 심박수 변화율]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(10) Model 10: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval harmonic average heart rate change] are used as input values, and Trained with a training data set that uses free T4 concentrations as labeling values.

(11) 모델 11: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 얕은 수면 비율]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(11) Model 11: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the first test day] as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(12) 모델 12: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 평균 심박수 변화율]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(12) Model 12: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval average heart rate change rate] are used as input values, and free corresponding to the first test day Trained with a training data set using T4 concentration as the labeling value.

(13) 모델 13: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 심박수의 첨도]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(13) Model 13: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, kurtosis of the interval heart rate of the first test day] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(14) 모델 14: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 심박수의 상대표준편차]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(14) Model 14: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, relative standard deviation of interval heart rate on the first test day] are used as input values, and 1 Trained as a learning data set that uses the free T4 concentration corresponding to the test day as a labeling value.

(15) 모델 15: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 심박수의 첨도의 변화량]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(15) Model 15: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, change in section average heart rate, change in kurtosis of section heart rate] are used as input values, and correspond to the first test day. Trained with a training data set that uses the free T4 concentration as a labeling value.

(16) 모델 16: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 구간 심박수의 왜도의 변화량]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(16) Model 16: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, change in section average heart rate, change in skewness of section heart rate] are used as input values, and on the first test day Trained with a training data set that uses the corresponding free T4 concentration as a labeling value.

(17) 모델 17: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제1 검사일의 구간 심박수의 왜도]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(17) Model 17: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, skewness of interval heart rate of the first test day] are used as input values, and the first It is trained as a learning data set that uses the free T4 concentration corresponding to the test date as a labeling value.

모델 9 내지 모델 17에서 이용된 입력값은 모델 2에서 이용된 입력값에 각각 제2 검사일의 구간 얕은 수면 비율, 구간 조화평균 심박수 변화율, 제1 검사일의 구간 얕은 수면 비율, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 심박수의 상대표준편차, 구간 심박수의 첨도의 변화량, 구간 심박수의 왜도의 변화량, 제1 검사일의 구간 심박수의 왜도를 더 포함한다.The input values used in Models 9 to 17 are the input values used in Model 2, respectively, the section light sleep rate, section harmonic average heart rate change rate on the second test day, section light sleep rate, section average heart rate change rate, and section average heart rate change rate on the first test day. It further includes kurtosis of the section heart rate on the first test day, relative standard deviation of the section heart rate on the first test day, change in kurtosis of the section heart rate, change in skewness of the section heart rate, and skewness of the section heart rate on the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 25]와 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 25] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 2model 2 0.47760.4776 0.35160.3516 모델 9model 9 0.53760.5376 0.35050.3505 모델 10model 10 0.50950.5095 0.34750.3475 모델 11model 11 0.50950.5095 0.35920.3592 모델 12model 12 0.5020.502 0.34860.3486 모델 13model 13 0.50080.5008 0.35610.3561 모델 14model 14 0.4580.458 0.36170.3617 모델 15model 15 0.45720.4572 0.3610.361 모델 16model 16 0.45650.4565 0.36230.3623 모델 17model 17 0.42720.4272 0.36220.3622

본 출원의 발명자들은 모델 2을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 14 내지 17을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 9 내지 13을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량과 함께 이용되는 제2 검사일의 구간 얕은 수면 비율, 구간 조화평균 심박수 변화율, 제1 검사일의 구간 얕은 수면 비율, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 첨도는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using Model 2, the Pearson Correlation was lowered when predicting thyroid hormone concentration using Models 14 to 17, while the thyroid hormone concentration was predicted using Models 9 to 13. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date , the interval light sleep ratio of the second test day, the interval harmonic average heart rate change rate, the interval light sleep ratio of the first test day, the interval average heart rate change rate, and the kurtosis of the interval heart rate of the first test day, used together with the interval average heart rate change, improve performance. It was confirmed that it was a significant variable.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(18) 모델 18: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(18) Model 18: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(19) 모델 19: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 REM 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(19) Model 19: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval REM sleep ratio of the second test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(20) 모델 20: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 구간 조화평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(20) Model 20: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change, section light sleep rate of the second test day, section harmonic average heart rate change rate] as input values It is learned as a learning data set using the free T4 concentration corresponding to the first test day as a labeling value.

(21) 모델 21: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(21) Model 21: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, relative standard deviation of the interval heart rate of the second test day ] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(22) 모델 22: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제2 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(22) Model 22: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, standard deviation of the interval heart rate of the second test day] is used as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(23) 모델 23: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 구간 평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(23) Model 23: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change, section light sleep ratio, section average heart rate change rate on the second test day] as input values And, it is learned as a learning data set that uses the free T4 concentration corresponding to the first inspection day as a labeling value.

(24) 모델 24: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 얕은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(24) Model 24: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, interval light sleep ratio of the first test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(25) 모델 25: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(25) Model 25: Enter [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change, section light sleep ratio of the second test day, section average heart rate of the first test day] It is trained as a learning data set using the free T4 concentration corresponding to the first test day as the labeling value.

(26) 모델 26: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(26) Model 26: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, relative standard deviation of the interval heart rate of the first test day ] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(27) 모델 27: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(27) Model 27: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, skewness of the interval heart rate of the first test day] is used as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

모델 18 내지 모델 27에서 이용된 입력값은 모델 9에서 이용된 입력값에 각각 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 REM 수면 비율, 구간 조화평균 심박수 변화율, 제2 검사일의 구간 심박수의 상대표준편차, 제2 검사일의 구간 심박수의 표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 왜도를 더 포함한다.The input values used in Models 18 to 27 are the input values used in Model 9, respectively, kurtosis of the section heart rate on the first test day, section REM sleep rate on the second test day, section harmonic mean heart rate change rate, and section heart rate on the second test day. Relative standard deviation of, standard deviation of interval heart rate on the second test day, interval average heart rate change rate, interval light sleep rate on the first test day, interval average heart rate on the first test day, relative standard deviation of the interval heart rate on the first test day, 1 It further includes the skewness of the zone heart rate on the test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다.The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 26]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 26] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 9model 9 0.53760.5376 0.35050.3505 모델 18model 18 0.5770.577 0.33820.3382 모델 19model 19 0.56680.5668 0.3430.343 모델 20model 20 0.56620.5662 0.33850.3385 모델 21model 21 0.56610.5661 0.33840.3384 모델 22model 22 0.56520.5652 0.33930.3393 모델 23model 23 0.56440.5644 0.33890.3389 모델 24model 24 0.56010.5601 0.34050.3405 모델 25model 25 0.52750.5275 0.33640.3364 모델 26model 26 0.51350.5135 0.34530.3453 모델 27model 27 0.50430.5043 0.34380.3438

본 출원의 발명자들은 모델 9을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 25 내지 27을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 18 내지 24을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율과 함께 이용되는 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 REM 수면 비율, 구간 조화평균 심박수 변화율, 제2 검사일의 구간 심박수의 상대표준편차, 제2 검사일의 구간 심박수의 표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 얕은 수면 비율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using Model 9, the Pearson Correlation was lowered when predicting thyroid hormone concentration using Models 25 to 27, while the thyroid hormone concentration was predicted using Models 18 to 24. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date , the interval average heart rate change, the kurtosis of the interval heart rate on the first test day used with the interval light sleep rate on the second test day, the interval REM sleep rate on the second test day, the interval harmonic mean heart rate change, the relative of the interval heart rate on the second test day The standard deviation, standard deviation of section heart rate on the second test day, rate of change in section average heart rate, and section light sleep ratio on the first test day were confirmed to be significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(28) 모델 28: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(28) Model 28: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of heart rate in the interval of the test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(29) 모델 29: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 구간 조화평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(29) Model 29: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, interval Harmonic average heart rate change rate] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(30) 모델 30: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(30) Model 30: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Standard deviation of the heart rate of the interval of the test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(31) 모델 31: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 구간 평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(31) Model 31: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, interval Average heart rate change rate] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(32) 모델 32: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(32) Model 32: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, It is learned as a learning data set that uses [section average heart rate of test day 1] as an input value and uses the free T4 concentration corresponding to the first test day as a labeling value.

(33) 모델 33: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(33) Model 33: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 1 Relative standard deviation of heart rate in the interval of test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(34) 모델 34: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(34) Model 34: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 1 Skewness of section heart rate on test day] is used as input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

모델 28 내지 모델 34에서 이용된 입력값은 모델 18에서 이용된 입력값에 각각 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화율, 제2 검사일의 구간 심박수의 표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 왜도를 더 포함한다.The input values used in Models 28 to 34 are the relative standard deviation of the section heart rate on the second test day, the section harmonic average heart rate change rate, the standard deviation of the section heart rate on the second test day, and the section average heart rate in the input values used in Model 18, respectively. It further includes the rate of change, the average heart rate of the section on the first test day, the relative standard deviation of the section heart rate on the first test day, and the skewness of the section heart rate on the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다.The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 27]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 27] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 18model 18 0.5770.577 0.33820.3382 모델 28model 28 0.59360.5936 0.33420.3342 모델 29model 29 0.58730.5873 0.33410.3341 모델 30model 30 0.5870.587 0.33520.3352 모델 31model 31 0.5810.581 0.33470.3347 모델 32model 32 0.55110.5511 0.33570.3357 모델 33model 33 0.52660.5266 0.33950.3395 모델 34model 34 0.51030.5103 0.34060.3406

본 출원의 발명자들은 모델 18을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 32 내지 34을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 28 내지 31을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도와 함께 이용되는 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화율, 제2 검사일의 구간 심박수의 표준편차, 구간 평균 심박수 변화율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using Model 18, the Pearson Correlation was lowered when predicting thyroid hormone concentration using Models 32 to 34, while the thyroid hormone concentration was predicted using Models 28 to 31. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date , interval average heart rate change, interval light sleep ratio on the second test day, relative standard deviation of the interval heart rate on the second test day used with the kurtosis of the interval heart rate on the first test day, interval harmonic mean heart rate change, interval heart rate on the second test day. It was confirmed that the standard deviation and section average heart rate change rate are significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(35) 모델 35: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(35) Model 35: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate of the test day, section average heart rate change rate] as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(36) 모델 36: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(36) Model 36: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval harmonic average heart rate change rate] is used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(37) 모델 37: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 조화평균 심박수 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(37) Model 37: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval harmonic average heart rate change] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(38) 모델 38: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(38) Model 38: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of the zone heart rate on the test day, standard deviation of the zone heart rate on the first test day] as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(39) 모델 39: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(39) Model 39: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of the section heart rate of the test day, section average heart rate of the first test day] as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(40) 모델 40: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(40) Model 40: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of the interval heart rate of the test day, skewness of the interval heart rate of the first test day] as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

모델 35 내지 모델 40에서 이용된 입력값은 모델 28에서 이용된 입력값에 각각 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, 구간 조화평균 심박수 변화량, 제1 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 심박수의 왜도를 더 포함한다.The input values used in Models 35 to 40 are the input values used in Model 28, respectively, interval average heart rate change rate, interval harmonic average heart rate change rate, interval harmonic average heart rate change amount, standard deviation of interval heart rate on the first test day, and first test day. It further includes the section average heart rate and the skewness of the section heart rate on the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 28]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 28] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 28model 28 0.59360.5936 0.33420.3342 모델 35model 35 0.59760.5976 0.33060.3306 모델 36model 36 0.59510.5951 0.33210.3321 모델 37model 37 0.58960.5896 0.33350.3335 모델 38model 38 0.56370.5637 0.33560.3356 모델 39model 39 0.55970.5597 0.33470.3347 모델 40model 40 0.53850.5385 0.33690.3369

본 출원의 발명자들은 모델 28을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 37 내지 40을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 35 내지 36을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.한편, 본 출원의 발명자들은 모델 28을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 37을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation은 상대적으로 작은 수치로 저하되지만, MAE는 향상되는 것을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using model 28, the Pearson Correlation was lowered when predicting thyroid hormone concentration using models 37 to 40, while the thyroid hormone concentration was predicted using models 35 to 36. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Meanwhile, the inventors of the present application found that compared to predicting the thyroid hormone concentration using Model 28, when predicting the thyroid hormone concentration using Model 37, the Pearson Correlation It was confirmed that the was reduced to a relatively small value, but the MAE was improved.

이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차와 함께 이용되는 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, 구간 조화평균 심박수 변화량은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.When confirmed based on this, the inventors of the present application predict the thyroid hormone concentration, the TSH concentration corresponding to the second test day, the free T4 concentration corresponding to the second test day, the average heart rate change in the section, and the rate of light sleep in the section on the second test day. , It was confirmed that the section average heart rate change rate, section harmonic average heart rate change rate, and section harmonic average heart rate change used along with the kurtosis of the section heart rate on the first test day and the relative standard deviation of the section heart rate on the second test day are significant variables that improve performance. did.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(41) 모델 41: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(41) Model 41: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval average heart rate change rate, and the interval harmonic average heart rate change rate] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(42) 모델 42: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 깊은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(42) Model 42: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section deep sleep rate on the second test day] as input values, and free T4 concentration corresponding to the first test day as a labeling value as a learning data set. Learned.

(43) 모델 43: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(43) Model 43: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval average heart rate change rate, and the interval harmonic average heart rate change] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(44) 모델 44: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(44) Model 44: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 A learning data set that uses the relative standard deviation of the interval heart rate of the test day, the rate of change of the interval average heart rate, and the skewness of the interval heart rate of the second test day as input values, and uses the free T4 concentration corresponding to the first test day as a labeling value. Learned with .

(45) 모델 45: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 대상의 성별] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(45) Model 45: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, and gender of the subject] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(46) 모델 46: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 심박수의 JS Divergence] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(46) Model 46: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the rate of change of the interval average heart rate, and the JS Divergence of the interval heart rate] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(47) 모델 47: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 REM 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(47) Model 47: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate of the test day, section average heart rate change rate, section REM sleep ratio of the second test day] as input values, and free T4 concentration corresponding to the first test day as a labeling value as a learning data set. Learned.

(48) 모델 48: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 심박수의 첨도의 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(48) Model 48: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate on the test day, the rate of change of the average heart rate of the interval, and the change in kurtosis of the interval heart rate] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value. .

(49) 모델 49: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제2 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(49) Model 49: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 A learning data set that uses the relative standard deviation of the section heart rate on the test day, the rate of change in section average heart rate, and the standard deviation of the section heart rate on the second test day as input values, and uses the free T4 concentration corresponding to the first test day as a labeling value. Learned with .

(50) 모델 50: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(50) Model 50: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 A learning data set that uses the relative standard deviation of the section heart rate on the test day, the rate of change of the average heart rate in the section, and the standard deviation of the section heart rate on the first test day as input values, and uses the free T4 concentration corresponding to the first test day as a labeling value. Learned with .

(51) 모델 51: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(51) Model 51: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the section heart rate of the test day, the rate of change of the section average heart rate, and the section harmonic average heart rate of the first test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value as a learning data set. Learned.

(52) 모델 52: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(52) Model 52: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 A learning data set that uses the relative standard deviation of the interval heart rate on the test day, the rate of change of the interval average heart rate, and the skewness of the interval heart rate on the first test day as input values, and uses the free T4 concentration corresponding to the first test day as a labeling value. Learned with .

(53) 모델 53: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(53) Model 53: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Learning data using the relative standard deviation of the section heart rate of the test day, the rate of change of the average heart rate of the section, and the relative standard deviation of the section heart rate of the first test day as input values, and the free T4 concentration corresponding to the first test day as the labeling value. Trained as a set.

모델 41 내지 모델 53에서 이용된 입력값은 모델 35에서 이용된 입력값에 각각 구간 조화평균 심박수 변화율, 제2 검사일의 구간 깊은 수면 비율, 구간 조화평균 심박수 변화량, 제2 검사일의 구간 심박수의 왜도, 대상의 성별, 구간 심박수의 JS Divergence, 제2 검사일의 구간 REM 수면 비율, 구간 심박수의 첨도의 변화량, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 조화평균 심박수, 제1 검사일의 구간 심박수의 왜도, 제1 검사일의 구간 심박수의 상대표준편차를 더 포함한다.The input values used in Models 41 to 53 are the input values used in Model 35, respectively, the interval harmonic average heart rate change rate, the interval deep sleep rate on the second test day, the interval harmonic average heart rate change, and the skewness of the interval heart rate on the second test day. , gender of the subject, JS Divergence of section heart rate, section REM sleep ratio on the second test day, change in kurtosis of section heart rate, standard deviation of section heart rate on the second test day, standard deviation of section heart rate on the first test day, first test day It further includes the interval harmonic average heart rate, the skewness of the interval heart rate on the first test day, and the relative standard deviation of the interval heart rate on the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다.The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 29]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 29] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 35model 35 0.59760.5976 0.33060.3306 모델 41model 41 0.61120.6112 0.33030.3303 모델 42model 42 0.60720.6072 0.33030.3303 모델 43model 43 0.60470.6047 0.33090.3309 모델 44model 44 0.60460.6046 0.33240.3324 모델 45model 45 0.6040.604 0.3310.331 모델 46model 46 0.60320.6032 0.33010.3301 모댈 47Model 47 0.60270.6027 0.33230.3323 모델 48model 48 0.60240.6024 0.33110.3311 모델 49model 49 0.60.6 0.33350.3335 모델 50model 50 0.56550.5655 0.33460.3346 모델 51model 51 0.56520.5652 0.3340.334 모델 52model 52 0.55280.5528 0.33240.3324 모델 53model 53 0.53730.5373 0.33440.3344

본 출원의 발명자들은 모델 35을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 50 내지 53을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 41 내지 49을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율과 함께 이용되는 구간 조화평균 심박수 변화율, 제2 검사일의 구간 깊은 수면 비율, 구간 조화평균 심박수 변화량, 제2 검사일의 구간 심박수의 왜도, 대상의 성별, 구간 심박수의 JS Divergence, 제2 검사일의 구간 REM 수면 비율, 구간 심박수의 첨도의 변화량, 제2 검사일의 구간 심박수의 표준편차는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using model 35, the Pearson Correlation was lowered when predicting thyroid hormone concentration using models 50 to 53, while the thyroid hormone concentration was predicted using models 41 to 49. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date , interval average heart rate change, interval light sleep ratio on the second test day, kurtosis of the interval heart rate on the first test day, relative standard deviation of the interval heart rate on the second test day, interval harmonic average heart rate change rate used with the interval average heart rate change rate, 2 The interval deep sleep ratio on the test day, the change in interval harmonic average heart rate, the skewness of the interval heart rate on the second test day, the subject's gender, the JS Divergence of the interval heart rate, the interval REM sleep ratio on the second test day, the change in kurtosis of the interval heart rate, The standard deviation of the zone heart rate on the second test day was confirmed to be a significant variable that improves performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(54) 모델 54: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(54) Model 54: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (free T4 concentration of the second test day x kurtosis of the interval heart rate of the first test day)] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(55) 모델 55: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 / 제2 검사일의 구간 얕은 수면 비율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(55) Model 55: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate on the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (free T4 concentration on the second test day / interval light sleep ratio on the second test day)] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(56) 모델 56: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(56) Model 56: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate on the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (free T4 concentration on the second test day x the relative standard deviation of the interval heart rate on the second test day)] are used as input values, and 1 Trained as a learning data set that uses the free T4 concentration corresponding to the test day as a labeling value.

(57) 모델 57: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(57) Model 57: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (section average heart rate change / free T4 concentration on the second test day)] are used as input values, and the free corresponding to the first test day Trained with a training data set using T4 concentration as the labeling value.

(58) 모델 58: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(58) Model 58: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (interval light sleep ratio of the second test day / free T4 concentration of the second test day)] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(59) 모델 59: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(59) Model 59: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the zone heart rate of the test day, the rate of change of the section average heart rate, the rate of change of the section harmonic average heart rate, (zone light sleep ratio of the second test day x the relative standard deviation of the section heart rate of the second test day)] are used as input values, Learned with a learning data set that uses the free T4 concentration corresponding to the first inspection day as a labeling value.

(60) 모델 60: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 TSH 농도 x 구간 평균 심박수 변화량)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(60) Model 60: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate on the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (TSH concentration on the second test day x the interval average heart rate change)] are used as input values, and free T4 corresponding to the first test day Trained with a training data set that uses concentration as the labeling value.

(61) 모델 61: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (구간 평균 심박수 변화량 x 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(61) Model 61: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the section heart rate on the test day, the section average heart rate change rate, the section harmonic average heart rate change rate, ((section average heart rate change x the relative standard deviation of the section heart rate on the second test day)] are used as input values, and on the first test day, Trained with a training data set that uses the corresponding free T4 concentration as a labeling value.

(62) 모델 62: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 구간 평균 심박수 변화량)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(62) Model 62: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of the interval heart rate of the test day, the interval average heart rate change rate, the interval harmonic average heart rate change rate, (free T4 concentration of the second test day x the interval average heart rate change)] are used as input values, and the free corresponding to the first test day Trained with a training data set using T4 concentration as the labeling value.

(63) 모델 63: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (구간 평균 심박수 변화율 / 구간 평균 심박수 변화량)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(63) Model 63: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 The relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (section average heart rate change rate / section average heart rate change amount)] are used as input values, and the free T4 concentration corresponding to the first test day is used as input. Trained with a training data set used as labeling values.

모델 54 내지 모델 63에서 이용된 입력값은 모델 41에서 이용된 입력값에 각각 (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 free T4 농도 / 제2 검사일의 구간 얕은 수면 비율), (제2 검사일의 free T4 농도 x 제2 검사일의 구간 심박수의 상대표준편차), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 TSH 농도 x 구간 평균 심박수 변화량), (구간 평균 심박수 변화량 x 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 free T4 농도 x 구간 평균 심박수 변화량), (구간 평균 심박수 변화율 / 구간 평균 심박수 변화량)을 더 포함한다.The input values used in Models 54 to 63 are the input values used in Model 41, respectively (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (free T4 concentration on the second test day / second Zone light sleep ratio on the test day), (free T4 concentration on the second test day Light sleep ratio/free T4 concentration on the second test day), (interval light sleep ratio on the second test day x relative standard deviation of the interval heart rate on the second test day), (TSH concentration on the second test day It further includes section average heart rate change x relative standard deviation of section heart rate on the second test day), (free T4 concentration on the second test day x section average heart rate change), (section average heart rate change rate / section average heart rate change).

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 30]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 30] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 41model 41 0.61120.6112 0.33030.3303 모델 54model 54 0.61660.6166 0.32760.3276 모델 55model 55 0.61270.6127 0.32910.3291 모델 56model 56 0.61010.6101 0.32770.3277 모델 57model 57 0.60890.6089 0.33010.3301 모델 58model 58 0.60870.6087 0.32980.3298 모델 59model 59 0.60820.6082 0.32890.3289 모댈 60modal 60 0.59470.5947 0.33170.3317 모델 61model 61 0.59310.5931 0.33290.3329 모델 62model 62 0.59120.5912 0.33080.3308 모델 63model 63 0.58460.5846 0.33210.3321

본 출원의 발명자들은 모델 41을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 56 내지 63을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 54 내지 55를 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.한편, 본 출원의 발명자들은 모델 41을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 56 내지 모델 59를 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation은 상대적으로 작은 수치로 저하되지만, MAE는 향상되는 것을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using model 41, the Pearson Correlation was lowered when predicting thyroid hormone concentration using models 56 to 63, while the thyroid hormone concentration was predicted using models 54 to 55. It was confirmed that Pearson Correlation is improved when predicting hormone concentration. Meanwhile, the inventors of the present application predicted thyroid hormone concentration using Model 56 to Model 59 compared to predicting thyroid hormone concentration using Model 41. It was confirmed that Pearson Correlation deteriorates to a relatively small value, but MAE improves.

이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율과 함께 이용되는 (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 free T4 농도 / 제2 검사일의 구간 얕은 수면 비율), (제2 검사일의 free T4 농도 x 제2 검사일의 구간 심박수의 상대표준편차), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차)는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.When confirmed based on this, the inventors of the present application predict the thyroid hormone concentration, the TSH concentration corresponding to the second test day, the free T4 concentration corresponding to the second test day, the average heart rate change in the section, and the rate of light sleep in the section on the second test day. , kurtosis of the interval heart rate on the first test day, relative standard deviation of the interval heart rate on the second test day, interval average heart rate change rate, and interval harmonic average heart rate change rate (free T4 concentration on the second test day x interval heart rate on the first test day) kurtosis), (free T4 concentration on the second test day / interval light sleep ratio on the second test day), (free T4 concentration on the second test day x relative standard deviation of the interval heart rate on the second test day), (interval average heart rate change / free T4 concentration on the second test day), (interval light sleep ratio on the second test day / free T4 concentration on the second test day), (interval light sleep ratio on the second test day x relative standard deviation of the interval heart rate on the second test day) It was confirmed to be a significant variable that improves performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(64) 모델 64: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화량 / 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(64) Model 64: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the section heart rate on the first test day), (section average heart rate change / second test day Relative standard deviation of section heart rate)] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(65) 모델 65: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 조화평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(65) Model 65: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of interval heart rate on the test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (interval harmonic average heart rate change rate / second test day Relative standard deviation of heart rate in the interval)] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(66) 모델 66: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(66) Model 66: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of the interval heart rate on the test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (zone light sleep ratio on the second test day / Free T4 concentration on the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(67) 모델 67: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(67) Model 67: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the section heart rate on the first test day), (section average heart rate change / second test day free T4 concentration)] as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(68) 모델 68: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(68) Model 68: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the section heart rate on the first test day), (section average heart rate change rate / second test day Relative standard deviation of section heart rate)] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(69) 모델 69: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(69) Model 69: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the section heart rate on the first test day), (zone light sleep rate on the second test day Relative standard deviation of the interval heart rate on the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(70) 모델 70: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (제2 검사일의 구간 심박수의 상대표준편차 / 제1 검사일의 구간 심박수의 첨도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(70) Model 70: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of the interval heart rate on the test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (relative standard of the interval heart rate on the second test day Deviation / kurtosis of the interval heart rate of the first test day) is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(71) 모델 71: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화율 / 구간 조화평균 심박수 변화율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(71) Model 71: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of interval heart rate on the test day, interval average heart rate change rate, interval harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the interval heart rate on the first test day), (section average heart rate change rate / interval harmonic average heart rate Rate of change] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(72) 모델 72: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화량 / 구간 평균 심박수 변화율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(72) Model 72: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change, interval light sleep ratio of the second test day, kurtosis of the interval heart rate of the first test day, 2 Relative standard deviation of section heart rate on the test day, section average heart rate change rate, section harmonic average heart rate change rate, (free T4 concentration on the second test day x kurtosis of the section heart rate on the first test day), (section average heart rate change / section average heart rate change rate )] as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

모델 64 내지 모델 72에서 이용된 입력값은 모델 54에서 이용된 입력값에 각각 (구간 평균 심박수 변화량 / 제2 검사일의 구간 심박수의 상대표준편차), (구간 조화평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 구간 심박수의 상대표준편차 / 제1 검사일의 구간 심박수의 첨도), (구간 평균 심박수 변화율 / 구간 조화평균 심박수 변화율), (구간 평균 심박수 변화량 / 구간 평균 심박수 변화율)을 더 포함한다.The input values used in Models 64 to 72 are the input values used in Model 54, respectively (section average heart rate change / relative standard deviation of section heart rate on the second test day), (section harmonic average heart rate change rate / section on the second test day) Relative standard deviation of heart rate), (section light sleep ratio on the second test day / free T4 concentration on the second test day), (section average heart rate change / free T4 concentration on the second test day), (section average heart rate change rate / second test day (Relative standard deviation of the zone heart rate of the second test day), (Relative standard deviation of the zone heart rate of the second test day × Zone light sleep ratio of the second test day), (Relative standard deviation of the zone heart rate of the second test day / Kurtosis), (section average heart rate change rate / section harmonic average heart rate change rate), (section average heart rate change / section average heart rate change rate) are further included.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was tested using the above-described test sets.

실험예#18에 따라 학습된 갑상선 호르몬 농도 예측모델의 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 31]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) of the thyroid hormone concentration prediction model learned according to Experimental Example #18 are shown in [Table 31] below.

Pearson CorrelationPearson Correlation MAEM.A.E. 모델 54model 54 0.61660.6166 0.32760.3276 모델 64model 64 0.62150.6215 0.32730.3273 모델 65model 65 0.62110.6211 0.32830.3283 모델 66model 66 0.62080.6208 0.32820.3282 모델 67model 67 0.62050.6205 0.32830.3283 모델 68model 68 0.61980.6198 0.32790.3279 모델 69model 69 0.61720.6172 0.32670.3267 모댈 70Model 70 0.60180.6018 0.32990.3299 모델 71model 71 0.60130.6013 0.32760.3276 모델 72model 72 0.59070.5907 0.33060.3306

본 출원의 발명자들은 모델 54을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 70 내지 71을 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 저하되는 반면에, 모델 64 내지 69를 이용하여 갑상선호르몬농도를 예측하면 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, (제2 검사일의 free T4 농도 x 제1 검사일의 구간 심박수의 첨도)와 함께 이용되는 (구간 평균 심박수 변화량 / 제2 검사일의 구간 심박수의 상대표준편차), (구간 조화평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 구간 얕은 수면 비율 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화량 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화율 / 제2 검사일의 구간 심박수의 상대표준편차), (제2 검사일의 구간 얕은 수면 비율 x 제2 검사일의 구간 심박수의 상대표준편차) 는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using model 54, the Pearson Correlation was lowered when predicting thyroid hormone concentration using models 70 to 71, while the thyroid hormone concentration was predicted using models 64 to 69. It was confirmed that the Pearson Correlation was improved when predicting the hormone concentration. Based on this, the inventors of the present application found that in predicting the thyroid hormone concentration, the TSH concentration corresponding to the second test date and the free T4 concentration corresponding to the second test date , section average heart rate change, section light sleep ratio on the second test day, kurtosis of the section heart rate on the first test day, relative standard deviation of the section heart rate on the second test day, section average heart rate change rate, section harmonic average heart rate change rate, (second test day of free T4 concentration Relative standard deviation), (section light sleep ratio on the second test day / free T4 concentration on the second test day), (section average heart rate change / free T4 concentration on the second test day), (section average heart rate change rate / section on the second test day Relative standard deviation of heart rate), (percentage of light sleep in the section on the second test day x relative standard deviation of the heart rate in the section on the second test day) were confirmed to be significant variables that improve performance.

실험예#19.Experimental Example #19.

학습 데이터 세트들의 준비Preparation of training data sets

총 360명의 환자를 대상으로 수집한 총 1,504개의 임상 데이터 세트들이 사용되었다.A total of 1,504 clinical data sets collected from a total of 360 patients were used.

전술한 학습방법 실시예#19에 기재한 갑상선 호르몬 농도 예측 모델의 학습 방법에 따라 모델이 학습되었다. 특히, free T4의 농도를 예측하기 위한 예측모델이 학습되었다.The model was learned according to the learning method of the thyroid hormone concentration prediction model described in Learning Method Example #19 described above. In particular, a prediction model was learned to predict the concentration of free T4.

구체적으로, 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.Specifically, each thyroid hormone concentration prediction model was trained using the following training data set.

(1) 모델 1: [제2 검사일에 대응되는 free T4 농도, 제2 검사일에 대응되는 TSH 농도]를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(1) Model 1: Learning data using [free T4 concentration corresponding to the second test day, TSH concentration corresponding to the second test day] as input values, and free T4 concentration corresponding to the first test day as the labeling value. Trained as a set.

(2) 모델 2: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(2) Model 2: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change rate] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained using a training data set.

(3) 모델 3: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화율]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(3) Model 3: Use [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic average heart rate change rate] as input, and label the free T4 concentration corresponding to the first test day. Trained with a training data set used as values.

(4) 모델 4: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 조화평균 심박수 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(4) Model 4: Use [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval harmonic average heart rate change] as input values, and label the free T4 concentration corresponding to the first test day. Trained with a training data set used as values.

(5) 모델 5: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화량]을 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(5) Model 5: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, section average heart rate change] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained using a training data set.

모델 2 내지 모델 5에서 이용된 입력값은 모델 1에서 이용된 입력값에 각각 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율, 구간 조화평균 심박수 변화량, 구간 평균 심박수 변화량을 더 포함한다.The input values used in Models 2 to 5 further include the section average heart rate change rate, section harmonic average heart rate change rate, section harmonic average heart rate change amount, and section average heart rate change amount to the input values used in Model 1, respectively.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 MAE (Mean Absolute Error) 및 Gray zone에 대한 Pearson Correlation을 확인하였다.The Pearson Correlation for the MAE (Mean Absolute Error) and gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was confirmed.

Gray zone은 갑상선기능항진증 판단의 기준이 되는 free T4 농도 값 근처의 구간을 의미한다. 갑상선기능항진증에 대해 오분류되는 경우 중 다수가 갑상선기능항진증의 판단 기준이 되는 free T4 농도 값 근처의 호르몬 수치를 예측하는 경우인 바, Gray zone에 대한 Pearson Correlation이 높을수록 오분류되는 경우가 줄어들 수 있다. 본 실험예#19에서는 갑상선기능항진증 판단의 기준이 되는 free T4 농도를 1.78ng/dL로 결정하고, Gray zone을 1.5ng/dL 내지 2.1ng/dL 구간으로 설정하였다.Gray zone refers to the area near the free T4 concentration value, which is the standard for determining hyperthyroidism. Many of the cases of misclassification for hyperthyroidism are cases of predicting hormone levels near the free T4 concentration value, which is the standard for judging hyperthyroidism, and the higher the Pearson Correlation for the gray zone, the fewer cases of misclassification. You can. In this Experimental Example #19, the free T4 concentration, which is the standard for determining hyperthyroidism, was determined to be 1.78ng/dL, and the gray zone was set to the range from 1.5ng/dL to 2.1ng/dL.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 MAE (Mean Absolute Error) 및 Gray zone에 대한 Pearson Correlation은 다음의 [표 32]과 같다.The Pearson Correlation for MAE (Mean Absolute Error) and gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 is shown in [Table 32] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 1model 1 0.12910.1291 0.43820.4382 모델 2model 2 0.20740.2074 0.35360.3536 모델 3model 3 0.20310.2031 0.35250.3525 모델 4model 4 0.19290.1929 0.35160.3516 모델 5model 5 0.19020.1902 0.35160.3516

본 출원의 발명자들은 모델 1을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 2 내지 5를 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도와 함께 이용되는 구간 평균 심박수 변화량, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율 또는 구간 조화평균 심박수 변화량은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application confirmed that the Pearson Correlation for the gray zone was improved when predicting thyroid hormone concentration using Models 2 to 5, compared to predicting thyroid hormone concentration using Model 1. Based on this, it can be confirmed When predicting thyroid hormone concentration, the inventors of the present application use the TSH concentration corresponding to the second test date, the free T4 concentration corresponding to the second test day, the interval average heart rate change, the interval average heart rate change rate, and the interval harmonic average. It was confirmed that heart rate change rate or interval harmonic average heart rate change is a significant variable that improves performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(6) 모델 6: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(6) Model 6: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day] as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(7) 모델 7: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제2 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(7) Model 7: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate of the second test day] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(8) 모델 8: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(8) Model 8: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate change] are used as input values, and Trained with a training data set that uses free T4 concentrations as labeling values.

(9) 모델 9: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 구간 조화평균 심박수 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(9) Model 9: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval harmonic average heart rate change rate] are used as input values, and Trained with a training data set that uses free T4 concentrations as labeling values.

(10) 모델 10: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(10) Model 10: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, kurtosis of the interval heart rate of the first test day] are used as input values, and the first test day It is trained as a training data set that uses the free T4 concentration corresponding to as a labeling value.

(11) 모델 11: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제2 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(11) Model 11: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, relative standard deviation of interval heart rate on the second test day] are used as input values, and 1 Trained as a learning data set that uses the free T4 concentration corresponding to the test date as a labeling value.

(12) 모델 12: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 구간 심박수의 JS Divergence] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(12) Model 12: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, JS Divergence of interval heart rate] are used as input values, and Trained with a training data set that uses free T4 concentrations as labeling values.

모델 6 내지 모델 12에서 이용된 입력값은 모델 2에서 이용된 입력값에 각각 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 조화평균 심박수, 구간 조화평균 심박수 변화량, 구간 조화평균 심박수 변화율, 제1 검사일의 구간 심박수의 첨도, 제2 검사일의 구간 심박수의 상대표준편차, 구간 심박수의 JS Divergence를 더 포함한다.The input values used in Models 6 to 12 are the input values used in Model 2, respectively, the section REM sleep ratio on the first test day, the section harmonic average heart rate on the second test day, the section harmonic average heart rate change, the section harmonic average heart rate change rate, It further includes kurtosis of the zone heart rate on the first test day, relative standard deviation of the zone heart rate on the second test day, and JS Divergence of the zone heart rate.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 33]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 33] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 2model 2 0.20740.2074 0.35360.3536 모델 6model 6 0.22710.2271 0.36630.3663 모델 7model 7 0.220.22 0.35240.3524 모델 8model 8 0.20790.2079 0.34940.3494 모델 9model 9 0.20740.2074 0.34990.3499 모델 10model 10 0.19010.1901 0.35860.3586 모델 11model 11 0.18390.1839 0.36660.3666 모델 12model 12 0.18340.1834 0.36060.3606

본 출원의 발명자들은 모델 2을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 10 내지 12를 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 6 내지 8을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.한편, 본 출원의 발명자들은 모델 2를 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 9을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation은 동일하지만, MAE는 향상되는 것을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using Model 2, when predicting thyroid hormone concentration using Models 10 to 12, the Pearson Correlation for the gray zone deteriorates, while Models 6 to 8 It was confirmed that the Pearson Correlation for the gray zone was improved when predicting the thyroid hormone concentration using When predicting thyroid hormone concentration, the Pearson Correlation for the gray zone was the same, but the MAE was confirmed to be improved.

이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율과 함께 이용되는 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 조화평균 심박수, 구간 조화평균 심박수 변화량, 구간 조화평균 심박수 변화율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.Based on this, the inventors of the present application believe that in predicting thyroid hormone concentration, the TSH concentration corresponding to the second test day, the free T4 concentration corresponding to the second test day, and the section of the first test day are used together with the average heart rate change rate of the section. REM sleep ratio, section harmonic average heart rate, section harmonic average heart rate change, and section harmonic average heart rate change rate on the second test day were confirmed to be significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(13) 모델 13: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(13) Model 13: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject] are used as input values , learned as a learning data set that uses the free T4 concentration corresponding to the first inspection day as a labeling value.

(14) 모델 14: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 구간 심박수의 상대표준편차 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(14) Model 14: Enter [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, relative standard deviation change in interval heart rate]. It is trained as a learning data set using the free T4 concentration corresponding to the first test day as the labeling value.

(15) 모델 15: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 깊은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(15) Model 15: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep rate of the first test day, interval deep sleep rate of the second test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(16) 모델 16: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제2 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(16) Model 16: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, standard deviation of the interval heart rate of the second test day] is used as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(17) 모델 17: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 구간 심박수의 왜도 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(17) Model 17: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep rate of the first test day, skewness change in interval heart rate] as input values and is learned as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(18) 모델 18: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(18) Model 18: Enter [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, interval average heart rate of the first test day] It is trained as a learning data set using the free T4 concentration corresponding to the first test day as the labeling value.

(19) 모델 19: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제1 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(19) Model 19: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, interval harmonic average heart rate of the first test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(20) 모델 20: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(20) Model 20: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, relative standard deviation of the interval heart rate of the first test day ] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(21) 모델 21: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(21) Model 21: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, kurtosis of the interval heart rate of the first test day] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

모델 13 내지 모델 21에서 이용된 입력값은 모델 6에서 이용된 입력값에 각각 대상의 성별, 구간 심박수의 상대표준편차 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 구간 심박수의 왜도 변화량, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 조화평균 심박수, 제1 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 첨도를 더 포함한다.The input values used in Models 13 to 21 are the input values used in Model 6, respectively, the subject's gender, relative standard deviation change in section heart rate, section deep sleep ratio on the second test day, and standard deviation of section heart rate on the second test day. , the amount of skewness change in the section heart rate, the section average heart rate of the first test day, the section harmonic average heart rate of the first test day, the relative standard deviation of the section heart rate of the first test day, and the kurtosis of the section heart rate of the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 34]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 34] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 6model 6 0.22710.2271 0.36630.3663 모델 13model 13 0.24510.2451 0.35160.3516 모델 14model 14 0.23940.2394 0.34990.3499 모델 15model 15 0.23530.2353 0.35280.3528 모델 16model 16 0.23430.2343 0.35160.3516 모델 17model 17 0.23420.2342 0.35140.3514 모델 18model 18 0.23170.2317 0.34740.3474 모델 19model 19 0.21620.2162 0.3480.348 모델 20model 20 0.2150.215 0.35150.3515 모델 21model 21 0.20010.2001 0.35040.3504

본 출원의 발명자들은 모델 6을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 19 내지 21을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 13 내지 18을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM과 함께 이용되는 대상의 성별, 구간 심박수의 상대표준편차 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 구간 심박수의 왜도 변화량, 제1 검사일의 구간 평균 심박수는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using Model 6, when predicting thyroid hormone concentration using Models 19 to 21, the Pearson Correlation for the gray zone is lowered, while Models 13 to 18 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Free T4 concentration corresponding to the test day, interval average heart rate change rate, gender of the subject used with interval REM on the first test day, relative standard deviation change in interval heart rate, interval deep sleep ratio on the second test day, interval on the second test day It was confirmed that the standard deviation of heart rate, skewness change in section heart rate, and section average heart rate on the first test day were significant variables that improved performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(22) 모델 22: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(22) Model 22: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate ] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(23) 모델 23: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 제2 검사일의 구간 깊은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(23) Model 23: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval deep of the second test day Sleep ratio] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(24) 모델 24: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 제1 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(24) Model 24: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval harmony of the first test day Average heart rate] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(25) 모델 25: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(25) Model 25: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate of the first test day Relative standard deviation of] is used as an input value, and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(26) 모델 26: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(26) Model 26: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, gender of the subject, interval heart rate of the first test day Kurtosis of] is used as an input value, and the free T4 concentration corresponding to the first inspection date is used as a labeling value.

모델 22 내지 모델 26에서 이용된 입력값은 모델 13에서 이용된 입력값에 각각 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제1 검사일의 구간 조화평균 심박수, 제1 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 심박수의 첨도를 더 포함한다.The input values used in Models 22 to 26 are the input values used in Model 13, respectively, including the skewness change in section heart rate, section deep sleep ratio on the second test day, section harmonic average heart rate on the first test day, and section on the first test day. It further includes the relative standard deviation of the heart rate and the kurtosis of the interval heart rate on the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 35]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 35] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 13model 13 0.24510.2451 0.35160.3516 모델 22model 22 0.25280.2528 0.34820.3482 모델 23model 23 0.24760.2476 0.35020.3502 모델 24model 24 0.22310.2231 0.34810.3481 모델 25model 25 0.21780.2178 0.34880.3488 모델 26model 26 0.21310.2131 0.34850.3485

본 출원의 발명자들은 모델 13을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 24 내지 26을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 22 내지 23을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별과 함께 이용되는 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using Model 13, when predicting thyroid hormone concentration using Models 24 to 26, the Pearson Correlation for the gray zone deteriorates, while Models 22 to 23 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 The free T4 concentration corresponding to the test day, the interval average heart rate change rate, the interval REM of the first test day, the skewness change of the interval heart rate used with the subject's gender, and the interval deep sleep ratio of the second test day are significant factors that lead to performance improvement. It was confirmed that it was a variable.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(27) 모델 27: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(27) Model 27: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , deep sleep ratio of the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(28) 모델 28: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(28) Model 28: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep rate of the first test day, subject's gender, skewness change in interval heart rate , kurtosis of the interval heart rate of the first test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(29) 모델 29: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(29) Model 29: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section average heart rate of the first test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(30) 모델 30: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제1 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(30) Model 30: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval harmonic average heart rate of the first test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

모델 27 내지 모델 30에서 이용된 입력값은 모델 22에서 이용된 입력값에 각각 제2 검사일의 구간 깊은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 평균 심박수를 더 포함한다.The input values used in Models 27 to 30 are the input values used in Model 22, respectively, the interval deep sleep ratio of the second test day, the kurtosis of the interval heart rate on the first test day, the interval average heart rate on the first test day, and the first test day. It also includes zone average heart rate.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 36]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 36] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 22model 22 0.25280.2528 0.34820.3482 모델 27model 27 0.25530.2553 0.34940.3494 모델 28model 28 0.22870.2287 0.34610.3461 모델 29model 29 0.22350.2235 0.34770.3477 모델 30model 30 0.21630.2163 0.34810.3481

본 출원의 발명자들은 모델 13을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 28 내지 30을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 27을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량과 함께 이용되는 제2 검사일의 구간 깊은 수면 비율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using Model 13, when predicting thyroid hormone concentration using Models 28 to 30, the Pearson Correlation for the gray zone was lowered, whereas using Model 27 It was confirmed that when predicting the thyroid hormone concentration, the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used the TSH concentration corresponding to the second test date and the second test date when predicting the thyroid hormone concentration. The corresponding free T4 concentration, section average heart rate change rate, section REM on the first test day, gender of the subject, and section deep sleep rate on the second test day used along with the skewness change in section heart rate are significant variables that improve performance. Confirmed.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(31) 모델 31: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(31) Model 31: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , the segment deep sleep ratio of the second test day, the standard deviation of the section heart rate of the second test day] as input values, and the free T4 concentration corresponding to the first test day as a labeling value.

(32) 모델 32: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 구간 조화평균 심박수의 변화율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(32) Model 32: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep rate of the second test day, rate of change of section harmonic average heart rate] as input values, and learned as a learning data set using the free T4 concentration corresponding to the first test day as a labeling value.

(33) 모델 33: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 얕은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(33) Model 33: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, section light sleep rate of the second test day] as input values, and the free T4 concentration corresponding to the first test day as a labeling value.

(34) 모델 34: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제1 검사일의 구간 심박수의 첨도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(34) Model 34: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, kurtosis of section heart rate of the first test day] as input values, and the free T4 concentration corresponding to the first test day as a labeling value.

(35) 모델 35: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제1 검사일의 구간 조화평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(35) Model 35: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, section harmonic average heart rate of the first test day] as input values, and the free T4 concentration corresponding to the first test day as a labeling value.

(36) 모델 36: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(36) Model 36: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, section average heart rate of the first test day] as input values, and the free T4 concentration corresponding to the first test day as a labeling value.

모델 31 내지 모델 36에서 이용된 입력값은 모델 27에서 이용된 입력값에 각각 제2 검사일의 구간 심박수의 표준편차, 구간 조화평균 심박수의 변화율, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 첨도, 제1 검사일의 구간 조화평균 심박수, 제1 검사일의 구간 평균 심박수를 더 포함한다.The input values used in Models 31 to 36 are the standard deviation of the section heart rate on the second test day, the rate of change of the section harmonic average heart rate, the section light sleep rate on the second test day, and the standard deviation of the section heart rate on the first test day in addition to the input values used in Model 27, respectively. It further includes kurtosis of the section heart rate, section harmonic average heart rate of the first test day, and section average heart rate of the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 37]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 37] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 22model 22 0.25530.2553 0.34940.3494 모델 31model 31 0.2620.262 0.34520.3452 모델 32model 32 0.25970.2597 0.34790.3479 모델 33model 33 0.25630.2563 0.340.34 모델 34model 34 0.23360.2336 0.34570.3457 모델 35model 35 0.22010.2201 0.34840.3484 모델 36model 36 0.21350.2135 0.3480.348

본 출원의 발명자들은 모델 22을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 34 내지 36을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 31 내지 33을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율과 함께 이용되는 제2 검사일의 구간 심박수의 표준편차, 구간 조화평균 심박수의 변화율, 제2 검사일의 구간 얕은 수면 비율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using model 22, when predicting thyroid hormone concentration using models 34 to 36, the Pearson Correlation for the gray zone deteriorates, while models 31 to 33 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Zone heart rate on the second test day used with the free T4 concentration corresponding to the test day, interval average heart rate change rate, interval REM on the first test day, subject's gender, skewness change in interval heart rate, and interval deep sleep ratio on the second test day. The standard deviation, rate of change of section harmonic average heart rate, and section light sleep rate on the second test day were confirmed to be significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(37) 모델 37: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(37) Model 37: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , segment deep sleep ratio of the second test day, standard deviation of the segment heart rate of the second test day, skewness of the segment heart rate of the first test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained with the training data set you use.

(38) 모델 38: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제2 검사일의 구간 얕은 수면 비율] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(38) Model 38: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, standard deviation of section heart rate of the second test day, section light sleep rate of the second test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained with a training data set that

(39) 모델 39: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 상대표준편차] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(39) Model 39: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , segment deep sleep ratio of the second test day, standard deviation of the segment heart rate of the second test day, relative standard deviation of the segment heart rate of the first test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained using a training data set.

(40) 모델 40: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(40) Model 40: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, standard deviation of section heart rate of the second test day, section average heart rate of the first test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value. Trained with the training data set.

(41) 모델 41: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 평균 심박수] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(41) Model 41: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , section deep sleep ratio of the second test day, standard deviation of section heart rate of the second test day, section average heart rate of the first test day] are used as input values, and the free T4 concentration corresponding to the first test day is used as a labeling value. Trained with the training data set.

모델 37 내지 모델 41에서 이용된 입력값은 모델 31에서 이용된 입력값에 각각 제1 검사일의 구간 심박수의 왜도, 제2 검사일의 구간 얕은 수면 비율, 제1 검사일의 구간 심박수의 상대표준편차, 제1 검사일의 구간 평균 심박수, 제1 검사일의 구간 평균 심박수를 더 포함한다.The input values used in Models 37 to 41 include the skewness of the zone heart rate on the first test day, the zone light sleep ratio on the second test day, the relative standard deviation of the zone heart rate on the first test day, and the input values used in Model 31, respectively. It further includes the section average heart rate of the first test day and the section average heart rate of the first test day.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 38]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 38] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 31model 31 0.2620.262 0.34520.3452 모델 37model 37 0.26650.2665 0.34280.3428 모델 38model 38 0.26640.2664 0.33860.3386 모델 39model 39 0.23620.2362 0.34680.3468 모델 40model 40 0.2360.236 0.34240.3424 모델 41model 41 0.22290.2229 0.34570.3457

본 출원의 발명자들은 모델 31을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 39 내지 41을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 37 내지 38을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차와 함께 이용되는 제1 검사일의 구간 심박수의 왜도, 제2 검사일의 구간 얕은 수면 비율은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using model 31, when predicting thyroid hormone concentration using models 39 to 41, the Pearson Correlation for the gray zone is lowered, while models 37 to 38 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Free T4 concentration corresponding to the test day, section average heart rate change rate, section REM on the first test day, gender of the subject, skewness change in section heart rate, section deep sleep ratio on the second test day, standard deviation of section heart rate on the second test day It was confirmed that the skewness of the heart rate in the section on the first test day and the rate of light sleep in the section on the second test day, used together, were significant variables that improved performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(42) 모델 42: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 제2 검사일의 구간 깊은 수면 비율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(42) Model 42: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , zone deep sleep ratio on the second test day, standard deviation of the zone heart rate on the second test day, skewness of the zone heart rate on the first test day, (free T4 concentration on the second test day x zone deep sleep rate on the second test day)] It is trained using a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

(43) 모델 43: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(43) Model 43: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate)] as input values. and is learned as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(44) 모델 44: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 TSH 농도 x 제2 검사일의 구간 깊은 수면 비율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(44) Model 44: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , zone deep sleep ratio of the second test day, standard deviation of the zone heart rate of the second test day, skewness of the zone heart rate of the first test day, (TSH concentration of the second test day x zone deep sleep rate of the second test day)]. It is trained as a learning data set using the free T4 concentration corresponding to the first test day as the labeling value.

(45) 모델 45: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 구간 심박수의 표준편차 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(45) Model 45: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , zone deep sleep ratio on the second test day, standard deviation of the zone heart rate on the second test day, skewness of the zone heart rate on the first test day, (standard deviation of the zone heart rate on the second test day / free T4 concentration on the second test day)] is used as an input value, and is trained as a learning data set using the free T4 concentration corresponding to the first inspection day as a labeling value.

(46) 모델 46: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 TSH 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(46) Model 46: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , zone deep sleep ratio of the second test day, standard deviation of the zone heart rate of the second test day, skewness of the zone heart rate of the first test day, (skewness of the zone heart rate of the first test day / TSH concentration of the second test day)] It is trained with a learning data set that is used as an input value and the free T4 concentration corresponding to the first inspection day is used as a labeling value.

모델 42 내지 모델 46에서 이용된 입력값은 모델 37에서 이용된 입력값에 각각 (제2 검사일의 free T4 농도 x 제2 검사일의 구간 깊은 수면 비율), (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제2 검사일의 TSH 농도 x 제2 검사일의 구간 깊은 수면 비율), (제2 검사일의 구간 심박수의 표준편차 / 제2 검사일의 free T4 농도), (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 TSH 농도)를 더 포함한다.The input values used in Models 42 to 46 are the input values used in Model 37, respectively (free T4 concentration on the second test day x interval deep sleep rate on the second test day), (free T4 concentration on the second test day x interval heart rate) skewness change), (TSH concentration on the second test day skewness of heart rate/TSH concentration on the second test day).

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 39]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 39] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 37model 37 0.26650.2665 0.34280.3428 모델 42model 42 0.26840.2684 0.34440.3444 모델 43model 43 0.26830.2683 0.34120.3412 모델 44model 44 0.26760.2676 0.34360.3436 모델 45model 45 0.24850.2485 0.34250.3425 모델 46model 46 0.24770.2477 0.34490.3449

본 출원의 발명자들은 모델 37을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 45 내지 46을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 42 내지 44를 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도와 함께 이용되는 (제2 검사일의 free T4 농도 x 제2 검사일의 구간 깊은 수면 비율), (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제2 검사일의 TSH 농도 x 제2 검사일의 구간 깊은 수면 비율)은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using model 37, when predicting thyroid hormone concentration using models 45 to 46, the Pearson Correlation for the gray zone deteriorates, while models 42 to 44 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Free T4 concentration corresponding to the test day, section average heart rate change rate, section REM on the first test day, gender of the subject, skewness change in section heart rate, section deep sleep ratio on the second test day, standard deviation of section heart rate on the second test day , (free T4 concentration on the second test day x interval deep sleep rate on the second test day), (free T4 concentration on the second test day TSH concentration on the second test day x deep sleep ratio on the second test day) was confirmed to be a significant variable that improves performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(47) 모델 47: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(47) Model 47: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Heart rate change rate / free T4 concentration on the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(48) 모델 48: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제2 검사일의 free T4 농도 / 제2 검사일의 TSH 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(48) Model 48: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (second test day) [free T4 concentration on the test day / TSH concentration on the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(49) 모델 49: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 심박수의 왜도의 변화량 x 제2 검사일의 구간 깊은 수면 비율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(49) Model 49: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval heart rate) It is learned as a learning data set that uses the change in skewness

(50) 모델 50: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제2 검사일의 구간 심박수 표준편차 / 제2 검사일의 free T4 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(50) Model 50: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (second test day) It is learned as a learning data set that uses the interval heart rate standard deviation of the test day / free T4 concentration of the second test day) as an input value, and uses the free T4 concentration corresponding to the first test day as a labeling value.

(51) 모델 51: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 구간 심박수의 표준편차)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(51) Model 51: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (first test day) Skewness of the section heart rate on the test day / standard deviation of the section heart rate on the second test day] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

(52) 모델 52: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 TSH 농도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(52) Model 52: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (first test day) Skewness of heart rate in the section of the test day / TSH concentration of the second test day)] is used as an input value, and the free T4 concentration corresponding to the first test day is used as a labeling value.

모델 47 내지 모델 52에서 이용된 입력값은 모델 43에서 이용된 입력값에 각각 (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 free T4 농도 / 제2 검사일의 TSH 농도), (구간 심박수의 왜도의 변화량 x 제2 검사일의 구간 깊은 수면 비율), (제2 검사일의 구간 심박수 표준편차 / 제2 검사일의 free T4 농도), (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 구간 심박수의 표준편차), (제1 검사일의 구간 심박수의 왜도 / 제2 검사일의 TSH 농도)를 더 포함한다.The input values used in Models 47 to 52 are the input values used in Model 43, respectively (interval average heart rate change rate / free T4 concentration on the second test day), (free T4 concentration on the second test day / TSH concentration on the second test day) ), (Change in skewness of section heart rate / Standard deviation of the zone heart rate on the second test day), (skewness of the zone heart rate on the first test day / TSH concentration on the second test day) are further included.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 40]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 40] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 43model 43 0.26830.2683 0.34120.3412 모델 47model 47 0.27460.2746 0.34130.3413 모델 48model 48 0.27350.2735 0.34340.3434 모델 49model 49 0.2710.271 0.3430.343 모델 50model 50 0.25490.2549 0.34220.3422 모델 51model 51 0.25350.2535 0.34360.3436 모델 52model 52 0.25240.2524 0.34350.3435

본 출원의 발명자들은 모델 43을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 50 내지 52을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 47 내지 49를 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량)와 함께 이용되는 (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 free T4 농도 / 제2 검사일의 TSH 농도), (구간 심박수의 왜도의 변화량 x 제2 검사일의 구간 깊은 수면 비율)은 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that compared to predicting thyroid hormone concentration using model 43, when predicting thyroid hormone concentration using models 50 to 52, the Pearson Correlation for the gray zone deteriorates, while models 47 to 49 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Free T4 concentration corresponding to the test day, section average heart rate change rate, section REM on the first test day, gender of the subject, skewness change in section heart rate, section deep sleep ratio on the second test day, standard deviation of section heart rate on the second test day , (interval average heart rate change rate / free T4 concentration on the second test day), (free T4 concentration on the second test day) used with the skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate) free T4 concentration / TSH concentration on the second test day), (change in skewness of segment heart rate x segment deep sleep ratio on the second test day) were confirmed to be significant variables that improve performance.

또한, 실험을 진행한 각 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.In addition, each thyroid hormone concentration prediction model that was tested was trained using the following training data set.

(53) 모델 53: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (구간 평균 심박수 변화율 / 제1 검사일의 구간 심박수의 왜도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(53) Model 53: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Heart rate change rate / free T4 concentration on the second test day), (section average heart rate change rate / skewness of section heart rate on the first test day)] are used as input values, and the free T4 concentration corresponding to the first test day is used as the labeling value. Trained with a training data set that

(54) 모델 54: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 구간 깊은 수면 비율 / 제1 검사일의 구간 심박수의 왜도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(54) Model 54: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Heart rate change rate / free T4 concentration on the second test day), (interval deep sleep rate on the second test day / skewness of the interval heart rate on the first test day)] are used as input values, and the free T4 concentration corresponding to the first test day is used as input values. Trained with a training data set used as labeling values.

(55) 모델 55: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제1 검사일의 구간 REM 수면 비율 / 제1 검사일의 구간 심박수의 왜도)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(55) Model 55: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Heart rate change rate / free T4 concentration on the second test day), (interval REM sleep rate on the first test day / skewness of the interval heart rate on the first test day)] are used as input values, and the free T4 concentration corresponding to the first test day is used as input values. Trained with a training data set used as labeling values.

(56) 모델 56: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 free T4 농도 x 구간 평균 심박수 변화율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(56) Model 56: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Learning to use heart rate change rate / free T4 concentration on the second test day), (free T4 concentration on the second test day Trained on a dataset.

(57) 모델 57: [제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM 수면 비율, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도), (제2 검사일의 free T4 농도 x 제1 검사일의 구간 REM 수면 비율)] 를 입력값으로 사용하고, 제1 검사일에 대응되는 free T4 농도를 라벨링값으로 사용하는 학습 데이터 세트로 학습됨.(57) Model 57: [TSH concentration corresponding to the second test day, free T4 concentration corresponding to the second test day, interval average heart rate change rate, interval REM sleep ratio of the first test day, subject's gender, skewness change in interval heart rate , interval deep sleep ratio on the second test day, standard deviation of the interval heart rate on the second test day, skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day x skewness change in interval heart rate), (interval average Heart rate change rate / free T4 concentration on the second test day), (free T4 concentration on the second test day Trained using a training data set.

모델 53 내지 모델 52에서 이용된 입력값은 모델 47에서 이용된 입력값에 각각 (구간 평균 심박수 변화율 / 제1 검사일의 구간 심박수의 왜도), (제2 검사일의 구간 깊은 수면 비율 / 제1 검사일의 구간 심박수의 왜도), (제1 검사일의 구간 REM 수면 비율 / 제1 검사일의 구간 심박수의 왜도), (제2 검사일의 free T4 농도 x 구간 평균 심박수 변화율), (제2 검사일의 free T4 농도 x 제1 검사일의 구간 REM 수면 비율)를 더 포함한다.The input values used in models 53 to 52 are the input values used in model 47, respectively (interval average heart rate change rate / skewness of interval heart rate on the first test day), (interval deep sleep ratio on the second test day / first test day) skewness of the interval heart rate), (interval REM sleep ratio on the first test day / skewness of the interval heart rate on the first test day), (free T4 concentration on the second test day x interval average heart rate change rate), (free on the second test day) T4 concentration x interval REM sleep rate on the first test day) is further included.

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 후술할 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning consisted of 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying two interpolation data generation, which will be described later.

갑상선 호르몬 농도 예측모델의 학습Learning of thyroid hormone concentration prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델의 테스트Test of thyroid hormone concentration prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the thyroid hormone concentration prediction model learned according to Experimental Example #19 was tested using the above-described test sets.

실험예#19에 따라 학습된 갑상선 호르몬 농도 예측모델의 Gray zone에 대한 Pearson Correlation, MAE (Mean Absolute Error)은 다음의 [표 41]과 같다.The Pearson Correlation and MAE (Mean Absolute Error) for the gray zone of the thyroid hormone concentration prediction model learned according to Experimental Example #19 are shown in [Table 41] below.

Gray zone
Pearson Correlation
Gray zone
Pearson Correlation
MAEM.A.E.
모델 47model 47 0.27460.2746 0.34130.3413 모델 53model 53 0.28330.2833 0.33850.3385 모델 54model 54 0.28330.2833 0.33970.3397 모델 55model 55 0.28260.2826 0.33950.3395 모델 56model 56 0.26730.2673 0.33950.3395 모델 57model 57 0.2630.263 0.34020.3402

본 출원의 발명자들은 모델 47을 이용하여 갑상선호르몬농도를 예측하였을 때와 비교할 때, 모델 56 내지 57을 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 저하되는 반면에, 모델 53 내지 55를 이용하여 갑상선호르몬농도를 예측하면 Gray zone에 대한 Pearson Correlation이 향상되는 것을 확인하였다.이를 토대로 확인할 때, 본 출원의 발명자들은 갑상선호르몬농도를 예측함에 있어서, 제2 검사일에 대응되는 TSH 농도, 제2 검사일에 대응되는 free T4 농도, 구간 평균 심박수 변화율, 제1 검사일의 구간 REM, 대상의 성별, 구간 심박수의 왜도 변화량, 제2 검사일의 구간 깊은 수면 비율, 제2 검사일의 구간 심박수의 표준편차, 제1 검사일의 구간 심박수의 왜도, (제2 검사일의 free T4 농도 x 구간 심박수의 왜도 변화량), (구간 평균 심박수 변화율 / 제2 검사일의 free T4 농도)와 함께 이용되는 (구간 평균 심박수 변화율 / 제1 검사일의 구간 심박수의 왜도), (제2 검사일의 구간 깊은 수면 비율 / 제1 검사일의 구간 심박수의 왜도), (제1 검사일의 구간 REM 수면 비율 / 제1 검사일의 구간 심박수의 왜도)는 성능 향상을 가져오는 유의미한 변수임을 확인하였다.The inventors of the present application found that, compared to predicting thyroid hormone concentration using model 47, when predicting thyroid hormone concentration using models 56 to 57, the Pearson Correlation for the gray zone deteriorates, while models 53 to 55 It was confirmed that when predicting thyroid hormone concentration using , the Pearson Correlation for the gray zone is improved. Based on this, the inventors of the present application used 2 Free T4 concentration corresponding to the test day, section average heart rate change rate, section REM on the first test day, gender of the subject, skewness change in section heart rate, section deep sleep ratio on the second test day, standard deviation of section heart rate on the second test day , skewness of the interval heart rate on the first test day, (free T4 concentration on the second test day Rate of change / skewness of zone heart rate on the first test day), (zone deep sleep rate on the second test day / skewness of the zone heart rate on the first test day), (zone REM sleep ratio on the first test day / zone heart rate on the first test day skewness) was confirmed to be a significant variable that improves performance.

갑상선 호르몬 농도를 예측하는 것은 갑상선 기능이상을 분류하는 것과 관련성이 있는 바, 갑상선 호르몬 농도 예측모델에서 유의미한 변수로 고려된 입력 변수들은 갑상선 기능이상 예측모델에 대한 유의미한 변수로 활용되는 것은 당연히 가능하다.Since predicting thyroid hormone concentration is related to classifying thyroid dysfunction, it is naturally possible that input variables considered as meaningful variables in the thyroid hormone concentration prediction model can be used as meaningful variables in the thyroid dysfunction prediction model.

9. 갑상선 기능 이상 예측 모델 및 갑상선 호르몬 예측 모델을 이용한 갑상선 기능이상증의 예측 방법9. Prediction method for thyroid dysfunction using thyroid dysfunction prediction model and thyroid hormone prediction model

본 출원의 출원인은 갑상선기능분류모델과 갑상선호르몬농도예측모델 사이의 연관성을 고려하여, 갑상선기능분류모델의 출력을 갑상선호르몬농도예측모델의 입력으로, 갑상선호르몬농도예측모델의 출력을 갑상선기능분류모델의 입력으로 사용한 유의미한 갑상선 기능 이상 또는 갑상선호르몬농도 예측 모듈을 확인하였고, 해당 모듈을 이하에서 개시한다.Considering the relationship between the thyroid function classification model and the thyroid hormone concentration prediction model, the applicant of this application has used the output of the thyroid function classification model as input to the thyroid hormone concentration prediction model and the output of the thyroid hormone concentration prediction model as the thyroid function classification model. A significant thyroid function abnormality or thyroid hormone concentration prediction module used as input was identified, and the corresponding module is disclosed below.

도 7은 갑상선 기능 이상 예측 모듈을 설명하기 위한 블록도이다.Figure 7 is a block diagram for explaining the thyroid dysfunction prediction module.

갑상선 기능 이상 예측 모듈(100)은 적어도 하나의 갑상선 기능 분류모델(1000) 및 적어도 하나의 갑상선호르몬농도 예측 모델(2000)을 포함할 수 있다.The thyroid function abnormality prediction module 100 may include at least one thyroid function classification model 1000 and at least one thyroid hormone concentration prediction model 2000.

갑상선 기능 분류모델(1000)은 전술한 실시예 1 내지 실시예 12 중 어느 하나의 방법으로 구현될 수 있다. 갑상선호르몬농도 예측 모델(2000)은 전술한 실시예 13 내지 실시예 19 중 어느 하나의 방법으로 구현될 수 있다. The thyroid function classification model 1000 may be implemented using any one of the above-described Examples 1 to 12. The thyroid hormone concentration prediction model (2000) can be implemented by any one of the above-described Examples 13 to 19.

갑상선 기능 이상 예측 모듈(100)이 둘 이상의 갑상선 기능 분류 모델(1000)을 포함할 때, 제1 갑상선 기능 분류 모델(1000)은 실시예 1 내지 실시예 11 중 어느 하나의 방법으로 구현되고, 제2 갑상선 기능 분류 모델(1000)은 실시예 12의 방법으로 구현되는 등, 서로 다른 형태로 학습된 갑상선 기능 분류 모델일 수 있다.When the thyroid function abnormality prediction module 100 includes two or more thyroid function classification models 1000, the first thyroid function classification model 1000 is implemented by any one of Examples 1 to 11, and 2 The thyroid function classification model 1000 may be a thyroid function classification model learned in different forms, such as implemented using the method of Example 12.

갑상선 기능 이상 예측 모듈(100)이 둘 이상의 갑상선호르몬농도 예측 모델(2000)을 포함할 때, 제1 갑상선호르몬농도 예측 모델(2000)은 실시예 13 내지 17 및 19 중 어느 하나의 방법으로 구현되고, 제2 갑상선호르몬농도 예측 모델(2000)은 실시예 18의 방법으로 구현되는 등, 서로 다른 형태로 학습된 갑상선 기능 예측 모델일 수 있다.When the thyroid dysfunction prediction module 100 includes two or more thyroid hormone concentration prediction models 2000, the first thyroid hormone concentration prediction model 2000 is implemented by any one of Examples 13 to 17 and 19, and , The second thyroid hormone concentration prediction model (2000) may be a thyroid function prediction model learned in different forms, such as implemented using the method of Example 18.

일 실시예에 따르면, 갑상선 기능 이상 예측 모듈(100)은 제1 갑상선 기능 분류모델(1000')의 출력값이 제1 갑상선호르몬농도 예측모델(2000')로 입력되는 형태일 수 있다.According to one embodiment, the thyroid function abnormality prediction module 100 may be configured such that the output value of the first thyroid function classification model 1000' is input to the first thyroid hormone concentration prediction model 2000'.

다른 실시예에 따르면, 갑상선 기능 이상 예측 모듈(100)은 제1 갑상선호르몬농도 예측모델(2000')의 출력값이 제1 갑상선 기능 분류 모델(1000')로 입력되는 형태일 수 있다.According to another embodiment, the thyroid function abnormality prediction module 100 may be configured such that the output value of the first thyroid hormone concentration prediction model 2000' is input to the first thyroid function classification model 1000'.

또 다른 실시예에 따르면, 갑상선 기능 이상 예측 모듈(100)은 제1 갑상선 기능 분류모델(1000')의 출력값이 제1 갑상선호르몬농도 예측모델(2000')로 입력되고, 제1 갑상선호르몬농도 예측모델(2000')이 제2 갑상선 기능 분류 모델(1000'')로 입력되는 형태일 수 있다.According to another embodiment, the thyroid function abnormality prediction module 100 inputs the output value of the first thyroid function classification model 1000' into the first thyroid hormone concentration prediction model 2000', and predicts the first thyroid hormone concentration. The model (2000') may be input as the second thyroid function classification model (1000'').

또 다른 실시예에 따르면, 갑상선 기능 이상 예측 모듈(100)은 제1 갑상선 기능 분류모델(1000')의 출력값이 제1 갑상선호르몬농도 예측모델(2000')로 입력되고, 제1 갑상선호르몬농도 예측모델(2000')이 제2 갑상선 기능 분류 모델(1000'')로 입력되고, 제2 갑상선 기능 분류 모델(1000'')의 출력값이 제3 갑상선 기능 분류 모델(1000''')로 입력되는 형태일 수 있다.According to another embodiment, the thyroid function abnormality prediction module 100 inputs the output value of the first thyroid function classification model 1000' into the first thyroid hormone concentration prediction model 2000', and predicts the first thyroid hormone concentration. The model (2000'') is input to the second thyroid function classification model (1000''), and the output value of the second thyroid function classification model (1000'') is input to the third thyroid function classification model (1000'''). It may be in the form.

이에 한정되지 않고, 갑상선 기능 이상 예측 모듈(100)은 N개의 갑상선 기능 분류모델(1000)과 M개의 갑상선호르몬농도 예측모델(2000)로 구성될 수 있으며(N과 M은 자연수임), 이에 대한 중복되는 설명은 생략하기로 한다.Not limited to this, the thyroid function abnormality prediction module 100 may be composed of N thyroid function classification models 1000 and M thyroid hormone concentration prediction models 2000 (N and M are natural numbers), and Redundant explanations will be omitted.

이 때, 갑상선 기능 분류 모델(1000)은 갑상선호르몬농도 예측모델(2000)의 출력값에 더하여, 심박수의 변화량, 심박수의 분포, 심박수간의 차이 등 입력값들을 함께 처리할 수 있다.At this time, the thyroid function classification model 1000 can process input values such as change in heart rate, heart rate distribution, and difference between heart rates in addition to the output value of the thyroid hormone concentration prediction model 2000.

이 때, 갑상선호르몬농도 예측 모델(2000)은 갑상선 기능 분류 모델(1000)의 출력값에 더하여, 심박수의 변화량, 심박수의 분포, 심박수간의 차이 등 입력값들을 함께 처리할 수 있다.At this time, the thyroid hormone concentration prediction model (2000) can process input values such as heart rate change, heart rate distribution, and difference between heart rates in addition to the output value of the thyroid function classification model (1000).

도 7을 참고하여 설명하면, 갑상선 기능 이상 예측 모듈(100)을 통해 갑상선 기능에 관한 정보 또는 갑상선호르몬 농도에 관한 정보를 출력할 때에, 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량을 갑상선 기능 분류모델(1000',1000'',,,) 및 갑상선호르몬농도 예측모델(2000',2000'',,,)에 입력하고, 갑상선 기능 분류모델(1000')은 상기 획득된 기준일의 호르몬 농도 및 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량을 처리하여 갑상선 기능 이상에 관한 값을 출력하고, 갑상선호르몬농도 예측모델(2000')은 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 갑상선 기능 분류모델(1000')에서 출력된 값을 처리하여 갑상선호르몬농도에 관한 값을 출력하고, 갑상선 기능 분류모델(1000'')은 상기 획득된 기준일의 호르몬 농도, 상기 획득된 대상일에 대응되는 구간 심박수들의 평균(대상일의 구간 평균 심박수)의 기준일에 대응되는 구간 심박수들의 평균(기준일의 구간 평균 심박수)에 대한 변화량 및 갑상선호르몬농도 예측모델(2000')에서 출력된 값을 처리하여 갑상선 기능에 관한 값을 출력할 수 있다.When explaining with reference to FIG. 7, when outputting information about thyroid function or information about thyroid hormone concentration through the thyroid function abnormality prediction module 100, the hormone concentration of the obtained reference date and the obtained target date are corresponded. The thyroid function classification model (1000', 1000'',,,) and the thyroid function classification model (1000', 1000'',,,) and the thyroid function classification model (1000', 1000'',,,) Input to the hormone concentration prediction model (2000',2000'',,,), and the thyroid function classification model (1000') is the average of the hormone concentration of the obtained reference day and the heart rate of the section corresponding to the obtained target date (target The change in the average of the section heart rates corresponding to the reference day (average heart rate of the section of the day) is processed to output a value related to thyroid dysfunction, and the thyroid hormone concentration prediction model (2000') is obtained above. Hormone concentration on the obtained reference day, the change in the average of section heart rates corresponding to the obtained target date (segment average heart rate on the target day) with respect to the average of section heart rates corresponding to the reference day (segment average heart rate on the reference day), and thyroid function classification model The value output from (1000'') is processed to output a value related to thyroid hormone concentration, and the thyroid function classification model (1000'') is used to calculate the hormone concentration of the obtained reference day and the heart rate interval corresponding to the obtained target date. Values related to thyroid function by processing the change in the average (average heart rate of the section on the target day) of the section heart rates corresponding to the reference day and the value output from the thyroid hormone concentration prediction model (2000') can be output.

갑상선 기능 이상 예측 모듈(100)에 있어서, 상기 갑상선 기능 분류모델(1000) 및 갑상선호르몬농도 예측모델(2000)은 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값을 처리할 수 있다. 상기 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 대상일에 대해 결정되는 제1 심박수 정보와 기준일에 대해 결정되는 제2 심박수 정보 간의 관계일 수 있다. 상기 관계는 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량 및 7) 상기 기준일에 대응되는 구간 심박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence로 이루어진 그룹에서 선택된 하나 또는 그들의 조합일 수 있으나, 이에 한정되지 않는다.In the thyroid function abnormality prediction module 100, the thyroid function classification model 1000 and the thyroid hormone concentration prediction model 2000 may process values related to heart rate information on the target date and heart rate information on the reference date. The values related to the heart rate information of the target date and the heart rate information of the reference date may be a relationship between the first heart rate information determined for the target date and the second heart rate information determined for the reference date. The relationship is 1) the amount of change in the average of section heart rates of the target day with respect to the reference date, 2) the rate of change in the average of section heart rates of the target day with respect to the reference day, 3) the standard deviation of the target date with respect to the reference date. Amount of change, 4) Amount of change in the relative standard deviation of the target date with respect to the base date, 5) Amount of change in skewness of the target date with respect to the base date, 6) Amount of change in kurtosis of the target date with respect to the base date, and 7) the above It may be one selected from the group consisting of section heart rates corresponding to the reference date and JS Divergence between section heart rates corresponding to the target date, or a combination thereof, but is not limited thereto.

갑상선 기능 이상 예측 모듈(100)에 있어서, 갑상선 기능 분류모델(1000)이 처리하는 값과 갑상선호르몬농도 예측모델(2000)이 처리하는 값은 동일하지 않을 수 있다. 예를 들어, 갑상선 기능 분류모델(1000)은 [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), (제2 검사일에 대응되는 free T4의 농도/제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도]을 처리하고, 갑상선호르몬농도 예측모델(2000)은 [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도]을 처리할 수 있다.In the thyroid function abnormality prediction module 100, the value processed by the thyroid function classification model 1000 and the value processed by the thyroid hormone concentration prediction model 2000 may not be the same. For example, the thyroid function classification model 1000 is [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (1st test day) Relative standard deviation of section heart rates corresponding to the test day - relative standard deviation of section heart rates corresponding to the second test day), (skewness of section heart rates corresponding to the second test day - skewness of section heart rates corresponding to the first test day) , (Kurtosis of the zone heart rates corresponding to the second test day - Kurtosis of the zone heart rates corresponding to the first test day), (JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day) , (Concentration of free T4 corresponding to the second test day/JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day), Concentration of free T4 corresponding to the second test day, 2 Concentration of TSH corresponding to the test day] is processed, and the thyroid hormone concentration prediction model (2000) is [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / second test day corresponding section average heart rate), (relative standard deviation of section heart rates corresponding to the first test day - relative standard deviation of section heart rates corresponding to the second test day), (skewness of section heart rates corresponding to the second test day - 1st Skewness of the interval heart rates corresponding to the test day), (Kurtosis of the interval heart rates corresponding to the second test day - Kurtosis of the interval heart rates corresponding to the first test day), (Kurtosis of the interval heart rates corresponding to the second test day and the first test day JS Divergence between corresponding interval heart rates), concentration of free T4 corresponding to the second test day, concentration of TSH corresponding to the second test day] can be processed.

갑상선 기능 이상 예측 모듈(100)에 있어서, 갑상선 기능 분류모델(1000)간 처리하는 값이 동일하지 않을 수 있다. 예를 들어, 제1 갑상선 기능 분류모델(1000)은 [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도]를 처리하고, 제N 갑상선 기능 분류모델(1000)은 [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), 제1 검사일에 대응되는 추정 호르몬 농도에 관한 값, 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도]를 처리할 수 있다.In the thyroid function abnormality prediction module 100, the values processed between the thyroid function classification models 1000 may not be the same. For example, the first thyroid function classification model 1000 is [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), 2 Concentration of free T4 corresponding to test day, TSH concentration corresponding to second test day] is processed, and the N thyroid function classification model (1000) is [((((section average heart rate corresponding to first test day - second test day) (area average heart rate corresponding to) / section average heart rate corresponding to the second test day), value regarding the estimated hormone concentration corresponding to the first test day, concentration of free T4 corresponding to the second test day, TSH corresponding to the second test day concentration] can be processed.

도 8 및 9는 학습시킬 때의 임상데이터를 선별함으로써 항진증, 정상 및/또는 저하증에 특화된 갑상선 기능 분류모델(1000) 및/또는 갑상선호르몬농도 예측모델(2000)을 포함하는 갑상선 기능 이상 예측 모듈(100)을 설명하기 위한 블록도이다.8 and 9 show a thyroid function abnormality prediction module (including a thyroid function classification model (1000) and/or a thyroid hormone concentration prediction model (2000) specialized for hyperactivity, normal and/or hypothyroidism by selecting clinical data during learning ( This is a block diagram to explain 100).

도 8을 참조하면, 갑상선 기능 분류모델(1000)은 임상 데이터를 항진 또는 항진아님으로 라벨링하여 학습시킨 갑상선 기능 항진증 분류모델 및 임상 데이터를 저하 또는 저하아님으로 라벨링하여 학습시킨 갑상선 기능 저하증 분류모델을 포함할 수 있다.Referring to FIG. 8, the thyroid function classification model 1000 is a hyperthyroidism classification model learned by labeling clinical data as hyper or non-hyperthyroid, and a hypothyroidism classification model learned by labeling clinical data as low or not low. It can be included.

이 때, 갑상선 기능 이상 예측 모듈(100)에서, 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값을 포함하는 입력 데이터 세트가 갑상선 기능 항진증 분류모델(1000') 및 갑상선 기능 저하증 분류모델(1000'')에 입력되고, 두 분류모델의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값이 갑상선호르몬농도예측모델(2000)'에 입력될 수 있다. 갑상선호르몬농도예측모델(2000')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선 기능 항진증 분류모델(1000''') 및 갑상선 기능 저하증 분류모델(1000'''')에 입력될 수 있다. 갑상선 기능 항진증 분류모델(1000''')의 출력값과 갑상선 기능 저하증 분류모델(1000'''')의 출력값을 기초로 대상의 갑상선 기능 이상에 관한 정보가 획득될 수 있다.At this time, in the thyroid function abnormality prediction module 100, an input data set containing values related to heart rate information of the target day and heart rate information of the reference day is used as a hyperthyroidism classification model (1000') and a hypothyroidism classification model (1000 ''), and the output values of the two classification models, the heart rate information on the target day, and the heart rate information on the reference day can be input into the 'Thyroid hormone concentration prediction model (2000)'. The output value of the thyroid hormone concentration prediction model (2000'), the heart rate information on the target day, and the heart rate information on the reference day are the hyperthyroidism classification model (1000''') and the hypothyroidism classification model (1000''''). can be entered. Information about the subject's thyroid dysfunction can be obtained based on the output value of the hyperthyroidism classification model (1000''') and the output value of the hypothyroidism classification model (1000'''').

또는, 갑상선호르몬농도예측모델(2000')의 출력값을 토대로 산출된 호르몬농도가 갑상선항진증에 대응되는 경우, 갑상선호르몬농도예측모델(2000')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선기능항진증분류모델(1000''')에 입력되어 그를 토대로 갑상선 기능 이상에 관한 정보가 획득되고, 갑상선호르몬농도예측모델(2000')의 출력값을 토대로 산출된 호르몬농도가 갑상선저하증에 대응되는 경우, 갑상선호르몬농도예측모델(2000')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선기능저하증분류모델(1000'''')에 입력되어 그를 토대로 갑상선 기능 이상에 관한 정보가 획득될 수 있다. Alternatively, if the hormone concentration calculated based on the output value of the thyroid hormone concentration prediction model (2000') corresponds to hyperthyroidism, the output value of the thyroid hormone concentration prediction model (2000'), the heart rate information on the target date, and the heart rate information on the reference date The related values are entered into the hyperthyroidism classification model (1000'''), based on which information on thyroid dysfunction is obtained, and the hormone concentration calculated based on the output value of the thyroid hormone concentration prediction model (2000'') is used to determine hypothyroidism. In case of correspondence, the output value of the thyroid hormone concentration prediction model (2000'), the heart rate information of the target day, and the heart rate information of the reference day are input into the hypothyroidism classification model (1000'''') and based on this, the thyroid function abnormality Information about can be obtained.

도 9을 참조하면, 갑상선 기능 분류모델(1000)은 임상 데이터를 항진 또는 항진아님으로 라벨링하여 학습시킨 갑상선 기능 항진증 분류모델 및 임상 데이터를 저하 또는 저하아님으로 라벨링하여 학습시킨 갑상선 기능 저하증 분류모델을 포함할 수 있다. 갑상선호르몬농도 예측모델(2000)은 임상 데이터 중 항진증에 대응되는 학습 데이터 세트만을 이용한 갑상선호르몬농도 예측모델(항진), 임상 데이터 중 정상에 대응되는 학습 데이터 세트만을 이용한 갑상선호르몬농도 예측모델(정상) 및 임상 데이터 중 저하증에 대응되는 학습 데이터 세트만을 이용한 갑상선호르몬농도 예측모델(저하)을 포함할 수 있다.Referring to FIG. 9, the thyroid function classification model 1000 is a hyperthyroidism classification model learned by labeling clinical data as hyper or not hyper and a hypothyroidism classification model learned by labeling clinical data as low or not low. It can be included. Thyroid hormone concentration prediction model (2000) is a thyroid hormone concentration prediction model using only the learning data set corresponding to hyperactivity among clinical data (hyperthyroidism), and a thyroid hormone concentration prediction model using only the learning data set corresponding to normal among clinical data (normal). And it may include a thyroid hormone concentration prediction model (lower) using only the learning data set corresponding to hypothyroidism among clinical data.

이 때, 갑상선 기능 이상 예측 모듈(100)에서, 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값을 포함하는 입력 데이터 세트가 갑상선호르몬농도 예측모델(2000')에 입력되고, 갑상선호르몬농도 예측모델의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값이 갑상선 기능 항진증 분류모델(1000') 및 갑상선 기능 저하증 분류모델(1000'')에 입력될 수 있다.At this time, in the thyroid function abnormality prediction module 100, an input data set containing values related to the heart rate information of the target day and the heart rate information of the reference day is input to the thyroid hormone concentration prediction model 2000', and the thyroid hormone concentration is predicted. The output value of the model, the heart rate information on the target day, and the heart rate information on the reference day can be input into the hyperthyroidism classification model (1000') and the hypothyroidism classification model (1000'').

갑상선 기능 항진증 분류모델(1000') 및 갑상선 기능 저하증 분류모델(1000'')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선호르몬농도 예측모델(항진)(2000''), 갑상선호르몬농도 예측모델(정상)(2000''') 및 갑상선호르몬농도 예측모델(저하)(2000'''')에 입력될 수 있다. 갑상선호르몬농도 예측모델(항진)(2000''), 갑상선호르몬농도 예측모델(정상)(2000''') 및 갑상선호르몬농도 예측모델(저하)(2000'''')의 출력값을 기초로 대상의 갑상선호르몬농도에 관한 정보를 획득할 수 있다.The output values of the hyperthyroidism classification model (1000') and the hypothyroidism classification model (1000''), as well as the heart rate information on the target day and the heart rate information on the reference day, are calculated from the thyroid hormone concentration prediction model (hyperthyroidism) (2000''). , can be entered into the thyroid hormone concentration prediction model (normal) (2000''') and the thyroid hormone concentration prediction model (decreased) (2000''''). Target based on the output values of thyroid hormone concentration prediction model (enhancement) (2000''), thyroid hormone concentration prediction model (normal) (2000'''), and thyroid hormone concentration prediction model (decreased) (2000''''). You can obtain information about thyroid hormone concentration.

또는, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값을 토대로 산출한 갑상선 기능이 갑상선 기능 항진증인 경우, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선호르몬농도 예측모델(항진)(2000'')에 입력되어 갑상선호르몬농도에 관한 정보가 획득될 수 있고, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값을 토대로 산출한 갑상선 기능이 갑상선 기능 저하증인 경우, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선호르몬농도 예측모델(저하)(2000'''')에 입력되어 갑상선호르몬농도에 관한 정보가 획득될 수 있고, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값을 토대로 산출한 갑상선 기능이 정상인 경우, 갑상선 기능 항진증 분류모델(1000') 및/또는 갑상선 기능 저하증 분류모델(1000'')의 출력값과 대상일의 심박수 정보 및 기준일의 심박수 정보에 관한 값은 갑상선호르몬농도 예측모델(정상)(2000''')에 입력되어 갑상선호르몬농도에 관한 정보가 획득될 수 있다.Alternatively, if the thyroid function calculated based on the output values of the hyperthyroidism classification model (1000') and/or the hypothyroidism classification model (1000'') is hyperthyroidism, the hyperthyroidism classification model (1000'') and/ Alternatively, the output value of the hypothyroidism classification model (1000''), the heart rate information on the target day, and the heart rate information on the reference day are input into the thyroid hormone concentration prediction model (hyperjin) (2000'') to provide information on thyroid hormone concentration. can be obtained, and if the thyroid function calculated based on the output values of the hyperthyroidism classification model (1000') and/or the hypothyroidism classification model (1000'') is hypothyroidism, the hyperthyroidism classification model (1000'') ') and/or the output values of the hypothyroidism classification model (1000''), the heart rate information on the target day, and the heart rate information on the reference day are input into the thyroid hormone concentration prediction model (low) (2000''''). If information on thyroid hormone concentration can be obtained and the thyroid function calculated based on the output values of the hyperthyroidism classification model (1000'') and/or the hypothyroidism classification model (1000'') is normal, hyperthyroidism is classified. The output values of the model (1000') and/or the hypothyroidism classification model (1000''), the heart rate information on the target day, and the heart rate information on the reference day are in the thyroid hormone concentration prediction model (normal) (2000'''). Information on thyroid hormone concentration can be obtained by entering the information.

도 7 내지 9를 통해, 기능분류-농도예측-기능분류 모델 또는 농도예측-기능분류-농도예측 모델로 총 3개의 계층이 있는 실시예를 기준으로 설명하였지만, 이는 설명의 편의를 위한 하나의 예시일 뿐, N개의 기능분류 모델 및 M개의 농도예측 모델을 포함하도록 여러 계층이 있는 순환 형태로 구현될 수 있음은 물론이다.7 to 9, the description is based on an embodiment in which there are a total of three layers as a functional classification-concentration prediction-functional classification model or a concentration prediction-functional classification-concentration prediction model, but this is an example for convenience of explanation. Of course, it can be implemented in a circular form with multiple layers to include N functional classification models and M concentration prediction models.

학습방법Learning method

갑상선 기능 이상 예측 모델 및 갑상선 호르몬 예측 모델이 전술한 여러 학습 데이터 세트에 의해 생성될 수 있음을 충분히 설명한바 있다.It has been sufficiently explained that a thyroid function abnormality prediction model and a thyroid hormone prediction model can be generated by the various learning data sets described above.

이하에서는, N개의 기능분류 모델 및 M개의 농도예측 모델을 포함하도록 형성된 갑상선 기능 이상 또는 갑상선호르몬농도 예측 모듈의 학습 순서에 대한 구체적인 실시예를 설명하기로 한다.Hereinafter, a specific embodiment of the learning sequence of the thyroid function abnormality or thyroid hormone concentration prediction module formed to include N functional classification models and M concentration prediction models will be described.

설명의 편의를 위해, 도 7에서 도시된 갑상선 기능 분류 모델(1000')의 출력값을 갑상선호르몬농도예측모델(2000')의 입력값으로, 갑상선호르몬농도예측모델(2000')의 출력값을 갑상선기능분류모델(1000'')의 입력값으로 하는 갑상선 기능 이상 예측 모듈을 이용하여 설명하기로 한다.For convenience of explanation, the output value of the thyroid function classification model (1000') shown in Figure 7 is used as the input value of the thyroid hormone concentration prediction model (2000'), and the output value of the thyroid hormone concentration prediction model (2000') is used as the thyroid function value. This will be explained using the thyroid function abnormality prediction module used as the input value of the classification model (1000'').

갑상선 기능 분류 모델(1000')의 출력값을 갑상선호르몬농도예측모델(2000')의 입력값으로, 갑상선호르몬농도예측모델(2000')의 출력값을 갑상선기능분류모델(1000'')의 입력값으로 하는 갑상선 기능 이상 예측 모듈의 경우, 갑상선 기능 분류 모델(1000')을 가장 먼저 학습시키고, 갑상선호르몬농도예측모델(2000')를 학습시킨 후 갑상선기능분류모델(1000'')를 학습시킬 수 있다.The output value of the thyroid function classification model (1000') is used as the input value of the thyroid hormone concentration prediction model (2000'), and the output value of the thyroid hormone concentration prediction model (2000') is used as the input value of the thyroid function classification model (1000''). In the case of the thyroid function abnormality prediction module, the thyroid function classification model (1000') can be trained first, the thyroid hormone concentration prediction model (2000') can be trained, and then the thyroid function classification model (1000'') can be trained. .

구체적인 예를 들어 설명하면, 갑상선 기능 분류모델 및 갑상선호르몬농도예측모델이 모두 i) 기준일 심박수와 대상일의 차이값, ii) 기준일의 호르몬 농도를 입력값으로 하여 대상일의 갑상선 기능 또는 갑상선 호르몬 농도를 출력하도록 학습되는 경우, To explain with a specific example, both the thyroid function classification model and the thyroid hormone concentration prediction model use i) the difference between the heart rate on the reference day and the target date, ii) the hormone concentration on the reference day as input values, and the thyroid function or thyroid hormone concentration on the target date. When learned to output,

갑상선기능분류모델(1000')은, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷으로 생성된 학습 데이터 세트들을 이용하여 학습될 수 있다.The thyroid function classification model (1000') is [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), thyroid hormone concentration corresponding to the second inspection day, and diagnosis corresponding to the first inspection day. Results (advanced or not) can be learned using training data sets generated in the same format.

이 후, 동일한 학습 데이터 세트를 기초로, 갑상선호르몬농도예측모델(2000')의 학습 데이터 세트를 생성할 수 있다.Afterwards, based on the same learning data set, a learning data set for the thyroid hormone concentration prediction model (2000') can be created.

갑상선호르몬농도예측모델(2000')의 학습 데이터 세트는, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 본 학습 데이터 세트에 포함된 심박수의 차이 및 갑상선 호르몬 농도를 갑상선기능분류모델(1000')에 입력하여 획득한 출력값, 제1 검사일에 대응되는 갑상선호르몬농도]와 같은 포맷으로 생성될 수 있다.The training data set of the thyroid hormone concentration prediction model (2000') is [(section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day), thyroid hormone concentration corresponding to the second inspection day, and this learning The output value obtained by inputting the difference in heart rate and thyroid hormone concentration included in the data set into the thyroid function classification model (1000'), thyroid hormone concentration corresponding to the first test date] may be generated in the same format.

이 후, 동일한 학습 데이터 세트를 기초로, 갑상선기능분류모델(1000'') 의 학습 데이터 세트를 생성할 수 있다.Afterwards, based on the same learning data set, a learning data set of the thyroid function classification model (1000'') can be created.

갑상선기능분류모델(1000'') 의 학습 데이터 세트는, [(제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, i) 본 학습 데이터 세트에 포함된 심박수의 차이, ii) 갑상선 호르몬 농도 및 iii) 본 학습 데이터 세트에 포함된 심박수의 차이, 갑상선 호르몬 농도를 갑상선기능분류모델(1000')에 입력하여 획득한 출력값을 갑상선호르몬농도예측모델(2000')에 입력하여 획득한 출력값, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]와 같은 포맷으로 생성될 수 있다.The training data set of the thyroid function classification model (1000'') is [(section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, i) The output values obtained by inputting the difference in heart rate included in this learning data set, ii) thyroid hormone concentration, and iii) the difference in heart rate and thyroid hormone concentration included in this learning data set into the thyroid function classification model (1000') are used as thyroid The output value obtained by inputting it into the hormone concentration prediction model (2000') and the diagnosis result (hyperactivity or not hyperactivity) corresponding to the first test date can be generated in the same format.

위와 같은 기술적 사상을 적용하여, 이를 토대로 N개의 기능 분류 모델 및 M개의 농도 예측 모델을 포함하는 모듈의 학습 데이터 세트를 생성하고 학습시키는 것은 당업자에 의해 충분히 용이하게 실시될 수 있는바, 구체적인 설명은 생략하기로 한다.Applying the above technical idea, generating and learning a training data set of a module including N functional classification models and M concentration prediction models based on this can be easily performed by a person skilled in the art. The detailed description is provided. Decided to omit it.

실험예#20.Experimental Example #20.

학습 데이터 세트들의 준비Preparation of training data sets

총 126명의 환자를 대상으로 수집한 총 464개의 임상 데이터 세트들이 사용되었다.A total of 464 clinical data sets collected from a total of 126 patients were used.

전술한 갑상선 기능 이상 예측 모듈의 형태로 실험이 진행되었다.An experiment was conducted in the form of the thyroid function abnormality prediction module described above.

갑상선 기능 이상 예측 모델에 포함된 갑상선 기능 분류모델의 경우,In the case of the thyroid function classification model included in the thyroid function abnormality prediction model,

[((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 진단결과][((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of section heart rates corresponding to the first test day - second 2 Relative standard deviation of section heart rates corresponding to the test day), (skewness of section heart rates corresponding to the second test day - skewness of section heart rates corresponding to the first test day), (kurtosis of section heart rates corresponding to the second test day - kurtosis of the zone heart rates corresponding to the first test day), (JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day), concentration of free T4 corresponding to the second test day, TSH concentration corresponding to the second test date, diagnostic result corresponding to the first test day]

갑상선 기능 이상 예측 모델에 포함된 갑상선호르몬농도 예측모델의 경우,In the case of the thyroid hormone concentration prediction model included in the thyroid function abnormality prediction model,

[((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 free T4의 농도]를 학습 데이터 세트로하여 학습이 진행되었다.[((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (relative standard deviation of section heart rates corresponding to the first test day - second 2 Relative standard deviation of section heart rates corresponding to the test day), (skewness of section heart rates corresponding to the second test day - skewness of section heart rates corresponding to the first test day), (kurtosis of section heart rates corresponding to the second test day - kurtosis of the zone heart rates corresponding to the first test day), (JS Divergence between the zone heart rates corresponding to the second test day and the zone heart rates corresponding to the first test day), concentration of free T4 corresponding to the second test day, Learning was conducted using [the concentration of TSH corresponding to the second test day and the concentration of free T4 corresponding to the first test day] as the learning data set.

특히, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In particular, each thyroid function prediction module tested was as follows.

(1) 갑상선항진증분류모델(1) Hyperthyroidism classification model

(2) 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델(2) Hyperthyroidism classification model, thyroid hormone prediction model connected to the output value, hyperthyroidism classification model connected to the output value

(3) 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델(3) Hyperthyroidism classification model, thyroid hormone prediction model connected to the output value, hyperthyroidism classification model connected to the output value, thyroid hormone prediction model connected to the output value, hyperthyroidism classification model connected to the output value

(4) 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델, 그 출력값과 연결된 갑상선호르몬예측모델, 그 출력값과 연결된 갑상선항진증분류모델(4) Hyperthyroidism classification model, thyroid hormone prediction model connected to the output value, hyperthyroidism classification model connected to the output value, thyroid hormone prediction model connected to the output value, hyperthyroidism classification model connected to the output value, thyroid hormone connected to the output value Prediction model, hyperthyroidism classification model connected to the output value

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 126명의 환자로부터 확보한 464개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,416개였다. 이 중 일부는 테스트 데이터 세트로 사용되었으며, 실험예#18에 따라 학습된 갑상선기능항진증 예측모델의 정확도 등은 테스트 세트들을 이용하여 확인되었다.At this time, the clinical data used for learning was 464 data obtained from 126 patients, and the total number of learning data sets created from this was 1,416. Some of these were used as test data sets, and the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #18 was confirmed using the test sets.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였고, 갑상선호르몬농도 예측모델은 extra trees을 이용하였다.The hyperthyroidism prediction model used a light gradient boosting machine, and the thyroid hormone concentration prediction model used extra trees.

실험예#20에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표 42]와 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #20 are shown in [Table 42] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV (1)(One) 83.4183.41 79.7379.73 84.1084.10 48.7648.76 95.6395.63 (2)(2) 87.2887.28 90.5490.54 86.6786.67 56.3056.30 97.9797.97 (3)(3) 86.6486.64 89.1989.19 86.1586.15 55.0055.00 97.6797.67 (4)(4) 86.6486.64 89.1989.19 86.1586.15 55.0055.00 97.6797.67

본 출원의 발명자들은, 갑상선기능항진증을 예측함에 있어서, 둘 이상의 갑상선기능항진증 분류모델 및 하나 이상의 갑상선호르몬농도 예측모델을 전술한 방식으로 연결하여 사용하였을 때 정확도, 민감도, 특이도, PPV 및 NPV 모두의 성능 향상이 있음을 확인할 수 있었다. 또한, 갑상선기능항진증에서 확인되는 성능 향상은 갑상선기능저하증을 예측함에 있어서도 유사하게 발현될 것으로 보인다.The inventors of the present application have found that in predicting hyperthyroidism, when using two or more hyperthyroidism classification models and one or more thyroid hormone concentration prediction models by connecting them in the above-described manner, the accuracy, sensitivity, specificity, PPV and NPV are all It was confirmed that there was an improvement in performance. In addition, the performance improvement observed in hyperthyroidism is expected to be similar in predicting hypothyroidism.

또한, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In addition, each thyroid function prediction module tested was as follows.

(1) 갑상선호르몬농도 예측모델(1) Thyroid hormone concentration prediction model

(2) 갑상선호르몬농도 예측모델, 그 출력값과 연결된 갑상선항진증 분류모델, 그 출력값과 연결된 갑상선호르몬농도 예측모델(2) Thyroid hormone concentration prediction model, hyperthyroidism classification model connected to the output value, and thyroid hormone concentration prediction model connected to the output value.

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 146명의 환자로부터 확보한 542개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,728개였다. 이 중 일부는 테스트 데이터 세트로 사용되었으며, 실험예#18에 따라 학습된 갑상선호르몬농도 예측모델의 MAE 등은 테스트 데이터 세트들을 이용하여 확인되었다.At this time, the clinical data used for learning was 542 data obtained from 146 patients, and the total number of learning data sets created from this was 1,728. Some of these were used as test data sets, and the MAE of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was confirmed using the test data sets.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였고, 갑상선호르몬농도 예측모델은 extra trees을 이용하였다.The hyperthyroidism prediction model used a light gradient boosting machine, and the thyroid hormone concentration prediction model used extra trees.

실험예#20에 따라 학습된 갑상선호르몬예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 43]와 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the thyroid hormone prediction model learned according to Experimental Example #20 are shown in [Table 43] below.

MAEM.A.E. MSEMSE MAPEMAPE (1)(One) 0.24210.2421 0.18300.1830 0.17220.1722 (2) (2) 0.24060.2406 0.17820.1782 0.17140.1714

본 출원의 발명자들은, 갑상선호르몬농도를 예측함에 있어서, 둘 이상의 갑상선호르몬농도 예측모델 및 하나 이상의 갑상선기능항진증 분류모델을 전술한 방식으로 연결하여 사용하였을 때 MAE, MSE 및 MAPE 의 성능 향상이 있음을 확인할 수 있었다. The inventors of the present application have found that, in predicting thyroid hormone concentration, there is an improvement in the performance of MAE, MSE, and MAPE when two or more thyroid hormone concentration prediction models and one or more hyperthyroidism classification models are connected and used in the above-described manner. I was able to confirm.

또한, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In addition, each thyroid function prediction module tested was as follows.

(1) 갑상선항진증 분류모델, 그 출력값을 입력으로 받는 갑상선호르몬농도 예측모델(1) Hyperthyroidism classification model and thyroid hormone concentration prediction model that receives the output value as input

(2) 갑상선항진증 분류모델 및 갑상선저하증 분류모델, 그 출력값을 입력으로 받는 갑상선호르몬농도 예측모델 (2) Hyperthyroidism classification model, hypothyroidism classification model, and thyroid hormone concentration prediction model that receives the output value as input

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 146명의 환자로부터 확보한 542개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,728개였다. 이 중 일부는 테스트 데이터 세트로 사용되었으며, 실험예#18에 따라 학습된 갑상선호르몬농도 예측모델의 MAE 등은 테스트 세트들을 이용하여 확인되었다.At this time, the clinical data used for learning was 542 data obtained from 146 patients, and the total number of learning data sets created from this was 1,728. Some of these were used as test data sets, and the MAE of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was confirmed using the test sets.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였고, 갑상선호르몬농도 예측모델은 extra trees을 이용하였다.The hyperthyroidism prediction model used a light gradient boosting machine, and the thyroid hormone concentration prediction model used extra trees.

실험예#20에 따르는 갑상선호르몬예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 44]와 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the thyroid hormone prediction model according to Experimental Example #20 are shown in [Table 44] below.

MAE(전체)MAE (all) MAE(항진)MAE (extension) MSEMSE MAPEMAPE (1)(One) 0.24040.2404 0.56080.5608 0.17800.1780 0.17120.1712 (2) (2) 0.24030.2403 0.55250.5525 0.17700.1770 0.17200.1720

본 출원의 발명자들은, 갑상선호르몬농도를 예측함에 있어서, MAE(항진)으로 본다면, 즉, 항진에 해당하는 여러 심박수들을 기반으로 갑상선호르몬농도를 예측하는 정확도를 평가한다면 갑상선기능항진증 분류모델의 출력값만을 이용하는 것에 비해 갑상선기능항진증 분류모델 및 갑상선기능저하증 분류모델을 함께 이용하는 경우 MAE 의 성능 향상이 있음을 확인할 수 있었지만,The inventors of the present application, when predicting thyroid hormone concentration, view it as MAE (hyperthyroidism), that is, when evaluating the accuracy of predicting thyroid hormone concentration based on various heart rates corresponding to hyperthyroidism, only the output value of the hyperthyroidism classification model is used. Compared to using the hyperthyroidism classification model and the hypothyroidism classification model together, it was confirmed that the performance of MAE improved.

MAE(전체)로 본다면, 즉, 항진/정상/저하에 해당하는 여러 심박수들을 기반으로 갑상선호르몬농도를 예측하는 정확도를 평가한다면 갑상선기능항진증분류모델 및 갑상선기능저하증분류모델을 모두 이용하는 경우와 갑상선기능항진증분류모델만을 이용하는 경우 사이에 유의미한 MAE의 성능 향상이 있다고 보기는 어려웠다.If you look at MAE (overall), that is, if you evaluate the accuracy of predicting thyroid hormone concentration based on various heart rates corresponding to hyper/normal/decreased heart rates, the case of using both hyperthyroidism classification model and hypothyroidism classification model and thyroid function It was difficult to see that there was a significant improvement in MAE performance between cases using only the hyperactivity disorder classification model.

따라서, 두가지 실험 결과 중 어느 한쪽이 절대적으로 좋다고 평가할 수 없으나, 본 출원의 발명자들은 실험을 진행하였던 학습 데이터들 중 '저하'에 대응되는 임상 데이터들이 다소 부족했던 점을 고려할 때, 학습데이터가 충분히 확보되고나면 갑상선기능항진증 분류모델의 출력값만을 이용하는 것에 비해 갑상선기능항진증 분류모델 및 갑상선기능저하증 분류모델을 함께 이용하는 경우에 더 나은 성능을 보여줄 것으로 기대하고 있다.Therefore, one of the two experimental results cannot be evaluated as absolutely good, but the inventors of this application believe that the learning data is sufficient, considering that the clinical data corresponding to 'deterioration' was somewhat lacking among the learning data for which the experiment was conducted. Once secured, it is expected to show better performance when using the hyperthyroidism classification model and hypothyroidism classification model together compared to using only the output value of the hyperthyroidism classification model.

또한, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In addition, each thyroid function prediction module tested was as follows.

(1) 갑상선항진증 분류모델, 그 출력값을 입력으로 받는 갑상선호르몬농도 예측모델(1) Hyperthyroidism classification model and thyroid hormone concentration prediction model that receives the output value as input

(2) 갑상선항진증 분류모델, 그 출력값을 입력으로 받는 갑상선호르몬농도 예측모델(항진) (2) Hyperthyroidism classification model, thyroid hormone concentration prediction model (hyperthyroidism) that receives the output value as input

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 146명의 환자로부터 확보한 542개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,728개였다. 이 중 일부는 테스트 데이터 세트로 사용되었으며, 실험예#18에 따라 학습된 갑상선호르몬농도 예측모델의 MAE 등은 테스트 세트들을 이용하여 확인되었다.At this time, the clinical data used for learning was 542 data obtained from 146 patients, and the total number of learning data sets created from this was 1,728. Some of these were used as test data sets, and the MAE of the thyroid hormone concentration prediction model learned according to Experimental Example #18 was confirmed using the test sets.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였고, 갑상선호르몬농도 예측모델은 extra trees을 이용하였다.The hyperthyroidism prediction model used a light gradient boosting machine, and the thyroid hormone concentration prediction model used extra trees.

실험예#20에 따르는 갑상선호르몬예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 45]와 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the thyroid hormone prediction model according to Experimental Example #20 are shown in [Table 45] below.

MAE(전체)MAE (all) MAE(항진)MAE (extension) MSEMSE MAPEMAPE (1)(One) 0.24040.2404 0.56080.5608 0.17800.1780 0.17120.1712 (2) (2) 0.27550.2755 0.50150.5015 0.20420.2042 0.20130.2013

마찬가지로, 본 출원의 발명자들은, 갑상선호르몬농도를 예측함에 있어서,MAE(전체)로 본다면, 즉, 항진/정상/저하에 해당하는 여러 심박수들을 기반으로 학습된 갑상선호르몬농도를 이용하는 것이 항진증에 특화된 갑상선호르몬농도를 이용하는 것에 비해 MAE의 성능 향상이 있는 것처럼 보이기도 했지만, MAE(항진)으로 본다면, 즉, 항진에 해당하는 여러 심박수들을 기반으로 갑상선호르몬농도를 예측하는 정확도를 평가한다면 갑상선기능항진증 분류모델의 출력값이 '항진'에 대응되는 경우 갑상선호르몬농도 예측모델(항진)을 사용하는 경우의 MAE의 성능 향상이 있음을 확인할 수 있었다. Likewise, when predicting thyroid hormone concentration, the inventors of the present application believe that when looking at MAE (overall), that is, using the thyroid hormone concentration learned based on various heart rates corresponding to hyperactivity/normality/decreased hyperactivity disorder, the thyroid gland specialized for hyperactivity disorder There seemed to be an improvement in the performance of MAE compared to using hormone concentration, but if it is viewed as MAE (hyperthyroidism), that is, if the accuracy of predicting thyroid hormone concentration based on various heart rates corresponding to hyperthyroidism is evaluated, the hyperthyroidism classification model It was confirmed that there was an improvement in the performance of MAE when using the thyroid hormone concentration prediction model (hyperactivity) when the output value corresponded to 'hyperactivity'.

이에 대하여, 두가지 실험 결과 중 어느 한쪽이 절대적으로 좋다고 평가할 수 없으나, 본 출원의 발명자들은 임상 데이터가 충분히 확보되고나면 갑상선호르몬농도 예측모델(항진)을 사용하는 경우 전반적으로 좋을 수 있지만, 현재는 실험을 진행하였던 학습 데이터들 중 '항진'에 대응되는 임상 데이터들도 그들만으로 충분한 연관성을 확인하기에는 부족한 수였기 때문에 이러한 결과가 나왔을 것으로 예측하고 있다. In this regard, one of the two experimental results cannot be evaluated as absolutely good, but the inventors of the present application believe that using the thyroid hormone concentration prediction model (hyperactivity) may be good overall once sufficient clinical data is obtained, but currently, the experimental results It is predicted that this result was obtained because the number of clinical data corresponding to 'hyperactivity' among the learning data from which the process was conducted was insufficient to confirm a sufficient correlation.

실험예#21.Experimental Example #21.

학습 데이터 세트들의 준비Preparation of training data sets

총 360명의 환자를 대상으로 수집한 총 1,504개의 임상 데이터 세트들이 사용되었다.A total of 1,504 clinical data sets collected from a total of 360 patients were used.

전술한 갑상선 기능이상 예측 모듈의 형태로 실험이 진행되었다,An experiment was conducted in the form of the thyroid dysfunction prediction module described above.

구체적으로, 갑상선 호르몬 농도 예측모델의 출력값과 연결된 갑상선 기능이상 예측모델에 대한 실험이 진행되었다. 특히, 갑상선 호르몬 농도 예측모델은 free T4 농도를 예측하기 위한 예측하기 위한 예측모델로 학습되었고, 갑상선 기능이상 예측모델은 갑상선 기능항진증 예측모델로 학습되었다.Specifically, an experiment was conducted on a thyroid dysfunction prediction model linked to the output value of the thyroid hormone concentration prediction model. In particular, the thyroid hormone concentration prediction model was learned as a prediction model for predicting free T4 concentration, and the thyroid dysfunction prediction model was learned as a hyperthyroidism prediction model.

갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다. The thyroid hormone concentration prediction model was trained using the following training data set.

제1 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 구간 얕은 수면 비율, 제1 검사일에 대응되는 free T4 농도]First thyroid hormone concentration prediction model: [Thyroid hormone concentration corresponding to the second test day, (average heart rate in the section corresponding to the first test day - average heart rate in the section corresponding to the second test day), light sleep in the section corresponding to the second test day Ratio, free T4 concentration corresponding to the first test day]

제2 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 조화평균 심박수 - 제2 검사일에 대응되는 구간 조화평균 심박수), 제1 검사일에 대응되는 free T4 농도]Prediction model for second thyroid hormone concentration: [Thyroid hormone concentration corresponding to the second test day, (average heart rate of the section corresponding to the first test day - average heart rate of the section corresponding to the second test day), (harmony of the section corresponding to the first test day) Average heart rate - section harmonic average heart rate corresponding to the second test day), free T4 concentration corresponding to the first test day]

제3 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), ((제1 검사일에 대응되는 구간 조화평균 심박수 - 제2 검사일에 대응되는 구간 조화평균 심박수) / 제2 검사일에 대응되는 구간 조화평균 심박수), 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차, 제1 검사일에 대응되는 구간 심박수들의 첨도, 제2 검사일에 대응되는 구간 얕은 수면 비율, (제2 검사일에 대응되는 free T4 농도 x 제1 검사일에 대응되는 구간 심박수들의 첨도), ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), 제1 검사일에 대응되는 free T4 농도]Third thyroid hormone concentration prediction model: [Thyroid hormone concentration corresponding to the second test day, (average heart rate in the section corresponding to the first test day - average heart rate in the section corresponding to the second test day), ((section corresponding to the first test day Average heart rate - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), ((section harmonic average heart rate corresponding to the first test day - section harmonic average heart rate corresponding to the second test day) / interval harmonic mean heart rate corresponding to the second test day), relative standard deviation of the interval heart rates corresponding to the second test day, kurtosis of the interval heart rates corresponding to the first test day, interval light sleep ratio corresponding to the second test day, (second Free T4 concentration corresponding to the test day relative standard deviation of heart rates), free T4 concentration corresponding to the first test day]

제4 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 왜도 - 제2 검사일에 대응되는 구간 심박수들의 왜도), 제2 검사일에 대응되는 구간 심박수들의 표준편차, 제1 검사일에 대응되는 구간 심박수들의 왜도, 제2 검사일에 대응되는 구간 깊은 수면 비율, 제1 검사일에 대응되는 구간 REM 수면 비율, 대상의 성별, (제2 검사일에 대응되는 free T4 농도 x (제1 검사일에 대응되는 구간 심박수들의 왜도 - 제2 검사일에 대응되는 구간 심박수들의 왜도)), (((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 free T4 농도), (제2 검사일에 대응되는 구간 깊은 수면 비율 / 제1 검사일에 대응되는 구간 심박수들의 왜도), 제1 검사일에 대응되는 free T4 농도]Fourth thyroid hormone concentration prediction model: [Thyroid hormone concentration corresponding to the second test day, ((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average corresponding to the second test day heart rate), (skewness of section heart rates corresponding to the first test day - skewness of section heart rates corresponding to the second test day), standard deviation of section heart rates corresponding to the second test day, standard deviation of section heart rates corresponding to the first test day Skewness, interval deep sleep ratio corresponding to the second test day, interval REM sleep ratio corresponding to the first test day, gender of the subject, (free T4 concentration corresponding to the second test day x (interval heart rate corresponding to the first test day) Skewness - Skewness of section heart rates corresponding to the second test day)), ((((segment average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day ) / free T4 concentration corresponding to the second test day), (interval deep sleep ratio corresponding to the second test day / skewness of interval heart rates corresponding to the first test day), free T4 concentration corresponding to the first test day]

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다.The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

사용된 제2 검사일에 대응되는 갑상선 호르몬 농도에는 free T4 농도 및 TSH 농도가 포함되었다.Thyroid hormone concentrations corresponding to the second test day used included free T4 concentration and TSH concentration.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 상술한 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning was 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying the two interpolation data generation described above.

갑상선 기능이상 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다. The thyroid dysfunction prediction model was trained using the following training data set.

(1) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)](1) [((section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day) / section average heart rate corresponding to the second inspection day), (standard deviation of section heart rates corresponding to the first inspection day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, first estimated thyroid hormone concentration, diagnosis corresponding to the first test date Results (hyperactivity or not hyperactivity)]

(2) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제2 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)](2) [((section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day) / section average heart rate corresponding to the second inspection day), (standard deviation of section heart rates corresponding to the first inspection day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, second estimated thyroid hormone concentration, diagnosis corresponding to the first test date Results (hyperactivity or not hyperactivity)]

(3) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제3 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)](3) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, third estimated thyroid hormone concentration, diagnosis corresponding to the first test day Results (hyperactivity or not hyperactivity)]

(4) [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제4 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)](4) [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, fourth estimated thyroid hormone concentration, diagnosis corresponding to the first test day Results (hyperactivity or not hyperactivity)]

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculating the average of the obtained resting heart rates.

사용된 제2 검사일에 대응되는 갑상선 호르몬 농도에는 free T4 농도 및 TSH 농도가 포함되었다.Thyroid hormone concentrations corresponding to the second test day used included free T4 concentration and TSH concentration.

제1 추정 갑상선 호르몬 농도는 제1 갑상선 호르몬 농도 예측모델로부터 획득되며, 제2 추정 갑상선 호르몬 농도는 제2 갑상선 호르몬 농도 예측모델로부터 획득되며, 제3 추정 갑상선 호르몬 농도는 제3 갑상선 호르몬 농도 예측모델로부터 획득되며, 제4 추정 갑상선 호르몬 농도는 제4 갑상선 호르몬 농도 예측모델로부터 획득되었다.The first estimated thyroid hormone concentration is obtained from the first thyroid hormone concentration prediction model, the second estimated thyroid hormone concentration is obtained from the second thyroid hormone concentration prediction model, and the third estimated thyroid hormone concentration is obtained from the third thyroid hormone concentration prediction model. and the fourth estimated thyroid hormone concentration was obtained from the fourth thyroid hormone concentration prediction model.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 상술한 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning was 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying the two interpolation data generation described above.

갑상선 호르몬 농도 예측모델 및 갑상선 기능이상 예측모델의 학습Learning of thyroid hormone concentration prediction model and thyroid dysfunction prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델이 먼저 학습된 후, Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 기능이상 예측모델이 학습되었다.A thyroid hormone concentration prediction model was first learned, and then a thyroid dysfunction prediction model was learned using the above-mentioned learning data sets using a light gradient boosting machine.

실험예#21에 따르는 갑상선 기능이상 예측 모듈의 AUC(Area Under the ROC Curve), 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value), F1 Score는 다음의 [표 46]과 같다.The AUC (Area Under the ROC Curve), accuracy, sensitivity, specificity, PPV (Positive Predictive Value), NPV (Negative Predictive Value), and F1 Score of the thyroid dysfunction prediction module according to Experimental Example #21 are as follows [Table 46] ] is the same as

AUCAUC 정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV F1 scoreF1 score (1)(One) 0.88810.8881 82.2182.21 81.7081.70 82.2982.29 40.5140.51 96.8296.82 0.54160.5416 (2)(2) 0.88170.8817 80.7380.73 80.4380.43 80.7880.78 38.1838.18 96.5596.55 0.51780.5178 (3)(3) 0.88500.8850 82.0582.05 81.7081.70 82.1082.10 40.2540.25 96.8196.81 0.5390.539 (4)(4) 0.88610.8861 82.3282.32 81.7081.70 82.4182.41 40.6840.68 96.8396.83 0.5430.543

본 출원의 발명자들은, 갑상선 기능이상을 예측함에 있어서, 갑상선 호르몬 농도 예측모델을 전술한 방식으로 연결하여 사용하였을 때 AUC, 정확도, 민감도, 특이도, PPV, NPV 및 F1 score 모두의 성능 향상이 있음을 확인할 수 있었다.The inventors of the present application found that when predicting thyroid dysfunction, the performance of AUC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score was improved when the thyroid hormone concentration prediction model was connected and used in the above-described manner. was able to confirm.

10. 갑상선 기능 이상 예측에 있어서, 학습 데이터의 증강(Data Augmentation)10. Data Augmentation of learning data in predicting thyroid dysfunction

갑상선 기능 분류 모델 및/또는 갑상선호르몬농도 예측 모델을 구현함에 있어서, 임상데이터를 확보하기 어려운 이유에 대해서는 전술한바 있다. The reason why it is difficult to secure clinical data when implementing a thyroid function classification model and/or a thyroid hormone concentration prediction model has been described above.

이에 따라, 보다 정확도 높은 갑상선 기능 분류 모델 및/또는 갑상선호르몬농도 예측 모델을 획득하기 위해서는 학습 데이터의 증강이 필요하다.Accordingly, augmentation of learning data is necessary to obtain a more accurate thyroid function classification model and/or thyroid hormone concentration prediction model.

실측데이터를 이용해서 혈액 검사 수치를 보간할 때, 보간 방법, 보간할 데이터의 수 및 보간 대상 데이터의 특정 등을 수행할 수 있다.When interpolating blood test values using actual measurement data, you can specify the interpolation method, the number of data to be interpolated, and the data to be interpolated.

보간 방법의 경우, Linear interpolation, Cubic interpolation, 혈액검사 결과와 심박수의 관계를 고려한 기계학습 알고리즘 등이 사용될 수 있으나 이에 한정되지 않는다.In the case of interpolation methods, linear interpolation, cubic interpolation, machine learning algorithms that consider the relationship between blood test results and heart rate, etc. may be used, but are not limited to these.

보간할 데이터의 수의 경우, 실측 데이터 사이의 간격(day gap)에 따라 결정될 수 있고, 또는, 고정된 개수만큼 보간할 수 있으나, 이에 한정되지 않는다. 일 예로, 실측 데이터 사이의 간격과 무관하게 제1 실측데이터와 제2 실측데이터 사이에 2개의 보간 데이터가 생성될 수 있다.The number of data to be interpolated may be determined depending on the day gap between measured data, or a fixed number may be interpolated, but is not limited thereto. As an example, two pieces of interpolation data may be generated between the first measured data and the second measured data regardless of the interval between the measured data.

보간 대상 데이터는 실측 데이터의 전부를 대상으로 할 수 있으나, 실측 데이터 사이의 간격(day gap)이 미리 정해진 기간(예, 60일)을 도과하는 경우 보간 대상 데이터로 사용되지 않을 수 있다.The interpolation target data may be all of the actual measured data, but if the day gap between the measured data exceeds a predetermined period (e.g., 60 days), it may not be used as the interpolation target data.

실험예#22.Experimental Example #22.

학습 데이터 세트들의 준비Preparation of training data sets

총 126명의 환자를 대상으로 수집한 총 464개의 임상 데이터 세트들이 사용되었다.A total of 464 clinical data sets collected from a total of 126 patients were used.

전술한 갑상선항진증 분류모델로 실험이 진행되었다.An experiment was conducted using the hyperthyroidism classification model described above.

갑상선항진증 분류모델은, [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 상대 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), (제2 검사일에 대응되는 구간 심박수들의 왜도 - 제1 검사일에 대응되는 구간 심박수들의 왜도), (제2 검사일에 대응되는 구간 심박수들의 첨도 - 제1 검사일에 대응되는 구간 심박수들의 첨도), (제2 검사일에 대응되는 구간 심박수들과 제1 검사일에 대응되는 구간 심박수들 사이의 JS Divergence), 제2 검사일에 대응되는 free T4의 농도, 제2 검사일에 대응되는 TSH의 농도, 제1 검사일에 대응되는 진단결과]를 학습 데이터 세트로 하여 학습이 진행되었다.The hyperthyroidism classification model is: [((section average heart rate corresponding to the first inspection day - section average heart rate corresponding to the second inspection day) / section average heart rate corresponding to the second inspection day), (section average heart rate corresponding to the first inspection day relative standard deviation of the interval heart rates corresponding to the second test day), (skewness of the interval heart rates corresponding to the second test day - skewness of the interval heart rates corresponding to the first test day), (skewness of the interval heart rates corresponding to the first test day), (skewness of the interval heart rates corresponding to the second test day) Kurtosis of the corresponding interval heart rates - kurtosis of the interval heart rates corresponding to the first test day), (JS Divergence between the interval heart rates corresponding to the second test day and the interval heart rates corresponding to the first test day), corresponding to the second test day Learning was conducted using the concentration of free T4, TSH concentration corresponding to the second test date, and diagnosis result corresponding to the first test day] as the learning data set.

특히, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In particular, each thyroid function prediction module tested was as follows.

(1) 갑상선항진증분류모델(실측데이터 간 interpolation 미진행)(1) Hyperthyroidism classification model (interpolation between actual data was not performed)

(2) 갑상선항진증분류모델(실측데이터 간 1개의 interpolation 데이터 생성)(2) Hyperthyroidism classification model (generating one interpolation data between actual measurement data)

(3) 갑상선항진증분류모델(실측데이터 간 2개의 interpolation 데이터 생성)(3) Hyperthyroidism classification model (generating two interpolation data between actual measurement data)

(4) 갑상선항진증분류모델(실측데이터 간 3개의 interpolation 데이터 생성)(4) Hyperthyroidism classification model (generating three interpolation data between actual measurement data)

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 126명의 환자로부터 확보한 464개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,416개(interpolation 이전)였고, 이들 중 일부를 테스트 데이터 세트로 사용하였다.At this time, the clinical data used for learning was 464 data obtained from 126 patients, and the total number of learning data sets created from this was 1,416 (before interpolation), and some of these were used as test data sets. used.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였다. The hyperthyroidism prediction model used a light gradient boosting machine.

실험예#22에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표 47]과 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #22 are shown in [Table 47] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV (1)(One) 84.3284.32 72.7372.73 86.0586.05 43.8443.84 95.4795.47 (2)(2) 82.3382.33 82.4382.43 82.3182.31 46.9246.92 96.1196.11 (3)(3) 84.4884.48 87.8487.84 83.8583.85 50.7850.78 97.3297.32 (4)(4) 81.9081.90 86.4986.49 81.0381.03 46.3846.38 96.9396.93

본 출원의 발명자들은, 혈액 검사 수치를 보간하여 학습 데이터를 사용할 때 민감도, PPV 및 NPV 모두의 성능 향상이 있음을 확인할 수 있었다.또한, 2회의 interpolation을 진행하였을 때에, interpolation을 하지 않았거나 1회 또는 3회의 interpolation을 진행하였을 때에 비해 정확도, 민감도, 특이도, PPV 및 NPV 모두의 성능 향상이 있음을 확인할 수 있었다. 따라서, 실측 혈액 검사 수치 사이에 2개의 혈액 검사 수치를 보간했을 때에 가장 정확도가 높은 것을 확인하였다.The inventors of this application were able to confirm that there was an improvement in the performance of both sensitivity, PPV, and NPV when using learning data by interpolating blood test values. In addition, when interpolation was performed twice, interpolation was not performed or interpolation was performed once. Alternatively, it was confirmed that there was an improvement in the performance of accuracy, sensitivity, specificity, PPV, and NPV compared to when interpolation was performed three times. Therefore, it was confirmed that the highest accuracy was achieved when two blood test values were interpolated between the actual blood test values.

또한, 갑상선기능항진증에서 확인되는 성능 향상은 갑상선기능저하증을 분류하고, 갑상선호르몬농도를 예측함에 있어서도 유사하게 발현될 것으로 보인다. 또한, 실험을 진행한 각각의 갑상선 기능 예측 모듈은 아래와 같았다.In addition, the performance improvement observed in hyperthyroidism is expected to be similar in classifying hypothyroidism and predicting thyroid hormone concentration. In addition, each thyroid function prediction module tested was as follows.

(1) 갑상선항진증분류모델(Cubic)(1) Hyperthyroidism classification model (Cubic)

(2) 갑상선항진증분류모델(Linear)(2) Hyperthyroidism classification model (Linear)

(3) 갑상선호르몬농도예측모델(Cubic)(3) Thyroid hormone concentration prediction model (Cubic)

(4) 갑상선호르몬농도예측모델(Linear)(4) Thyroid hormone concentration prediction model (Linear)

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 10일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 10 days based on the test date, and calculating the average of the obtained resting heart rates.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 146명의 환자로부터 확보한 542개의데이터들이 활용되었으며, 이로부터 생성된 학습 데이터 세트들의 개수는 총 1,728개였다. 이 중 일부는 테스트 데이터 세트로 사용되었으며, 실험예#22에 따라 학습된 갑상선기능항진증 예측모델의 정확도 및 갑상선호르몬농도 예측모델의 MAE 등은 테스트 세트들을 이용하여 확인되었다.At this time, the clinical data used for learning was 542 data obtained from 146 patients, and the total number of learning data sets created from this was 1,728. Some of these were used as test data sets, and the accuracy of the hyperthyroidism prediction model learned according to Experimental Example #22 and the MAE of the thyroid hormone concentration prediction model were confirmed using the test sets.

갑상선기능항진증 예측모델은, Light gradient boosting machine을 이용하였고, 갑상선호르몬농도 예측모델은 extra trees을 이용하였다. The hyperthyroidism prediction model used a light gradient boosting machine, and the thyroid hormone concentration prediction model used extra trees.

실험예#22에 따르는 갑상선기능항진증 예측모델의 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value)는 다음의 [표 48]과 같다.The accuracy, sensitivity, specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value) of the hyperthyroidism prediction model according to Experimental Example #22 are shown in [Table 48] below.

정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV (1)(One) 79.5279.52 87.5087.50 77.9777.97 43.5043.50 96.9996.99 (2)(2) 80.6380.63 85.2385.23 77.9777.97 43.5043.50 96.9996.99

실험예#19에 따르는 갑상선호르몬예측모델의 MSE (Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error)는 다음의 [표 49]과 같다.The MSE (Mean Squared Error), MAE (Mean Absolute Error), and MAPE (Mean Absolute Percentage Error) of the thyroid hormone prediction model according to Experimental Example #19 are shown in [Table 49] below.

MAEM.A.E. MSEMSE MAPEMAPE (3)(3) 0.24230.2423 0.17930.1793 0.17280.1728 (4) (4) 0.24210.2421 0.18300.1830 0.17220.1722

본 출원의 발명자들은, 혈액 검사 수치를 보간하여 학습 데이터로 사용할 때에 갑상선기능항진증예측모델과 갑상선호르몬예측모델 모두 성능 향상이 있음을 확인하였다. 또한, 보간 방법에 관계없이 데이터 보간을 통한 성능 향상이 있을 것으로 실험 결과를 기반으로 강하게 추정하고 있다.분류모델의 경우 갑상선기능항진증예측모델을 이용하여 실험을 진행하였으나, 갑상선기능저하증예측모델의 경우에도 유사한 경향을 보일 것으로 판단된다.The inventors of this application confirmed that both the hyperthyroidism prediction model and the thyroid hormone prediction model improved performance when blood test values were interpolated and used as learning data. In addition, it is strongly assumed based on the experimental results that there will be performance improvement through data interpolation regardless of the interpolation method. In the case of the classification model, the experiment was conducted using the hyperthyroidism prediction model, but in the case of the hypothyroidism prediction model, the experiment was conducted using the hyperthyroidism prediction model. It is expected to show a similar trend.

11. 복수의 기준일에 기초한 갑상선기능이상증의 예측 방법11. Prediction method for thyroid dysfunction based on multiple reference dates

본 출원의 출원인은 선택될 수 있는 기준일이 복수개인 대상에 대하여, 복수의 기준일 각각에 기초하여 복수의 갑상선 기능이상 예측값을 획득하고, 복수의 예측값들의 평균값에 기초하여 상기 대상의 갑상선 기능이상증에 대한 결과를 획득하는 유의미한 갑상선 기능이상 예측 방법을 확인하였고, 해당 방법을 이하에서 개시한다.The applicant of the present application obtains a plurality of thyroid dysfunction prediction values based on each of the plurality of reference dates for a subject with a plurality of reference dates that can be selected, and determines the thyroid dysfunction of the subject based on the average value of the plurality of prediction values. A method for predicting significant thyroid dysfunction that obtains results was identified, and the method is disclosed below.

도 10은 갑상선 기능이상 예측 방법을 설명하기 위한 도면(40)이다.Figure 10 is a diagram 40 for explaining a method for predicting thyroid dysfunction.

도 10을 참고하면, 대상(subject)에 대한 기준일(410, 420)은 두 개 있을 수 있다. 예를 들어, 대상에 대한 갑상선호르몬농도를 알고 있는 제1 기준일(410) 및 제2 기준일(420)이 존재할 수 있다.Referring to FIG. 10, there may be two reference dates 410 and 420 for the subject. For example, there may be a first reference date (410) and a second reference date (420) for which the thyroid hormone concentration for the subject is known.

구간 심박수들의 전처리 결과(411, 421) 및 호르몬 농도(412, 422)는 두 개의 기준일(410, 420) 각각에 대응될 수 있다. 구간 심박수들의 전처리 결과 및 호르몬 농도와 관련해서는, 4. 갑상선기능이상증의 예측방법 및 도 5에서 상술한 바 있으므로 중복 설명은 생략한다.The preprocessing results of section heart rates (411, 421) and hormone concentrations (412, 422) may correspond to each of the two reference days (410, 420). Regarding the preprocessing results of the heart rates and hormone concentrations, this has been described in detail in 4. Prediction method for thyroid dysfunction and FIG. 5, so redundant explanation will be omitted.

대상에 대한 갑상선 기능이상증에 대한 결과(450)를 획득하기 위해 적어도 하나의 갑상선 기능이상 예측모델(430)이 이용될 수 있다. 갑상선 기능이상 예측모델(430)은 상술한 갑상선 기능이상 예측모델 중 적어도 하나의 학습방법으로 구현될 수 있으며, 학습방법에 대한 중복 설명은 생략한다.At least one thyroid dysfunction prediction model 430 may be used to obtain a result 450 for thyroid dysfunction for the subject. The thyroid dysfunction prediction model 430 can be implemented using at least one learning method among the thyroid dysfunction prediction models described above, and redundant description of the learning method will be omitted.

도 10을 참고하면, 두 개의 기준일(410, 420)이 있는 대상에 대한 갑상선 기능이상증에 대한 결과(450)를 획득하기 위해, 제1 예측값(441)은 대상일(400)에 대응되는 구간 심박수들의 전처리 결과(401), 제1 기준일(410)에 대응되는 구간 심박수들의 전처리 결과(411), 제1 기준일(410)에 대응되는 호르몬 농도(412) 및 갑상선 기능이상 예측모델(430)을 이용하여 획득되고, 제2 예측값(442)은 대상일(400)에 대응되는 구간 심박수들의 전처리 결과(401), 제2 기준일(420)에 대응되는 구간 심박수들의 전처리 결과(421), 제2 기준일(420)에 대응되는 호르몬 농도(422) 및 갑상선 기능이상 예측모델(430)을 이용하여 획득될 수 있다. 예측값을 획득하는 구체적인 내용은 4. 갑상선기능이상증의 예측방법 및 도 5에서 상술한 바 있으므로 중복 설명은 생략한다.Referring to FIG. 10, in order to obtain a result 450 for thyroid dysfunction for a subject with two reference days 410 and 420, the first predicted value 441 is the interval heart rate corresponding to the target date 400. Preprocessing results (401), preprocessing results (411) of section heart rates corresponding to the first reference date (410), hormone concentration (412) corresponding to the first reference date (410), and thyroid dysfunction prediction model (430) It is obtained, and the second predicted value 442 is the preprocessing result 401 of the section heart rates corresponding to the target date 400, the preprocessing result 421 of the section heart rates corresponding to the second reference date 420, and the second reference date ( It can be obtained using the hormone concentration 422 and the thyroid dysfunction prediction model 430 corresponding to 420). The specific details of obtaining the predicted value are described in detail in 4. Prediction method for thyroid dysfunction and FIG. 5, so redundant explanation will be omitted.

이 때, 제1 예측값(441) 및 제2 예측값(442)은 갑상선 기능이상증에 대한 확률값일 수 있다. 예를 들어, 제1 예측값(441)은 0 내지 1 사이의 값일 수 있고, 제2 예측값(442)은 0 내지 1 사이의 값일 수 있으며, 값의 범위는 이에 한정되지 않고 확률값을 나타낼 수 있는 다른 수치 범위로 나타낼 수도 있다.At this time, the first predicted value 441 and the second predicted value 442 may be probability values for thyroid dysfunction. For example, the first predicted value 441 may be a value between 0 and 1, and the second predicted value 442 may be a value between 0 and 1, but the range of values is not limited to this and may be other values that may represent probability values. It can also be expressed as a numerical range.

대상에 대한 갑상선 기능이상증에 대한 결과(450)는 제1 예측값(441) 및 제2 예측값(442)의 평균값에 기초하여 판단될 수 있다. 구체적으로, 제1 예측값(441) 및 제2 예측값(442)의 평균값이 임계값(threshold) 이상이면 대상은 갑상선 기능이상증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능이상증이 없는 것으로 판단될 수 있다.The result 450 regarding thyroid dysfunction for the subject may be determined based on the average value of the first predicted value 441 and the second predicted value 442. Specifically, if the average value of the first predicted value 441 and the second predicted value 442 is greater than or equal to the threshold, the subject is determined to have dysthyroidism, and if the average value is less than the threshold, the subject is determined to not have dysthyroidism. It can be judged that

예를 들어, 갑상선 기능이상 예측모델(430)이 갑상선 기능항진증 예측모델인 경우, 제1 예측값(441) 및 제2 예측값(442)의 평균값이 임계값이상이면 대상은 갑상선 기능항진증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능항진증이 없는 것으로 판단될 수 있다.For example, if the thyroid dysfunction prediction model 430 is a hyperthyroidism prediction model, if the average value of the first prediction value 441 and the second prediction value 442 is greater than the threshold, the subject is determined to have hyperthyroidism. If the average value is below the threshold, the subject can be judged to not have hyperthyroidism.

다른 예를 들어, 갑상선 기능이상 예측모델(430)이 갑상선 기능저하증 예측모델인 경우, 제1 예측값(441) 및 제2 예측값(442)의 평균값이 임계값이상이면 대상은 갑상선 기능저하증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능저하증이 없는 것으로 판단될 수 있다.For another example, when the thyroid dysfunction prediction model 430 is a hypothyroidism prediction model, if the average value of the first prediction value 441 and the second prediction value 442 is greater than the threshold, the subject is considered to have hypothyroidism. If it is determined that the average value is less than the threshold, the subject can be determined to not have hypothyroidism.

갑상선 기능이상증에 대한 결과(450)를 판단하기 위해 이용되는 임계값(threshold)은 갑상선 기능이상 예측모델(430)의 학습 결과에 기초하여 결정될 수 있다. 예를 들어, 학습된 갑상선 기능이상 예측모델에 검증 데이터들을 입력하여 획득된 예측값들로 갑상선 기능이상 예측모델의 성능을 평가할 때 설정된 임계값에 따라 갑상선 기능이상 예측모델의 민감도 및 특이도가 달라질 수 있다. 이에 따라, 임계값을 임의로 설정하여 갑상선 기능이상 예측모델의 성능을 평가하고, 그 중 민감도 및 특이도 수치가 높은 임계값을 최종 임계값으로 결정할 수 있다.The threshold used to determine the result 450 for thyroid dysfunction may be determined based on the learning results of the thyroid dysfunction prediction model 430. For example, when evaluating the performance of a thyroid dysfunction prediction model using the prediction values obtained by inputting validation data into the learned thyroid dysfunction prediction model, the sensitivity and specificity of the thyroid dysfunction prediction model may vary depending on the set threshold. there is. Accordingly, the performance of the thyroid dysfunction prediction model can be evaluated by arbitrarily setting the threshold, and the threshold with the highest sensitivity and specificity values can be determined as the final threshold.

도 11은 갑상선 기능이상 예측 방법을 설명하기 위한 도면(50)이다.Figure 11 is a diagram 50 for explaining a method for predicting thyroid dysfunction.

도 11을 참고하면, 대상(subject)에 대한 기준일(510, 520)은 두 개 있을 수 있다. 예를 들어, 대상에 대한 갑상선호르몬농도를 알고 있는 제1 기준일(510) 및 제2 기준일(520)이 존재할 수 있다.Referring to FIG. 11, there may be two reference dates 510 and 520 for the subject. For example, there may be a first reference date (510) and a second reference date (520) for which the thyroid hormone concentration for the subject is known.

구간 심박수들의 전처리 결과(511, 521) 및 호르몬 농도(512, 522)는 두 개의 기준일(510, 520) 각각에 대응될 수 있다. 구간 심박수들의 전처리 결과 및 호르몬 농도와 관련해서는, 4. 갑상선기능이상증의 예측방법 및 도 5에서 상술한 바 있으므로 중복 설명은 생략한다.The preprocessing results of section heart rates (511, 521) and hormone concentrations (512, 522) may correspond to each of the two reference days (510, 520). Regarding the preprocessing results of the heart rates and hormone concentrations, this has been described in detail in 4. Prediction method for thyroid dysfunction and FIG. 5, so redundant explanation will be omitted.

대상에 대한 갑상선 기능이상증에 대한 결과(550)를 획득하기 위해 적어도 하나의 갑상선 기능이상 예측모델(530) 및 적어도 하나의 갑상선 호르몬 농도 예측모델(535)이 이용될 수 있다. 갑상선 기능이상 예측모델(430) 은 상술한 갑상선 기능이상 예측모델 중 적어도 하나의 학습방법으로 구현될 수 있으며, 학습방법에 대한 중복 설명은 생략한다. 갑상선 호르몬 농도 예측모델(430) 은 상술한 갑상선 호르몬 농도 예측모델 중 적어도 하나의 학습방법으로 구현될 수 있으며, 학습방법에 대한 중복 설명은 생략한다.At least one thyroid dysfunction prediction model 530 and at least one thyroid hormone concentration prediction model 535 may be used to obtain a result 550 for thyroid dysfunction for the subject. The thyroid dysfunction prediction model 430 can be implemented using at least one learning method among the thyroid dysfunction prediction models described above, and redundant description of the learning method will be omitted. The thyroid hormone concentration prediction model 430 can be implemented using at least one learning method among the thyroid hormone concentration prediction models described above, and redundant description of the learning method will be omitted.

도 10을 참고하면, 두 개의 기준일(510, 520)이 있는 대상에 대한 갑상선 기능이상증에 대한 결과(550)를 획득하기 위해 제1 예측값(541) 및 제2 예측값(542)이 획득될 수 있다.Referring to FIG. 10, a first predicted value 541 and a second predicted value 542 may be obtained to obtain a result 550 for thyroid dysfunction for a subject with two reference dates 510 and 520. .

구체적으로, 제1 추정 호르몬값은 대상일(500)에 대응되는 구간 심박수들의 전처리 결과(501), 제1 기준일(510)에 대응되는 구간 심박수들의 전처리 결과(511), 제1 기준일(510)에 대응되는 호르몬 농도(512) 및 갑상선 호르몬 농도 예측모델(535)을 이용하여 획득되고, 제1 예측값(541)은 획득된 추정 호르몬값, 대상일(500)에 대응되는 구간 심박수들의 전처리 결과(501), 제1 기준일(510)에 대응되는 구간 심박수들의 전처리 결과(511), 제1 기준일(510)에 대응되는 호르몬 농도(512) 및 갑상선 기능이상 예측모델(530)을 이용하여 획득될 수 있다. 제2 추정 호르몬값은 대상일(500)에 대응되는 구간 심박수들의 전처리 결과(501), 제2 기준일(520)에 대응되는 구간 심박수들의 전처리 결과(521), 제2 기준일(520)에 대응되는 호르몬 농도(522) 및 갑상선 호르몬 농도 예측모델(535)을 이용하여 획득되고, 제2 예측값(542)은 획득된 추정 호르몬값, 대상일(500)에 대응되는 구간 심박수들의 전처리 결과(501), 제2 기준일(520)에 대응되는 구간 심박수들의 전처리 결과(521), 제2 기준일(520)에 대응되는 호르몬 농도(522) 및 갑상선 기능이상 예측모델(530)을 이용하여 획득될 수 있다. 추정 호르몬값을 획득하는 구체적인 내용은 7. 갑상선 호르몬 예측모델에 기초한 갑상선기능이상증의 예측방법 및 도 6에서 상술한 바 있으므로 중복 설명은 생략한다. 예측값을 획득하는 구체적인 내용은 4. 갑상선기능이상증의 예측방법 및 도 5에서 상술한 바 있으므로 중복 설명은 생략한다.Specifically, the first estimated hormone value is the preprocessing result 501 of section heart rates corresponding to the target date 500, the preprocessing result 511 of section heart rates corresponding to the first reference date 510, and the first reference date 510. is obtained using the corresponding hormone concentration 512 and the thyroid hormone concentration prediction model 535, and the first predicted value 541 is the obtained estimated hormone value and the preprocessing result of the heart rates in the section corresponding to the target date 500 ( 501), preprocessing results 511 of heart rates corresponding to the first reference date 510, hormone concentration 512 corresponding to the first reference date 510, and thyroid dysfunction prediction model 530. there is. The second estimated hormone value is the preprocessing result 501 of section heart rates corresponding to the target date 500, the preprocessing result 521 of section heart rates corresponding to the second reference date 520, and the preprocessing result 521 of section heart rates corresponding to the second reference date 520. It is obtained using the hormone concentration 522 and the thyroid hormone concentration prediction model 535, and the second predicted value 542 is the obtained estimated hormone value, the preprocessing result 501 of the heart rates in the section corresponding to the target date 500, It can be obtained using the preprocessing result 521 of the heart rates in the section corresponding to the second reference date 520, the hormone concentration 522 corresponding to the second reference date 520, and the thyroid dysfunction prediction model 530. The specific details of obtaining the estimated hormone value are described in detail in 7. Prediction method of thyroid dysfunction based on thyroid hormone prediction model and FIG. 6, so redundant explanation will be omitted. The specific details of obtaining the predicted value are described in detail in 4. Prediction method for thyroid dysfunction and FIG. 5, so redundant explanation will be omitted.

이 때, 제1 예측값(541) 및 제2 예측값(542)은 갑상선 기능이상증에 대한 확률값일 수 있다. 예를 들어, 제1 예측값(541)은 0 내지 1 사이의 값일 수 있고, 제2 예측값(542)은 0 내지 1 사이의 값일 수 있으며, 값의 범위는 이에 한정되지 않고 확률값을 나타낼 수 있는 다른 수치 범위로 나타낼 수도 있다.At this time, the first predicted value 541 and the second predicted value 542 may be probability values for thyroid dysfunction. For example, the first predicted value 541 may be a value between 0 and 1, and the second predicted value 542 may be a value between 0 and 1, but the range of values is not limited to this and may be other values that may represent probability values. It can also be expressed as a numerical range.

대상에 대한 갑상선 기능이상증에 대한 결과(550)는 제1 예측값(541) 및 제2 예측값(542)의 평균값에 기초하여 판단될 수 있다. 구체적으로, 제1 예측값(541) 및 제2 예측값(542)의 평균값이 임계값(threshold) 이상이면 대상은 갑상선 기능이상증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능이상증이 없는 것으로 판단될 수 있다. The result 550 regarding thyroid dysfunction for the subject may be determined based on the average value of the first predicted value 541 and the second predicted value 542. Specifically, if the average value of the first predicted value 541 and the second predicted value 542 is greater than or equal to the threshold, the subject is determined to have dysthyroidism, and if the average value is less than the threshold, the subject is determined to not have dysthyroidism. It can be judged that

예를 들어, 갑상선 기능이상 예측모델(430)이 갑상선 기능항진증 예측모델인 경우, 제1 예측값(441) 및 제2 예측값(442)의 평균값이 임계값이상이면 대상은 갑상선 기능항진증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능항진증이 없는 것으로 판단될 수 있다.For example, if the thyroid dysfunction prediction model 430 is a hyperthyroidism prediction model, if the average value of the first prediction value 441 and the second prediction value 442 is greater than the threshold, the subject is determined to have hyperthyroidism. If the average value is below the threshold, the subject can be judged to not have hyperthyroidism.

다른 예를 들어, 갑상선 기능이상 예측모델(430)이 갑상선 기능저하증 예측모델인 경우, 제1 예측값(441) 및 제2 예측값(442)의 평균값이 임계값이상이면 대상은 갑상선 기능저하증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능저하증이 없는 것으로 판단될 수 있다.For another example, when the thyroid dysfunction prediction model 430 is a hypothyroidism prediction model, if the average value of the first prediction value 441 and the second prediction value 442 is greater than the threshold, the subject is considered to have hypothyroidism. If it is determined that the average value is less than the threshold, the subject can be determined to not have hypothyroidism.

갑상선 기능이상증에 대한 결과(550)를 판단하기 위해 이용되는 임계값(threshold)은 갑상선 기능이상 예측모델(530)의 학습 결과에 기초하여 결정될 수 있으며, 구체적인 내용은 전술한 바 있으므로 중복 설명은 생략한다.The threshold used to determine the result 550 for thyroid dysfunction can be determined based on the learning results of the thyroid dysfunction prediction model 530, and since the specific details have been described above, redundant explanation is omitted. do.

전술한 갑상선 기능이상 예측 방법은 상술한 시스템(1)에 의해 수행될 수 있다.The above-described thyroid dysfunction prediction method can be performed by the above-described system (1).

도 12는 전술한 갑상선 기능이상 예측 방법을 설명하기 위한 흐름도이다.Figure 12 is a flowchart for explaining the method for predicting thyroid dysfunction described above.

도 12를 참고하면, 본 출원에 의해 개시되는 대상(subject)에 대한 갑상선 기능이상증을 예측하는 방법은, 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)을 포함한다.Referring to FIG. 12, the method for predicting thyroid dysfunction for a subject disclosed by the present application includes obtaining a trigger signal (S310), heart rate in the section corresponding to the target date determined by the trigger signal, Obtain pre-processing results for (S315), Obtain hormone concentrations corresponding to the first reference date (S320), Obtain pre-processing results for section heart rates corresponding to the first reference date (S325), The subject Obtaining a first difference value between the preprocessing result of the section heart rates corresponding to the day and the preprocessing result of the section heart rates corresponding to the first reference day (S330), the obtained first difference value corresponding to the first reference date Obtain the first predicted value by processing the hormone concentration with a thyroid dysfunction prediction model (S335), obtain the hormone concentration corresponding to the second reference date (S340), and pre-process the heart rates in the section corresponding to the second reference date. Obtaining a result (S345), obtaining a second difference value between the preprocessing result of the zone heart rates corresponding to the target date and the preprocessing result of the zone heart rates corresponding to the second reference day (S350), the obtained first 2 The difference value and the hormone concentration corresponding to the second standard are processed into a thyroid dysfunction prediction model to obtain a second prediction value (S355). Based on the average value of the first prediction value and the second prediction value, the second prediction value for dysthyroidism is obtained. Includes obtaining results (S360).

도 12의 트리거 신호의 획득(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330) 및 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335)은 도 5에서 설명한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150) 및 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)과 유사하게 수행될 수 있으므로 중복 설명은 생략한다.Acquisition of the trigger signal of FIG. 12 (S310), acquisition of preprocessing results for heart rates in the section corresponding to the target day determined by the trigger signal (S315), acquisition of hormone concentration corresponding to the first reference day (S320) ), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and preprocessing results for section heart rates corresponding to the first reference date Obtaining a first difference value (S330) and processing the obtained first difference value and the hormone concentration corresponding to the first reference date with a thyroid dysfunction prediction model to obtain a first prediction value (S335) are shown in Figure 5. Obtaining the described trigger signal (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), preprocessing the section heart rates corresponding to the target date (S120), corresponding to the reference date Obtaining the hormone concentration (S130), obtaining preprocessing results for section heart rates corresponding to the reference date (S140), preprocessing results for section heart rates corresponding to the target date and section heart rates corresponding to the reference date Obtain the difference between the pre-processing results (S150) and process the obtained difference and the hormone concentration corresponding to the reference date using a thyroid dysfunction prediction model and obtain the result of thyroid dysfunction for the subject (S160) ), so redundant description is omitted.

마찬가지로, 도 12의 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350) 및 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355)은 도 5에서 설명한 트리거 신호를 획득함(S100), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들을 획득함(S110), 상기 대상일에 대응되는 상기 구간 심박수들을 전처리함(S120), 기준일에 대응되는 호르몬 농도를 획득함(S130), 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S140), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 차이를 획득함(S150) 및 상기 획득된 차이 및 상기 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델을 이용하여 처리하고 상기 대상에 대한 갑상선기능이상증에 대한 결과를 획득함(S160)과 유사하게 수행될 수 있으므로 중복 설명은 생략한다.Similarly, the hormone concentration corresponding to the second reference day in FIG. 12 is obtained (S340), the preprocessing results for the section heart rates corresponding to the second reference day are obtained (S345), and the section heart rates corresponding to the target date are obtained (S345). A second difference value between the preprocessing result and the preprocessing result for the heart rates corresponding to the second reference date is obtained (S350), and the hormone concentration corresponding to the obtained second difference value and the second reference date is calculated for thyroid function. Obtaining a second predicted value by processing it with an abnormal prediction model (S355), obtaining the trigger signal described in FIG. 5 (S100), obtaining section heart rates corresponding to the target date determined by the trigger signal (S110), Preprocess the section heart rates corresponding to the target date (S120), obtain the hormone concentration corresponding to the reference date (S130), obtain preprocessing results for the section heart rates corresponding to the reference date (S140), and the target Obtain the difference between the preprocessing results of the section heart rates corresponding to the day and the preprocessing result of the section heart rates corresponding to the reference day (S150), and use the obtained difference and the hormone concentration corresponding to the reference day to create a thyroid dysfunction prediction model. Since it can be performed similarly to processing using and obtaining results on thyroid dysfunction for the subject (S160), redundant description will be omitted.

이 때, S335 단계에서 이용되는 갑상선기능이상 예측모델과 S355 단계에서 이용되는 갑상선기능이상 예측모델은 동일한 모델일 수 있다.At this time, the thyroid dysfunction prediction model used in step S335 and the thyroid dysfunction prediction model used in step S355 may be the same model.

이 때, S335 단계 및 S355 단계에서 획득되는 제1 예측값 및 제2 예측값은 갑상선 기능이상증 여부에 대한 확률값일 수 있다. 예를 들어, 제1 예측값(541)은 0 내지 1 사이의 값일 수 있고, 제2 예측값(542)은 0 내지 1 사이의 값일 수 있으며, 값의 범위는 이에 한정되지 않고 확률값을 나타낼 수 있는 다른 수치 범위로 나타낼 수도 있다.At this time, the first and second predicted values obtained in steps S335 and S355 may be probability values for whether thyroid dysfunction is present. For example, the first predicted value 541 may be a value between 0 and 1, and the second predicted value 542 may be a value between 0 and 1, but the range of values is not limited to this and may be other values that may represent probability values. It can also be expressed as a numerical range.

본 출원에 의해 개시되는 대상(subject)에 대한 갑상선 기능이상증을 예측하는 방법은 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)을 포함할 수 있다.The method for predicting thyroid dysfunction for a subject disclosed by the present application may include obtaining a result for thyroid dysfunction based on the average value of the first predicted value and the second predicted value (S360).

대상에 대한 갑상선 기능이상증에 대한 결과는 제1 예측값 및 제2 예측값의 평균값에 기초하여 판단될 수 있다. 구체적으로, 제1 예측값 및 제2 예측값의 평균값이 임계값(threshold) 이상이면 대상은 갑상선 기능이상증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능이상증이 없는 것으로 판단될 수 있다. The outcome for thyroid dysfunction for the subject may be determined based on the average of the first predicted value and the second predicted value. Specifically, if the average value of the first predicted value and the second predicted value is greater than or equal to the threshold, the subject may be determined to have dysthyroidism, and if the average value is less than the threshold, the subject may be determined to not have dysthyroidism.

예를 들어, 갑상선 기능이상 예측모델이 갑상선 기능항진증 예측모델인 경우, 제1 예측값 및 제2 예측값의 평균값이 임계값이상이면 대상은 갑상선 기능항진증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능항진증이 없는 것으로 판단될 수 있다.For example, if the thyroid dysfunction prediction model is a hyperthyroidism prediction model, if the average value of the first and second prediction values is above the threshold, the subject is judged to have hyperthyroidism, and if the average value is below the threshold, the subject is judged to have hyperthyroidism. can be judged as not having hyperthyroidism.

다른 예를 들어, 갑상선 기능이상 예측모델이 갑상선 기능저하증 예측모델인 경우, 제1 예측값 및 제2 예측값의 평균값이 임계값이상이면 대상은 갑상선 기능저하증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능저하증이 없는 것으로 판단될 수 있다.For another example, when the thyroid dysfunction prediction model is a hypothyroidism prediction model, if the average value of the first prediction value and the second prediction value is more than the threshold, the subject is judged to have hypothyroidism, and if the average value is less than the threshold value, the subject is judged to have hypothyroidism. The subject may be determined to not have hypothyroidism.

갑상선 기능이상증에 대한 결과를 판단하기 위해 이용되는 임계값(threshold)은 갑상선 기능이상 예측모델의 학습 결과에 기초하여 결정될 수 있으며, 구체적인 내용은 전술한 바 있으므로 중복 설명은 생략한다.The threshold used to determine the outcome of thyroid dysfunction can be determined based on the learning results of the thyroid dysfunction prediction model, and since the specific details have been described above, redundant explanation will be omitted.

전술한 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 모두 상기 심박수 측정 디바이스(10)에 의해 수행될 수 있다.Obtaining the above-described trigger signal (S310), obtaining preprocessing results for heart rates in the section corresponding to the target date determined by the trigger signal (S315), and obtaining hormone concentration corresponding to the first reference day (S320) ), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and preprocessing results for section heart rates corresponding to the first reference date Obtaining a first difference value (S330), processing the obtained first difference value and the hormone concentration corresponding to the first reference date with a thyroid dysfunction prediction model to obtain a first prediction value (S335), second reference date Obtaining the hormone concentration corresponding to (S340), obtaining preprocessing results for the zone heart rates corresponding to the second reference date (S345), preprocessing results for the zone heart rates corresponding to the target date and the second reference date Obtaining a second difference value of the preprocessing results for the corresponding section heart rates (S350), processing the obtained second difference value and the hormone concentration corresponding to the second standard as a thyroid dysfunction prediction model to obtain a second difference value (S350). Obtaining a predicted value (S355) and obtaining a result for thyroid dysfunction based on the average value of the first predicted value and the second predicted value (S360) can all be performed by the heart rate measuring device 10.

전술한 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 모두 상기 사용자 단말기(20)에 의해 수행될 수 있다.Obtaining the above-described trigger signal (S310), obtaining preprocessing results for heart rates in the section corresponding to the target date determined by the trigger signal (S315), and obtaining hormone concentration corresponding to the first reference day (S320) ), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and preprocessing results for section heart rates corresponding to the first reference date Obtaining a first difference value (S330), processing the obtained first difference value and the hormone concentration corresponding to the first reference date with a thyroid dysfunction prediction model to obtain a first prediction value (S335), second reference date Obtaining the hormone concentration corresponding to (S340), obtaining preprocessing results for the zone heart rates corresponding to the second reference date (S345), preprocessing results for the zone heart rates corresponding to the target date and the second reference date Obtaining a second difference value of the preprocessing results for the corresponding section heart rates (S350), processing the obtained second difference value and the hormone concentration corresponding to the second standard as a thyroid dysfunction prediction model to obtain a second difference value (S350). Obtaining a predicted value (S355) and obtaining a result for thyroid dysfunction based on the average value of the first predicted value and the second predicted value (S360) can all be performed by the user terminal 20.

전술한 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 모두 상기 서버(30)에 의해 수행될 수 있다.Obtaining the above-described trigger signal (S310), obtaining preprocessing results for heart rates in the section corresponding to the target date determined by the trigger signal (S315), and obtaining hormone concentration corresponding to the first reference day (S320) ), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and preprocessing results for section heart rates corresponding to the first reference date Obtaining a first difference value (S330), processing the obtained first difference value and the hormone concentration corresponding to the first reference date with a thyroid dysfunction prediction model to obtain a first prediction value (S335), second reference date Obtaining the hormone concentration corresponding to (S340), obtaining preprocessing results for the zone heart rates corresponding to the second reference date (S345), preprocessing results for the zone heart rates corresponding to the target date and the second reference date Obtaining a second difference value of the preprocessing results for the corresponding section heart rates (S350), processing the obtained second difference value and the hormone concentration corresponding to the second standard as a thyroid dysfunction prediction model to obtain a second difference value (S350). Obtaining a predicted value (S355) and obtaining a result for thyroid dysfunction based on the average value of the first predicted value and the second predicted value (S360) may all be performed by the server 30.

전술한 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 상기 심박수 측정 디바이스(10), 상기 사용자 단말기(20) 및 상기 서버(30)에 의해 적절히 분산되어 수행될 수 있다.Obtaining the above-described trigger signal (S310), obtaining preprocessing results for heart rates in the section corresponding to the target date determined by the trigger signal (S315), and obtaining hormone concentration corresponding to the first reference day (S320) ), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and preprocessing results for section heart rates corresponding to the first reference date Obtaining a first difference value (S330), processing the obtained first difference value and the hormone concentration corresponding to the first reference date with a thyroid dysfunction prediction model to obtain a first prediction value (S335), second reference date Obtaining the hormone concentration corresponding to (S340), obtaining preprocessing results for the zone heart rates corresponding to the second reference date (S345), preprocessing results for the zone heart rates corresponding to the target date and the second reference date Obtaining a second difference value of the preprocessing results for the corresponding section heart rates (S350), processing the obtained second difference value and the hormone concentration corresponding to the second standard as a thyroid dysfunction prediction model to obtain a second difference value (S350). Obtaining a predicted value (S355) and obtaining a result for thyroid dysfunction based on the average value of the first predicted value and the second predicted value (S360) are performed by the heart rate measuring device 10, the user terminal 20, and the server. It can be performed by appropriately dispersing by (30).

예를 들어, 트리거 신호를 획득함(S310), 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315)은 상기 사용자 단말기(20)에 의해 수행되고, 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 상기 서버(30)에 의해 수행될 수 있다.For example, acquiring a trigger signal (S310) and obtaining preprocessing results for heart rates in the section corresponding to the target date determined by the trigger signal (S315) are performed by the user terminal 20, Obtaining the hormone concentration corresponding to the first reference date (S320), obtaining preprocessing results for the section heart rates corresponding to the first reference date (S325), preprocessing results for section heart rates corresponding to the upper extremity target date and the first reference date 1 Obtaining the first difference value of the preprocessing results for the heart rates in the section corresponding to the reference day (S330), processing the obtained first difference value and the hormone concentration corresponding to the first reference day with a thyroid dysfunction prediction model Obtain the first predicted value (S335), obtain the hormone concentration corresponding to the second reference date (S340), obtain preprocessing results for the zone heart rates corresponding to the second reference date (S345), on the target date Obtaining a second difference value between the preprocessing result of the corresponding section heart rates and the preprocessing result of the section heart rates corresponding to the second reference date (S350), the obtained second difference value and the second reference date are obtained The hormone concentration is processed into a thyroid dysfunction prediction model to obtain a second predicted value (S355), and a result for thyroid dysfunction is obtained based on the average of the first and second predicted values (S360). The server 30 ) can be performed by.

다른 예를 들어, 트리거 신호를 획득함(S310)은 상기 사용자 단말기(20)에 의해 수행되고, 상기 트리거 신호에 의해 결정되는 대상일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S315), 제1 기준일에 대응되는 호르몬 농도를 획득함(S320), 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S325), 상지 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제1 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제1 차이값을 획득함(S330), 상기 획득된 제1 차이값 및 상기 제1 기준일에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제1 예측값을 획득함(S335), 제2 기준일에 대응되는 호르몬 농도를 획득함(S340), 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과를 획득함(S345), 상기 대상일에 대응되는 구간 심박수들의 전처리 결과와 상기 제2 기준일에 대응되는 구간 심박수들에 대한 전처리 결과의 제2 차이값을 획득함(S350), 상기 획득된 제2 차이값 및 상기 제2 기준을에 대응되는 호르몬 농도를 갑상선기능이상 예측모델로 처리하여 제2 예측값을 획득함(S355), 제1 예측값 및 제2 예측값의 평균값에 기초하여 갑상선기능이상증에 대한 결과를 획득함(S360)은 상기 서버(30)에 의해 수행될 수 있다.For another example, acquiring a trigger signal (S310) is performed by the user terminal 20, and preprocessing results for the heart rates corresponding to the target date determined by the trigger signal are obtained (S315). , Obtaining the hormone concentration corresponding to the first reference date (S320), Obtaining the preprocessing results of the section heart rates corresponding to the first reference date (S325), Preprocessing results of the section heart rates corresponding to the upper extremity target date and the above Obtain the first difference value of the preprocessing results for the heart rates in the section corresponding to the first reference date (S330), and process the obtained first difference value and the hormone concentration corresponding to the first reference date as a thyroid dysfunction prediction model. Obtain the first predicted value (S335), obtain the hormone concentration corresponding to the second reference date (S340), obtain preprocessing results for the heart rates in the section corresponding to the second reference date (S345), and the target date Obtaining a second difference value between the preprocessing result of the section heart rates corresponding to and the preprocessing result of the section heart rates corresponding to the second reference date (S350), corresponding to the obtained second difference value and the second reference The hormone concentration is processed with a thyroid dysfunction prediction model to obtain a second predicted value (S355), and a result for thyroid dysfunction is obtained based on the average of the first and second predicted values (S360). The server (S360) 30).

다만, 전술한 분산 수행의 형태에 국한되지 않고, 더 다양한 형태로 분산 수행될 수 있다.However, it is not limited to the form of distributed execution described above, and distributed execution may be performed in more diverse forms.

전술한 갑상선 기능이상 예측 방법은 최종 예측 결과가 갑상선 기능이상증에 대한 결과인 것으로 설명하였으나, 최종 예측 결과가 갑상선 호르몬 농도인 경우에도 유사한 방법이 적용될 수 있다. 구체적으로, 전술한 갑상선 기능이상 예측 방법에서 갑상선 기능이상 예측모델 대신 갑상선 호르몬 농도 예측 모델을 이용할 수 있으며, 보다 구체적인 설명은 생략한다.The above-described method for predicting thyroid dysfunction was explained as the final prediction result being a result of thyroid dysfunction, but a similar method can be applied even when the final prediction result is thyroid hormone concentration. Specifically, in the above-described thyroid dysfunction prediction method, a thyroid hormone concentration prediction model can be used instead of a thyroid dysfunction prediction model, and a more detailed description will be omitted.

전술한 갑상선 기능이상 예측 방법에서는 기준일이 2개인 것으로 설명하였으나, 이에 한정되지 않고 기준일은 3개, 4개, 5개, 6개, 7개, 8개, 9개 및 10개 중 하나일 수도 있다.In the above-mentioned thyroid dysfunction prediction method, it was explained that there are two reference dates, but the reference date is not limited to this and may be one of 3, 4, 5, 6, 7, 8, 9, and 10. .

예를 들어, 기준일이 제1 기준일, 제2 기준일 및 제3 기준일로 총 3개인 경우, 제1 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제1 기준일에 대응되는 구간 심박수들의 전처리 결과, 제1 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득되고, 제2 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제2 기준일에 대응되는 구간 심박수들의 전처리 결과, 제2 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득되고, 제3 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제3 기준일에 대응되는 구간 심박수들의 전처리 결과, 제3 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득될 수 있다.For example, if there are a total of three reference dates, including the first reference date, the second reference date, and the third reference date, the first predicted value is the preprocessing result of the zone heart rates corresponding to the target date, the preprocessing result of the zone heart rates corresponding to the first reference date, It is obtained using a prediction model for hormone concentration and thyroid dysfunction corresponding to the first reference date, and the second prediction value is the preprocessing result of the zone heart rates corresponding to the target date, the preprocessing result of the zone heart rates corresponding to the second reference date, and the second reference date. It is obtained using the corresponding hormone concentration and thyroid dysfunction prediction model, and the third prediction value is the preprocessing result of the section heart rates corresponding to the target date, the preprocessing result of the section heart rates corresponding to the third reference date, and the third reference date. It can be obtained using prediction models for hormone concentration and thyroid dysfunction.

이 때, 제1 예측값, 제2 예측값 및 제3 예측값은 갑상선 기능이상증에 대한 확률값일 수 있으며, 대상에 대한 갑상선 기능이상증에 대한 결과는 제1 예측값, 제2 예측값 및 제3 예측값의 평균값에 기초하여 판단될 수 있다. 구체적으로, 제1 예측값, 제2 예측값 및 제3 예측값의 평균값이 임계값(threshold) 이상이면 대상은 갑상선 기능이상증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능이상증이 없는 것으로 판단될 수 있다.At this time, the first predicted value, second predicted value, and third predicted value may be probability values for thyroid dysfunction, and the result for thyroid dysfunction for the subject is based on the average value of the first predicted value, second predicted value, and third predicted value. This can be judged. Specifically, if the average value of the first predicted value, second predicted value, and third predicted value is more than the threshold, the subject is judged to have dysthyroidism, and if the average value is less than the threshold, the subject is judged to not have dysthyroidism. It can be.

다른 예를 들어, 기준일이 제1 기준일, 제2 기준일, 제3 기준일 및 제4 기준일로 총 4개인 경우, 제1 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제1 기준일에 대응되는 구간 심박수들의 전처리 결과, 제1 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득되고, 제2 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제2 기준일에 대응되는 구간 심박수들의 전처리 결과, 제2 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득되고, 제3 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제3 기준일에 대응되는 구간 심박수들의 전처리 결과, 제3 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득되고, 제4 예측값은 대상일에 대응되는 구간 심박수들의 전처리 결과, 제4 기준일에 대응되는 구간 심박수들의 전처리 결과, 제4 기준일에 대응되는 호르몬 농도 및 갑상선 기능이상 예측모델을 이용하여 획득될 수 있다.For another example, if there are a total of four reference days, including the first reference date, the second reference date, the third reference date, and the fourth reference date, the first predicted value is the preprocessing result of the heart rates in the section corresponding to the target date, the section corresponding to the first reference date. As a result of pre-processing the heart rates, it is obtained using a prediction model for hormone concentration and thyroid dysfunction corresponding to the first reference date, and the second predicted value is the result of pre-processing the section heart rates corresponding to the target date, pre-processing the section heart rates corresponding to the second reference date. The result is obtained using a prediction model for hormone concentration and thyroid dysfunction corresponding to the second reference date, and the third prediction value is the preprocessing result of the zone heart rates corresponding to the target date, the preprocessing result of the zone heart rates corresponding to the third reference date, and the third prediction value. 3 It is obtained using a prediction model for hormone concentration and thyroid dysfunction corresponding to the reference date, and the fourth predicted value is the preprocessing result of the zone heart rates corresponding to the target date, the preprocessing result of the zone heart rates corresponding to the fourth reference date, and the fourth reference date. It can be obtained using the corresponding hormone concentration and thyroid dysfunction prediction model.

이 때, 제1 예측값, 제2 예측값, 제3 예측값 및 제4 예측값은 갑상선 기능이상증에 대한 확률값일 수 있으며, 대상에 대한 갑상선 기능이상증에 대한 결과는 제1 예측값, 제2 예측값, 제3 예측값 및 제4 예측값의 평균값에 기초하여 판단될 수 있다. 구체적으로, 제1 예측값, 제2 예측값, 제3 예측값 및 제4 예측값의 평균값이 임계값(threshold) 이상이면 대상은 갑상선 기능이상증이 있는 것으로 판단되고, 평균값이 임계값 미만이면 대상은 갑상선 기능이상증이 없는 것으로 판단될 수 있다.At this time, the first predicted value, the second predicted value, the third predicted value, and the fourth predicted value may be probability values for thyroid dysfunction, and the results for thyroid dysfunction for the subject are the first predicted value, the second predicted value, and the third predicted value. And it may be determined based on the average value of the fourth predicted value. Specifically, if the average value of the first predicted value, second predicted value, third predicted value, and fourth predicted value is more than the threshold, the subject is judged to have dysthyroidism, and if the average value is less than the threshold, the subject is judged to have dysthyroidism. It may be judged that there is no such thing.

기준일이 5개 이상인 경우에도 유사한 방법이 적용될 수 있으며, 구체적인 설명은 생략한다.A similar method can be applied even when there are five or more reference dates, and detailed explanations are omitted.

실험예#23.Experimental Example #23.

학습 데이터 세트들의 준비Preparation of training data sets

총 360명의 환자를 대상으로 수집한 총 1,504개의 임상 데이터 세트들이 사용되었다.A total of 1,504 clinical data sets collected from a total of 360 patients were used.

상술한 갑상선 기능항진증 예측모델의 학습 방법에 따라 모델이 학습되었다. The model was learned according to the learning method of the hyperthyroidism prediction model described above.

구체적으로, 갑상선 기능항진증 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.Specifically, the hyperthyroidism prediction model was trained using the following training data set.

모델 1 : [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]Model 1: [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, first Diagnosis result corresponding to test date (hyperactivity or non-hyperactivity)]

모델 2: [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]Model 2: [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of the section heart rates corresponding to the first test day - Standard deviation of zone heart rates corresponding to the second test day), thyroid hormone concentration corresponding to the second test day, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first test day]

모델 3 : [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]Model 3: [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of section heart rates corresponding to the first test day - standard deviation of zone heart rates corresponding to the second test day), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test day, diagnostic result (hyperactivity or not hyperactivity) corresponding to the first test day ]

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculating the average of the obtained resting heart rates.

사용된 제2 검사일에 대응되는 갑상선 호르몬 농도에는 free T4 농도 및 TSH 농도가 포함되었다.Thyroid hormone concentrations corresponding to the second test day used included free T4 concentration and TSH concentration.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 상술한 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning was 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying the two interpolation data generation described above.

갑상선기능항진증 예측모델의 학습Learning of hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선기능항진증 예측모델이 학습되었다.Using a light gradient boosting machine, a hyperthyroidism prediction model was learned using the aforementioned training data sets.

갑상선기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#23에 따라 학습된 갑상선 기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy of the hyperthyroidism prediction model learned according to Experimental Example #23 was tested using the test sets described above.

실험을 진행함에 있어서, 복수의 기준일에 기초한 갑상선 기능이상증의 예측 방법이 유의미한 성능 향상을 가져오는지 확인하기 위해서,In conducting the experiment, in order to determine whether the prediction method for thyroid dysfunction based on multiple reference dates results in significant performance improvement,

(1) 모델 1을 이용하여 하나의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(1) Evaluate the accuracy of thyroid dysfunction classification based on one reference date using Model 1,

(2) 모델 1을 이용하여 두 개의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(2) evaluate the accuracy of thyroid dysfunction classification based on two reference dates using Model 1;

(3) 모델 2을 이용하여 하나의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(3) Evaluate the accuracy of thyroid dysfunction classification based on one reference date using Model 2,

(4) 모델 2을 이용하여 두 개의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(4) Evaluate the accuracy of thyroid dysfunction classification based on two reference dates using Model 2;

(5) 모델 3을 이용하여 하나의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(5) Evaluate the accuracy of thyroid dysfunction classification based on one reference date using Model 3,

(6) 모델 3을 이용하여 두 개의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하였다.(6) Model 3 was used to evaluate the accuracy of thyroid dysfunction classification based on two reference dates.

두 개의 기준일에 기초한 갑상선 기능이상 분류는 도 12에 따른 방법으로 수행되었으며, 한 개의 기준일에 기초한 갑상선 기능이상 분류는 도 5에 따른 방법으로 수행되었다.Classification of thyroid dysfunction based on two reference dates was performed by the method according to FIG. 12, and classification of thyroid dysfunction based on one reference date was performed by the method according to FIG. 5.

실험예#23에 따르는 갑상선기능항진증 예측모델의 AUC(Area Under the ROC Curve), 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value), F1 Score는 다음의 [표 50]과 같다.The AUC (Area Under the ROC Curve), accuracy, sensitivity, specificity, PPV (Positive Predictive Value), NPV (Negative Predictive Value), and F1 Score of the hyperthyroidism prediction model according to Experimental Example #23 are as follows [Table 50] ] is the same as

AUCAUC 정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV F1 scoreF1 score (1)(One) 0.8670.867 78.7278.72 79.5979.59 79.7179.71 36.2736.27 96.2096.20 0.4970.497 (2)(2) 0.9490.949 89.9489.94 89.8189.81 89.9789.97 64.0364.03 97.8097.80 0.7480.748 (3)(3) 0.8740.874 81.0681.06 80.4380.43 81.1681.16 38.6538.65 96.5696.56 0.5220.522 (4)(4) 0.9520.952 89.9089.90 89.8189.81 89.9289.92 63.9263.92 97.8097.80 0.7470.747 (5)(5) 0.8820.882 81.6681.66 80.4380.43 81.8581.85 39.5439.54 96.5996.59 0.530.53 (6)(6) 0.9560.956 90.7190.71 90.5190.51 90.7590.75 66.0566.05 97.9697.96 0.7640.764

본 출원의 발명자들은 복수의 기준일에 기초하여 갑상선기능이상증을 예측할 경우 예측 성능이 향상됨을 알 수 있었다. The inventors of the present application found that prediction performance improved when predicting thyroid dysfunction based on multiple reference dates.

실험예#24.Experimental Example #24.

학습 데이터 세트들의 준비Preparation of training data sets

총 360명의 환자를 대상으로 수집한 총 1,504개의 임상 데이터 세트들이 사용되었다.A total of 1,504 clinical data sets collected from a total of 360 patients were used.

상술한 갑상선 호르몬 농도 예측모델의 학습 방법에 따라 갑상선 호르몬 농도 예측모델이 학습되었다. 특히, free T4의 농도를 예측하기 위한 예측모델이 학습되었다.A thyroid hormone concentration prediction model was learned according to the learning method of the thyroid hormone concentration prediction model described above. In particular, a prediction model was learned to predict the concentration of free T4.

구체적으로, 갑상선 호르몬 농도 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.Specifically, the thyroid hormone concentration prediction model was trained using the following training data set.

제1 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, (제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수), ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), ((제1 검사일에 대응되는 구간 조화평균 심박수 - 제2 검사일에 대응되는 구간 조화평균 심박수) / 제2 검사일에 대응되는 구간 조화평균 심박수), 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차, 제1 검사일에 대응되는 구간 심박수들의 첨도, 제2 검사일에 대응되는 구간 얕은 수면 비율, (제2 검사일에 대응되는 free T4 농도 x 제1 검사일에 대응되는 구간 심박수들의 첨도), ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 심박수들의 상대 표준편차), 제1 검사일에 대응되는 free T4 농도]Prediction model for first thyroid hormone concentration: [Thyroid hormone concentration corresponding to the second test day, (average heart rate of the section corresponding to the first test day - average heart rate of the section corresponding to the second test day), ((section corresponding to the first test day Average heart rate - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), ((section harmonic average heart rate corresponding to the first test day - section harmonic average heart rate corresponding to the second test day) / interval harmonic mean heart rate corresponding to the second test day), relative standard deviation of the interval heart rates corresponding to the second test day, kurtosis of the interval heart rates corresponding to the first test day, interval light sleep ratio corresponding to the second test day, (second Free T4 concentration corresponding to the test day relative standard deviation of heart rates), free T4 concentration corresponding to the first test day]

제2 갑상선 호르몬 농도 예측모델: [제2 검사일에 대응되는 갑상선 호르몬 농도, ((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 왜도 - 제2 검사일에 대응되는 구간 심박수들의 왜도), 제2 검사일에 대응되는 구간 심박수들의 표준편차, 제1 검사일에 대응되는 구간 심박수들의 왜도, 제2 검사일에 대응되는 구간 깊은 수면 비율, 제1 검사일에 대응되는 구간 REM 수면 비율, 대상의 성별, (제2 검사일에 대응되는 free T4 농도 x (제1 검사일에 대응되는 구간 심박수들의 왜도 - 제2 검사일에 대응되는 구간 심박수들의 왜도)), (((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 free T4 농도), (제2 검사일에 대응되는 구간 깊은 수면 비율 / 제1 검사일에 대응되는 구간 심박수들의 왜도), 제1 검사일에 대응되는 free T4 농도]Second thyroid hormone concentration prediction model: [Thyroid hormone concentration corresponding to the second test day, ((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average corresponding to the second test day heart rate), (skewness of section heart rates corresponding to the first test day - skewness of section heart rates corresponding to the second test day), standard deviation of section heart rates corresponding to the second test day, standard deviation of section heart rates corresponding to the first test day Skewness, interval deep sleep ratio corresponding to the second test day, interval REM sleep ratio corresponding to the first test day, gender of the subject, (free T4 concentration corresponding to the second test day x (interval heart rate corresponding to the first test day) Skewness - Skewness of section heart rates corresponding to the second test day)), ((((segment average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day ) / free T4 concentration corresponding to the second test day), (interval deep sleep ratio corresponding to the second test day / skewness of interval heart rates corresponding to the first test day), free T4 concentration corresponding to the first test day]

상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균으로 산출되었다. 상기 검사일에 대응되는 구간 조화평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 조화평균으로 산출되었다. 상기 검사일에 대응되는 구간 깊은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 깊은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 깊은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 얕은 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 얕은 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 얕은 수면 시간의 비율로 산출되었다. 상기 검사일에 대응되는 구간 REM 수면 비율은 상기 검사일을 기준으로 15일간의 총 수면 시간 및 총 REM 수면 시간을 획득하고, 상기 획득된 총 수면 시간에서 총 REM 수면 시간의 비율로 산출되었다.The section average heart rate corresponding to the test date was calculated by obtaining resting heart rates for 15 days based on the test date and averaging the obtained resting heart rates. The section harmonic average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculated as the harmonic average of the obtained resting heart rates. The section deep sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total deep sleep time for 15 days based on the test date, and calculated as the ratio of the total deep sleep time to the obtained total sleep time. The section light sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total light sleep time for 15 days based on the test date, and calculated as the ratio of the total light sleep time to the obtained total sleep time. The section REM sleep ratio corresponding to the test date was obtained by obtaining the total sleep time and total REM sleep time for 15 days based on the test date, and calculated as the ratio of the total REM sleep time to the total sleep time obtained.

사용된 제2 검사일에 대응되는 갑상선 호르몬 농도에는 free T4 농도 및 TSH 농도가 포함되었다.Thyroid hormone concentrations corresponding to the second test day used included free T4 concentration and TSH concentration.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 상술한 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning was 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying the two interpolation data generation described above.

상술한 갑상선 기능항진증 예측모델의 학습 방법에 따라 갑상선 기능항진증 예측모델이 학습되었다. A hyperthyroidism prediction model was learned according to the learning method of the hyperthyroidism prediction model described above.

구체적으로, 갑상선 기능항진증 예측모델은 아래와 같은 학습 데이터 세트로 학습이 진행되었다.Specifically, the hyperthyroidism prediction model was trained using the following training data set.

모델 1: [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제1 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]Model 1: [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of the section heart rates corresponding to the first test day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, first estimated thyroid hormone concentration, diagnosis corresponding to the first test day Results (hyperactivity or not hyperactivity)]

모델 2: [((제1 검사일에 대응되는 구간 평균 심박수 - 제2 검사일에 대응되는 구간 평균 심박수) / 제2 검사일에 대응되는 구간 평균 심박수), (제1 검사일에 대응되는 구간 심박수들의 표준편차 - 제2 검사일에 대응되는 구간 심박수들의 표준편차), 제1 시점에 대응되는 구간 심박수들의 표준편차, 제2 검사일에 대응되는 갑상선 호르몬 농도, 제2 추정 갑상선 호르몬 농도, 제1 검사일에 대응되는 진단결과(항진 또는 항진아님)]Model 2: [((section average heart rate corresponding to the first test day - section average heart rate corresponding to the second test day) / section average heart rate corresponding to the second test day), (standard deviation of the section heart rates corresponding to the first test day - standard deviation of zone heart rates corresponding to the second test date), standard deviation of zone heart rates corresponding to the first time point, thyroid hormone concentration corresponding to the second test date, second estimated thyroid hormone concentration, diagnosis corresponding to the first test date Results (hyperactivity or not hyperactivity)]

이때, 상기 검사일에 대응되는 구간 평균 심박수는 상기 검사일을 기준으로 15일간의 휴지기 심박수들을 획득하고, 상기 획득된 휴지기 심박수들의 평균을 산출하였다. At this time, the section average heart rate corresponding to the test date was obtained by obtaining resting heart rates for 15 days based on the test date, and calculating the average of the obtained resting heart rates.

사용된 제2 검사일에 대응되는 갑상선 호르몬 농도에는 free T4 농도 및 TSH 농도가 포함되었다.Thyroid hormone concentrations corresponding to the second test day used included free T4 concentration and TSH concentration.

제1 추정 갑상선 호르몬 농도는 제1 갑상선 호르몬 농도 예측모델로부터 획득되며, 제2 추정 갑상선 호르몬 농도는 제2 갑상선 호르몬 농도 예측모델로부터 획득되었다.The first estimated thyroid hormone concentration was obtained from the first thyroid hormone concentration prediction model, and the second estimated thyroid hormone concentration was obtained from the second thyroid hormone concentration prediction model.

이때, 서로 다른 환자들의 임상 데이터들 사이의 "차이"를 이용하여 학습 데이터 세트가 생성되지는 않았으며, 한 명의 환자로부터 획득된 임상 데이터 세트들을 이용하여 학습 데이터 세트가 생성되었다.At this time, a learning data set was not created using “differences” between clinical data of different patients, but a learning data set was created using clinical data sets obtained from one patient.

이 때, 학습에 활용된 임상 데이터들은 360명의 환자로부터 확보한 1,504개의 데이터들이 활용되었으며, 상술한 2개의 보간 데이터 생성을 적용하여 총 184,936개의 학습 데이터 세트들이 생성되었다.At this time, the clinical data used for learning was 1,504 data obtained from 360 patients, and a total of 184,936 learning data sets were created by applying the two interpolation data generation described above.

갑상선 호르몬 농도 예측모델 및 갑상선 기능항진증 예측모델의 학습Learning of thyroid hormone concentration prediction model and hyperthyroidism prediction model

Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 호르몬 농도 예측모델이 학습되었다.Using a light gradient boosting machine, a thyroid hormone concentration prediction model was learned using the above-described training data sets.

갑상선 호르몬 농도 예측모델이 먼저 학습된 후, Light gradient boosting machine을 이용하여, 전술한 학습 데이터 세트들을 이용하여 갑상선 기능항진증 예측모델이 학습되었다.A thyroid hormone concentration prediction model was first learned, and then a hyperthyroidism prediction model was learned using the above-mentioned learning data sets using a light gradient boosting machine.

갑상선 기능항진증 예측모델의 테스트Test of hyperthyroidism prediction model

학습된 예측 모델의 정확도를 판단하기 위하여 사용된 테스트 데이터 세트들은 총 351명의 환자로부터 확보된 1,148개의 임상 데이터들로부터 생성되었으며, 총 1,827개의 테스트 데이터 세트들이 생성되었다.The test data sets used to determine the accuracy of the learned prediction model were created from 1,148 clinical data obtained from a total of 351 patients, and a total of 1,827 test data sets were created.

실험예#24에 따라 학습된 갑상선기능항진증 예측모델의 정확도의 테스트는 전술한 테스트 세트들을 이용하여 진행되었다.The accuracy test of the hyperthyroidism prediction model learned according to Experimental Example #24 was conducted using the test sets described above.

실험을 진행함에 있어서, 복수의 기준일에 기초한 갑상선 기능이상증의 예측 방법이 유의미한 성능 향상을 가져오는지 확인하기 위해서,In conducting the experiment, in order to determine whether the prediction method for thyroid dysfunction based on multiple reference dates results in significant performance improvement,

(1) 모델 1을 이용하여 하나의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(1) Evaluate the accuracy of thyroid dysfunction classification based on one reference date using Model 1,

(2) 모델 1을 이용하여 두 개의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(2) evaluate the accuracy of thyroid dysfunction classification based on two reference dates using Model 1;

(3) 모델 2을 이용하여 하나의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하고,(3) Evaluate the accuracy of thyroid dysfunction classification based on one reference date using Model 2,

(4) 모델 2을 이용하여 두 개의 기준일에 기초한 갑상선 기능이상 분류에 대한 정확도를 평가하였다.(4) Model 2 was used to evaluate the accuracy of thyroid dysfunction classification based on two reference dates.

두 개의 기준일에 기초한 갑상선 기능이상 분류는 도 11에 따른 방법으로 수행되었으며, 한 개의 기준일에 기초한 갑상선 기능이상 분류는 도 5에 따른 방법으로 수행되었다.Classification of thyroid dysfunction based on two reference dates was performed by the method according to FIG. 11, and classification of thyroid dysfunction based on one reference date was performed by the method according to FIG. 5.

실험예#24에 따르는 갑상선 기능항진증 예측모델의 AUC(Area Under the ROC Curve), 정확도, 민감도, 특이도, PPV (Positive Predictive Value), NPV (Negative Predictive Value), F1 Score는 다음의 [표 51]와 같다.The AUC (Area Under the ROC Curve), accuracy, sensitivity, specificity, PPV (Positive Predictive Value), NPV (Negative Predictive Value), and F1 Score of the hyperthyroidism prediction model according to Experimental Example #24 are as follows [Table 51] ] is the same as

AUCAUC 정확도accuracy 민감도responsiveness 특이도specificity PPVPPV NPVNPV F1 scoreF1 score (1)(One) 0.88500.8850 82.0582.05 81.7081.70 82.1082.10 40.2540.25 96.8196.81 0.5390.539 (2)(2) 0.95160.9516 89.7989.79 89.3589.35 89.8889.88 63.7063.70 97.7097.70 0.7440.744 (3)(3) 0.88610.8861 82.3282.32 81.7081.70 82.4182.41 40.6840.68 96.8396.83 0.5430.543 (4)(4) 0.95690.9569 91.0991.09 90.7490.74 91.1691.16 67.1267.12 98.0298.02 0.7720.772

본 출원의 발명자들은 복수의 기준일에 기초하여 갑상선 기능이상증을 예측할 경우 예측 성능이 향상됨을 알 수 있었다. The inventors of the present application found that prediction performance improved when predicting thyroid dysfunction based on multiple reference dates.

Claims (15)

하나 이상의 프로세서에 의해 수행되는 대상에 대한 갑상선기능이상증의 예측 방법으로,A method for predicting dysthyroidism for a subject performed by one or more processors, comprising: 트리거 신호를 획득함;Acquire trigger signal; 상기 획득된 트리거 신호에 기초하여 결정된 대상일을 결정함;determining a target date determined based on the obtained trigger signal; i) 대상일에 대해 결정되는 제1 심박수 정보와 기준일에 대해 결정되는 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 제1 예측모델에 입력하여 상기 대상일에 대응되는 갑상선기능이상에 대한 가예측 결과를 획득함-상기 제1 심박수 정보 및 상기 제2 심박수 정보는 상기 대상에 관한 심박수 정보임-;i) the relationship between the first heart rate information determined for the target date and the second heart rate information determined for the reference date, and ii) the thyroid hormone concentration corresponding to the target date by inputting the reference day hormone concentration corresponding to the reference date into the first prediction model Obtaining a tentative prediction result for dysfunction - the first heart rate information and the second heart rate information are heart rate information about the subject; i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iii) 상기 가예측 결과를 제2 예측모델에 입력하여 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 획득함; 및i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the tentative prediction result is input into a second prediction model to produce a result for the predicted hormone concentration corresponding to the target date. acquired; and i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iv) 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 제3 예측모델에 입력하여 갑상선기능이상에 대한 예측 결과를 획득함;을 포함하는 The results of i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iv) the predicted hormone concentration corresponding to the target date are input into a third prediction model to determine thyroid dysfunction. Obtaining predicted results; including 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 가예측 결과를 획득함에 있어서,In obtaining the tentative prediction result, 상기 대상일에 대응되는 가예측 호르몬 농도에 대한 결과가 더 입력되고,The results of the tentatively predicted hormone concentration corresponding to the target date are further entered, 상기 가예측 호르몬 농도는, The tentatively predicted hormone concentration is, i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계 및 ii) 상기 기준일 호르몬 농도를 제4 예측모델에 입력하여 획득되는,Obtained by inputting i) the relationship between the first heart rate information and the second heart rate information and ii) the reference day hormone concentration into the fourth prediction model, 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 예측 호르몬 농도에 대한 결과는 대상일에 대응되는 갑상선 호르몬 농도 예측값인,The result of the predicted hormone concentration is the predicted thyroid hormone concentration corresponding to the target date, 대상의 갑상선기능이상을 예측하는 방법.How to predict thyroid dysfunction in a subject. 제1 항에 있어서,According to claim 1, 상기 예측 호르몬 농도에 대한 결과는 대상일의 갑상선 호르몬 농도와 상기 기준일 호르몬 농도 사이의 관계에 대응되는,The results for the predicted hormone concentration correspond to the relationship between the thyroid hormone concentration on the target day and the hormone concentration on the reference day, 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 관계는,The above relationship is, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량 및 7) 상기 기준일에 대응되는 구간 심박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence로 이루어진 그룹에서 선택된 하나 또는 그들의 조합인1) the amount of change in the average of the section heart rates of the target day with respect to the reference date, 2) the rate of change in the average of the section heart rates of the target day with respect to the reference date, 3) the amount of change in the standard deviation of the target day with respect to the reference date, 4 ) The amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) the amount of change in kurtosis of the target date with respect to the reference date, and 7) corresponding to the reference date One or a combination thereof selected from the group consisting of JS Divergence between the zone heart rates corresponding to the target date and the zone heart rates corresponding to the target date. 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 예측 호르몬 농도는,The predicted hormone concentration is, 1) 상기 대상의 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 2) 상기 대상의 사요오드티로닌(tetraiodothyronine, T4)의 농도, 3) 상기 대상의 혈청에 유리된 T4 (free T4)의 농도, 4) 상기 대상의 삼요오드티로닌(triiodothyronine, T3)의 농도, 5) 상기 대상의 혈청에 유리된 T3 (free T3)의 농도 및 6) 상기 대상의 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 하나 또는 그들의 조합에 대응되는1) Concentration of thyroid stimulating hormone (TSH) in the subject, 2) Concentration of tetraiodothyronine (T4) in the subject, 3) Concentration of free T4 (free T4) in the serum of the subject concentration, 4) the concentration of triiodothyronine (T3) in the subject, 5) the concentration of free T3 in the serum of the subject, and 6) the thyroid-stimulating hormone-releasing hormone (Thyrotropin-releasing hormone) in the subject. Releasing Hormone (TRH) concentration corresponding to one or a combination thereof selected from the group consisting of 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 제1 예측모델에는, 1) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화량, 2) 상기 기준일에 대한 상기 대상일의 구간 심박수들의 평균의 변화율, 3) 상기 기준일에 대한 상기 대상일의 표준편차의 변화량, 4) 상기 기준일에 대한 상기 대상일의 상대 표준편차의 변화량, 5) 상기 기준일에 대한 상기 대상일의 왜도의 변화량, 6) 상기 기준일에 대한 상기 대상일의 첨도의 변화량, 7) 상기 기준일에 대응되는 구간 심박수들과 상기 대상일에 대응되는 구간 심박수들 간의 JS Divergence, 8) 상기 기준일에 대응되는 상기 대상의 갑상선 자극 호르몬(thyroid stimulating hormone, TSH)의 농도, 9) 상기 기준일에 대응되는 상기 대상의 사요오드티로닌(tetraiodothyronine, T4)의 농도, 10) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T4 (free T4)의 농도, 11) 상기 기준일에 대응되는 상기 대상의 삼요오드티로닌(triiodothyronine, T3)의 농도, 12) 상기 기준일에 대응되는 상기 대상의 혈청에 유리된 T3 (free T3)의 농도 및 13) 상기 기준일에 대응되는 상기 대상의 갑상샘자극호르몬-방출 호르몬(Thyrotropin-Releasing Hormone, TRH)의 농도로 이루어진 그룹에서 선택된 제1 값 및 제2 값을 곱하거나 나눈 조합 값이 더 입력되는The first prediction model includes: 1) the amount of change in the average heart rates of the section of the target day with respect to the reference date, 2) the rate of change of the average of section heart rates of the target day with respect to the reference date, 3) the target date with respect to the reference date , 4) the amount of change in the relative standard deviation of the target date with respect to the reference date, 5) the amount of change in skewness of the target date with respect to the reference date, 6) the amount of change in kurtosis of the target date with respect to the reference date , 7) JS Divergence between the zone heart rates corresponding to the reference date and the zone heart rates corresponding to the target date, 8) the concentration of thyroid stimulating hormone (TSH) of the subject corresponding to the reference date, 9) Concentration of tetraiodothyronine (T4) of the subject corresponding to the reference date, 10) Concentration of free T4 (free T4) in the serum of the subject corresponding to the reference date, 11) The concentration of triiodothyronine (T3) of the subject, 12) the concentration of free T3 (free T3) in the serum of the subject corresponding to the reference date, and 13) the thyroid-stimulating hormone of the subject corresponding to the reference date- A combination value obtained by multiplying or dividing the first value and the second value selected from the group consisting of the concentration of Thyrotropin-Releasing Hormone (TRH) is further input. 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 제2 예측모델에는 상기 조합 값이 입력되지 않고, 상기 제3 예측모델에는 상기 조합 값이 더 입력되는The combination value is not input to the second prediction model, and the combination value is further input to the third prediction model. 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제1 항에 있어서,According to claim 1, 상기 가예측 결과를 획득함에 있어서,In obtaining the tentative prediction result, 상기 대상일에 대응되는 구간심박수들의 표준편차, 왜도, 첨도 및 상기 대상일과 상기 기준일 사이의 일차(day gap)로 이루어진 그룹에서 선택된 하나 또는 그들의 조합이 더 입력되는,One or a combination thereof selected from the group consisting of the standard deviation, skewness, kurtosis, and day gap between the target date and the reference date of the section heart rates corresponding to the target date is further input, 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction. 제5 항에 있어서,According to clause 5, 상기 대상일에 대응되는 구간 심박수들은 상기 대상일을 기준으로 미리 결정된 기간에 대응되는 모든 휴지기 심박수(resting heart rate)이고,The zone heart rates corresponding to the target date are all resting heart rates corresponding to a predetermined period based on the target date, 상기 기준일에 대응되는 구간 심박수들은 상기 기준일을 기준으로 상기 미리 결정된 기간에 대응되는 모든 휴지기 심박수인The zone heart rates corresponding to the reference date are all resting heart rates corresponding to the predetermined period based on the reference date. 대상에 대한 갑상선기능이상증의 예측방법.Method for predicting thyroid dysfunction for a subject. 제1 항에 있어서,According to claim 1, 상기 제1 예측모델은 갑상선 기능 항진증 또는 갑상선 기능 항진증 아님으로 분류하는 모델이고,The first prediction model is a model that classifies hyperthyroidism or non-hyperthyroidism, 상기 제3 예측모델은 갑상선 기능 항진증 또는 갑상선 기능 항진증 아님으로 분류하는 모델인The third prediction model is a model that classifies hyperthyroidism or non-hyperthyroidism. 대상에 대한 갑상선기능이상증의 예측방법.Method for predicting thyroid dysfunction for a subject. 제11 항에 있어서,According to claim 11, 상기 제2 예측모델은 갑상선 기능 항진증으로 확인된 환자들의 심박수 정보 및 그때의 호르몬 수치를 학습 데이터 세트로 하여 학습된 모델인The second prediction model is a model learned using the heart rate information and hormone levels of patients confirmed to have hyperthyroidism as a learning data set. 대상에 대한 갑상선기능이상증의 예측방법.Method for predicting thyroid dysfunction for a subject. 제1 항에 있어서,According to claim 1, 상기 제1 예측모델은 갑상선 기능 항진증, 정상 또는 갑상선 기능 저하증으로 분류하는 모델이고,The first prediction model is a model that classifies hyperthyroidism, normal thyroid function, or hypothyroidism, 상기 제3 예측모델은 갑상선 기능 항진증, 정상 또는 갑상선 기능 저하증으로 분류하는 모델인The third prediction model is a model that classifies hyperthyroidism, normal or hypothyroidism. 대상에 대한 갑상선기능이상증의 예측방법.Method for predicting thyroid dysfunction for a subject. 제1 항에 있어서,According to claim 1, 상기 제1 예측모델은 갑상선 기능 항진증으로 확인된 환자들의 심박수 정보 및 그때의 갑상선 기능 이상 여부를 학습 데이터 세트로 하여 학습된 모델이고,The first prediction model is a model learned using the heart rate information of patients confirmed to have hyperthyroidism and the presence of abnormal thyroid function at that time as a learning data set, 갑상선 기능 저하증으로 확인된 환자들의 심박수 정보 및 그때의 갑상선 기능 이상 여부를 학습 데이터 세트로 하여 학습된 제5 예측 모델은 i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 입력받아 갑상선기능이상에 대한 가예측 결과를 획득하고,The fifth prediction model learned using the heart rate information of patients confirmed to have hypothyroidism and whether or not the thyroid function is abnormal at that time as a learning data set is i) the relationship between the first heart rate information and the second heart rate information, and ii) the reference date. Obtain a tentative prediction result for thyroid dysfunction by inputting the hormone concentration on the corresponding reference day, 상기 제2 예측 모델은 상기 제5 예측 모델에서 출력된 상기 가예측 결과를 더 입력받아 예측 호르몬 농도에 대한 결과를 획득하는The second prediction model further receives the tentative prediction result output from the fifth prediction model and obtains a result for the predicted hormone concentration. 대상에 대한 갑상선기능이상증의 예측방법.Method for predicting thyroid dysfunction for a subject. 하나 이상의 프로세서에 의해 수행되는 대상에 대한 갑상선기능이상증의 예측 방법으로,A method for predicting dysthyroidism for a subject performed by one or more processors, comprising: 트리거 신호를 획득함;Acquire trigger signal; 상기 획득된 트리거 신호에 기초하여 결정된 대상일을 결정함;determining a target date determined based on the obtained trigger signal; i) 대상일에 대해 결정되는 대상일 심박수에 관한 제1 심박수 정보와 기준일에 대해 결정되는 기준일 심박수에 관한 제2 심박수 정보 간의 관계 및 ii) 상기 기준일에 대응되는 기준일 호르몬 농도를 제1 예측모델에 입력하여 상 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 획득함-상기 제1 심박수 정보 및 상기 제2 심박수 정보는 상기 대상에 관한 심박수 정보임-;i) the relationship between the first heart rate information regarding the target date heart rate determined for the target date and the second heart rate information regarding the reference date heart rate determined for the reference date, and ii) the reference day hormone concentration corresponding to the reference date to the first prediction model Input to obtain a result of the predicted hormone concentration corresponding to the target date - the first heart rate information and the second heart rate information are heart rate information related to the target; i) 상기 제1 심박수 정보와 상기 제2 심박수 정보 간의 관계, ii) 상기 기준일 호르몬 농도 및 iii) 상기 대상일에 대응되는 예측 호르몬 농도에 대한 결과를 제2 예측모델에 입력하여 갑상선기능이상에 대한 예측 결과를 획득함;을 포함하는 i) the relationship between the first heart rate information and the second heart rate information, ii) the hormone concentration on the reference day, and iii) the results of the predicted hormone concentration corresponding to the target date are input into the second prediction model to determine thyroid dysfunction. Obtaining predicted results; including 대상의 갑상선기능이상을 예측하는 방법.A method of predicting a subject's thyroid dysfunction.
PCT/KR2024/001308 2023-01-27 2024-01-26 Method and system for predicting thyroid dysfunction of subject Ceased WO2024158259A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR10-2023-0010807 2023-01-27
KR20230010807 2023-01-27

Publications (1)

Publication Number Publication Date
WO2024158259A1 true WO2024158259A1 (en) 2024-08-02

Family

ID=91970877

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2024/001308 Ceased WO2024158259A1 (en) 2023-01-27 2024-01-26 Method and system for predicting thyroid dysfunction of subject

Country Status (1)

Country Link
WO (1) WO2024158259A1 (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005523426A (en) * 2002-03-28 2005-08-04 アボット・ラボラトリーズ System and method for managing a patient care program
KR20170017648A (en) * 2015-08-07 2017-02-15 엘지전자 주식회사 A mehthod of measuring a resting heart rate
CN106419848A (en) * 2016-10-21 2017-02-22 北京通和堂生物科技有限公司 Device and method for monitoring thyroid function
US20170307629A1 (en) * 2007-02-16 2017-10-26 Genzyme Corporation Method of identifying risk for thyroid disorder
KR20190032668A (en) * 2017-09-18 2019-03-28 서울대학교병원 System for predicting thyrotoxicosis using a wearable device and computer program using the same

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005523426A (en) * 2002-03-28 2005-08-04 アボット・ラボラトリーズ System and method for managing a patient care program
US20170307629A1 (en) * 2007-02-16 2017-10-26 Genzyme Corporation Method of identifying risk for thyroid disorder
KR20170017648A (en) * 2015-08-07 2017-02-15 엘지전자 주식회사 A mehthod of measuring a resting heart rate
CN106419848A (en) * 2016-10-21 2017-02-22 北京通和堂生物科技有限公司 Device and method for monitoring thyroid function
KR20190032668A (en) * 2017-09-18 2019-03-28 서울대학교병원 System for predicting thyrotoxicosis using a wearable device and computer program using the same

Similar Documents

Publication Publication Date Title
WO2020159154A1 (en) Method for encoding images and corresponding terminals
WO2023191206A1 (en) Exploratory data analysis automation system and method based on variable attributes
WO2022039521A1 (en) Method for determining whether medication has been administered and server using same
Talib et al. Fuzzy decision-making framework for sensitively prioritizing autism patients with moderate emergency level
WO2018147643A2 (en) Thoracic measuring device, scoliosis correction system, remote spinal diagnostic system, and wearable measuring device
WO2017175922A1 (en) Cpr training system and method
WO2021132851A1 (en) Electronic device, scalp care system, and control method for same
WO2022182156A1 (en) Method for measuring physiological indices
WO2017034141A1 (en) Method and system for determining a length of an object using an electronic devices
WO2012050262A1 (en) Method and system for monitoring the performance of plant instruments using ffvr and glrt
WO2016204496A1 (en) System and method of providing information of peripheral device
WO2020230952A1 (en) Pain monitoring apparatus and method
WO2021251593A1 (en) Method for monitoring thyroid function according drug administration, and monitoring server and user terminal for performing same
WO2023277589A1 (en) Method for guiding visit for active thyroid eye disease examination, and system for performing same
WO2023008938A1 (en) Thyroid dysfunction prediction method and system for subject
WO2024158259A1 (en) Method and system for predicting thyroid dysfunction of subject
WO2022145544A1 (en) Method for providing diagnosis assistance information, and device performing same
WO2024035131A1 (en) Method for monitoring thyroid eye disease condition, and system for performing same
WO2022234952A1 (en) Wearable device for providing haptic feedback and operating method therefor
WO2024035043A1 (en) Method for monitoring sleep through sleep prediction model, and device for performing same
WO2023096082A1 (en) Wearable device and method for providing service on basis of user body temperature
WO2022182067A1 (en) Method and device for providing hair loss state information
WO2019045320A1 (en) Method and electronic device for predicting electronic structure of material
WO2022145538A1 (en) Medical image analysis method considering feature information, medical image analysis device, and medical image analysis system
EP4158646A1 (en) Digital apparatus and application for cancer cachexia therapy and methods of use thereof

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24747492

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE