KR20220130360A - Atrial fibrillation prognosis prediction device and prediction method - Google Patents

Abstract

The method for predicting the prognosis of atrial fibrillation according to an embodiment of the present invention comprises the steps of (a) receiving, by an atrial fibrillation prognosis predicting apparatus, any one or more of non-invasive clinical information and electrocardiogram information about a patient, (b) the atrial fibrillation performing pre-processing on any one or more of the received non-invasive clinical information and electrocardiogram information for the patient by the prognosis prediction device of and inputting one or more of the invasive clinical information and the electrocardiogram information into the prediction model and outputting any one of a prediction result of an invasive clinical prognosis for the patient and a prediction result of the invasive clinical information.

Description

Atrial fibrillation prognosis prediction device and method

The present invention relates to a prognosis prediction apparatus and method for atrial fibrillation. More particularly, it relates to an apparatus and method for predicting the prognosis of atrial fibrillation using non-invasive clinical information and electrocardiogram information that can be easily obtained through a simple examination.

Arrhythmia is a symptom in which the heart beats abnormally faster, slower, or irregularly due to the failure of regular contractions to continue due to poor electrical stimulation in the heart or improper delivery of the stimulus. Atrial fibrillation is the main cause, and in severe cases, it can lead to sudden death or even a stroke.

As a treatment for arrhythmias, there are surgical therapies that can block the electrical conduction of the heart by cauterizing the heart tissue, such as high-frequency electrode catheter ablation, to prevent arrhythmias, but this is a case in which atrial fibrillation has already occurred and has spread to arrhythmias. It is a treatment and does not correspond to a preventive strategy to prevent the possibility of atrial fibrillation in advance.

On the other hand, in order to block the possibility of atrial fibrillation in advance, it is necessary to accurately predict the prognosis of atrial fibrillation and give a prescription or procedure accordingly. It was common to give prediction results by However, since this conventional prediction method is based on empirical rules, there is a possibility that the accuracy of the judgment made by the clinician may be lacking. There is a problem that it is difficult to do this quickly with only human thinking or ability. In addition, in order for clinicians to obtain invasive clinical information, which is the basis for predicting the prognosis of atrial fibrillation, various tests must be performed. There is a problem that it can.

The present invention relates to a new and innovative technology that can minimize the physical and economic burden of a patient quickly and with high accuracy in predicting the prognosis of atrial fibrillation by reflecting these conventional problems.

Republic of Korea Patent Publication No. 10-2020-0084561 (2020.07.13)

The technical problem to be solved by the present invention is to provide an atrial fibrillation prognosis prediction apparatus and prediction method, which can rapidly and with high accuracy predict the atrial fibrillation prognosis.

Another technical problem to be solved by the present invention is to provide an atrial fibrillation prognosis prediction apparatus and prediction method capable of minimizing the physical and economic burden of a patient in predicting the atrial fibrillation prognosis.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above technical problem, a method for predicting atrial fibrillation prognosis according to an embodiment of the present invention includes the steps of: (a) receiving, by an atrial fibrillation prognosis prediction apparatus, any one or more of non-invasive clinical information and electrocardiogram information about a patient , (b) performing, by the atrial fibrillation prognosis predicting device, pre-processing on any one or more of the received non-invasive clinical information and electrocardiogram information for the patient, and (c) the atrial fibrillation prognosis predicting device performing the pre-processing inputting any one or more of non-invasive clinical information and electrocardiogram information about the patient who performed .

According to an embodiment, when the information received in step (a) is non-invasive clinical information about the patient, step (b) includes: (b-1) of the received non-invasive clinical information about the patient Determining whether a missing value exists and (b-2) if there is a missing value as a result of the determination in step (b-1), estimating and writing the missing value based on training data. .

According to an embodiment, after step (b-2), the method may further include (b-3) performing at least one of recursive analysis and correction on the estimated and written missing values.

According to an embodiment, after step (b-2), (b-4) performing normalization on the non-invasive clinical information written by estimating the missing value, and (b-5) the normalization It may further include the step of encoding the non-invasive clinical information performed (Encoding).

According to an embodiment, the information received in step (a) is the electrocardiogram information for the patient, and the electrocardiogram information for the patient is measured at 10 second intervals. Normal Sinus Rhythm 12-lead (Lead) In the case of electrocardiogram information, the step (b) includes: (b-1) extracting single rhythm electrocardiogram information from the received electrocardiogram information on the patient, (b-2) the extracted single rhythm electrocardiogram information extracting at least one electrocardiogram indices, and (b-3) extracting information per one or more peaks from the extracted single rhythm electrocardiogram information by using the one or more extracted electrocardiogram indicators.

According to an embodiment, the information for each peak may be either a P-wave or a PR-Wave.

According to an embodiment, the information received in step (a) is the electrocardiogram information for the patient, and the electrocardiogram information for the patient is measured at 10 second intervals. Normal Sinus Rhythm 12-lead (Lead) In the case of electrocardiogram information, step (b) includes (b-4) receiving a selection of any one or more of a high-frequency filter and a low-frequency filter from a user, and (b-5) any one or more of the selected high-frequency filter and low-frequency filter. It may include the step of receiving the cutoff frequency for the input.

According to an embodiment, the step (c) includes: (c-1) inputting one or more of non-invasive clinical information and electrocardiogram information about the patient who has performed the pre-processing to the predictive model, (c-2) It may include outputting a prediction result of invasive clinical information for the patient as it is input to the predictive model, and (c-3) learning the steps (C-1) and (C-2).

According to an embodiment, the predictive model may include: (M-1) receiving one or more selections of information received from non-invasive clinical information and electrocardiogram information from a user, (M-2) using the selected information from the user Selecting a predictor variable for invasive clinical information to be predicted, (M-3) receiving a range of hyperparameters from the user, and (M-4) specifying the range through a grid search manager It can be created by performing a search for the received hyperparameter.

According to an embodiment, when the information received in step (a) is non-invasive clinical information for the patient, step (c) is, (c-1) non-invasive clinical information for the patient who has performed the pretreatment The step of inputting information into the predictive model and extending it to a first format for a convolution operation, (c-2) normalizing non-invasive clinical information about patients expanded to the first format based on the number of patients step, (c-3) applying a convolution layer to the non-invasive clinical information about the normalized patient, performing batch normalization, and applying a Leaky ReLU activation function to extend to the second form and (c-4) flattening the non-invasive clinical information about the patient expanded to the second format for neural network computation.

According to an embodiment, if the number of patients is M, the number of non-invasive clinical information is N _n, and the number of kernels applied to the predictive model is C _N , the first form is [M × N _n × 1], and The second format may be [M × N _n × C _N-1 ].

According to an embodiment, (c-7) applying a fully connected layer to the non-invasive clinical information about the flattened patient, batch normalization, applying a ReLU activation function, and dropping Step of applying the dropout layer and (c-8) applying the output layer to the non-invasive clinical information about the patient to which the dropout layer is applied to predict the invasive clinical prognosis for the patient and the invasive The method may further include outputting any one of prediction results of clinical information.

According to an embodiment, when the information received in step (a) is electrocardiogram information for the patient, step (c) includes, (c-1') electrocardiogram information for the patient who has performed the pre-processing Entering the predictive model and extending it to a third format for a convolution operation, (c-2') Normalizing the ECG information for the patient expanded to the third format based on the number of patients and the number of leads step, (c-3'-1) applying a convolution layer to the normalized electrocardiogram information for the patient, batch normalization, and applying a Leaky ReLU activation function to expand to a fourth form step, (c-3'-2) applying a pooling layer to the electrocardiogram information for the patient expanded in the fourth format to expand to a fifth format, and (c-4') to the fifth format It may include the step of flattening the electrocardiogram information for the expanded patient in the form of a neural network calculation.

According to an embodiment, the number of patients is M, the length of the electrocardiogram information is E _n , the number of electrocardiogram leads is L _n , the number of kernels applied to the predictive model is C _{N , and} the size of the pooling layer is P _s , the third form is [M × E _n × L _n × 1], the fourth form is [M × E _n × L _n × C _N-1 ], and the fifth form is, [M × (E _n /P _s ) × (L _n /P _s ) × C _N-1 ].

According to an embodiment, (c-7') applying a fully connected layer to the electrocardiogram information for the flattened patient, batch normalization, applying a ReLU activation function, and dropout Step of applying the layer (Dropout Layer) and (c-8') applying the output layer to the electrocardiogram information for the patient to which the dropout layer is applied to predict the invasive clinical prognosis for the patient and invasive clinical information The method may further include outputting any one of the prediction results of .

According to an embodiment, when the information received in step (a) is non-invasive clinical information and electrocardiogram information about the patient, step (c) includes: (c-5) of the information input to the predictive model. Determining whether there are a plurality of types and (c-6) If, as a result of the determination in step (d-1), there are a plurality of types of input information, the non-invasive clinical information and the electrocardiogram for the flattened patient are used as the predictive model It may include the step of applying a concatenate layer by inputting into .

According to an embodiment, (d) further comprising the step of learning the steps (a) to (c), wherein the step (d) is, (d-1) any one of the non-invasive clinical information and the electrocardiogram information Classifying the patient ID including at least one into a class according to the risk of a predictor variable for any one of invasive clinical prognosis and invasive clinical information for the patient, (d-2) encoding the classified class step, (d-3) configuring a batch unit for any one or more of the non-invasive clinical information and electrocardiogram information 1:1, and (d-4) configuring the batch unit 1:1 The method may include performing learning by applying a gradient descent method (Adam Optimizer) to any one or more of the non-invasive clinical information and the electrocardiogram information.

Atrial fibrillation prognosis prediction apparatus according to another embodiment of the present invention for achieving the above technical problem is one or more processors, a network interface, a memory and large-capacity network data for loading a computer program executed by the processor, and storage for storing the computer program, wherein the computer program is configured to: (A) receive, by the one or more processors, any one or more of non-invasive clinical information and electrocardiogram information about the patient; (B) the received patient An operation to perform preprocessing on any one or more of non-invasive clinical information and electrocardiogram information for An operation for outputting any one of a prediction result of an invasive clinical prognosis for a patient and a prediction result of an invasive clinical information is performed.

The computer program stored in the medium according to another embodiment of the present invention for achieving the above technical problem is combined with a computing device, (AA) receiving any one or more of non-invasive clinical information and electrocardiogram information about the patient, (BB) performing pre-processing on any one or more of the received non-invasive clinical information and electrocardiogram information for the patient; is input to the prediction model and outputting any one of the prediction result of the invasive clinical prognosis for the patient and the prediction result of the invasive clinical information for the patient is executed.

According to the present invention as described above, if the user selects the type of information received and the predictor variables related to the invasive clinical prognosis and invasive clinical information, which are the prognosis of atrial fibrillation to be predicted, the atrial fibrillation prognosis prediction apparatus has very high accuracy. This has the effect of being able to quickly output the prediction results.

In addition, both non-invasive clinical information and electrocardiogram information about the patient, which are necessary information, are information that can be easily obtained through general outpatient treatment or simple examination, minimizing the patient's physical and economic burden in predicting the prognosis of atrial fibrillation. It has the effect of being able to do it.

In addition, it is possible to learn about the prediction process and the prediction result while using the atrial fibrillation prognosis prediction apparatus, so that the accuracy of the prediction result output can be improved as the use is repeated.

Effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

1 is a view showing the overall configuration included in the apparatus for predicting the prognosis of atrial fibrillation according to a first embodiment of the present invention.
2 is a flowchart illustrating representative steps of a method for predicting the prognosis of atrial fibrillation according to a second embodiment of the present invention.
FIG. 3 is a flowchart detailing step S220 of performing pre-processing when the information received in step S210 is non-invasive clinical information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.
4 is a diagram illustrating non-invasive clinical information received for a specific patient in a table format.
FIG. 5 is a diagram exemplarily illustrating a state in which missing values are estimated and written in the non-invasive clinical information shown in FIG. 4 in a table.
FIG. 6 is a diagram exemplarily illustrating a state in which missing values entered by table estimation of the non-invasive clinical information shown in FIG. 5 are changed.
7 is a flowchart detailing step S220 of performing pre-processing when the information received in step S210 is electrocardiogram information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.
8 is a diagram exemplarily illustrating sinus rhythm 12-lead ECG information measured at 10-second intervals and single rhythm ECG information extracted therefrom.
9 is a diagram illustrating one or more ECG indicators extracted from single rhythm ECG information together with their values in a table format.
10 is a diagram exemplarily illustrating extraction of P-wave information using P onset and P offset, which are ECG indicators.
11 is a diagram exemplarily illustrating extraction of PR-wave information using P onset and Q onset, which are electrocardiogram indicators.
12 is a diagram exemplarily showing a result of preprocessing holistic electrocardiogram information.
13 is a flowchart of generating a predictive model in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.
14 is a diagram illustrating a case in which the information received in step S210 is non-invasive clinical information about a patient or when the information selected by the user is only non-invasive clinical information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention; It is a flowchart showing the process of outputting the prediction result to .
15 is a diagram illustrating a prediction result when the information received in step S210 is electrocardiogram information for a patient or when the information selected by the user is only electrocardiogram information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention; This is a flowchart showing the printing process.
16 is a diagram illustrating a case in which the information received in step S210 is both non-invasive clinical information and electrocardiogram information about the patient in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention, or the information selected by the user is non-invasive clinical information; It is a flowchart illustrating a process of outputting a prediction result in the case of both information and electrocardiogram information.
17 is a diagram exemplarily showing a schematic diagram of a predictive model.
18 is a diagram exemplarily showing the prediction result of the invasive clinical prognosis and the prediction result of the invasive clinical information that are finally output.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments published below, but may be implemented in various different forms, and only these embodiments allow the publication of the present invention to be complete, and common knowledge in the technical field to which the present invention pertains. It is provided to fully inform the possessor of the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout.

Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used with the meaning commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in a commonly used dictionary are not to be interpreted ideally or excessively unless clearly defined in particular. The terminology used herein is for the purpose of describing the embodiments and is not intended to limit the present invention. In this specification, the singular also includes the plural, unless specifically stated otherwise in the phrase.

As used herein, "comprises" and/or "comprising" refers to the presence of one or more other components, steps, operations and/or elements mentioned. or addition is not excluded.

1 is a view showing the overall configuration included in the apparatus 100 for predicting the prognosis of atrial fibrillation according to a first embodiment of the present invention.

However, this is only a preferred embodiment for achieving the object of the present invention, some components may be added or deleted as necessary, and of course, a role performed by one component may be performed by another component.

The apparatus 100 for predicting atrial fibrillation prognosis according to the first embodiment of the present invention includes a processor 10 , a network interface 20 , a memory 30 , a storage 40 , and a data bus 50 connecting them. can do.

The processor 10 controls the overall operation of each component. The processor 10 may be any one of a central processing unit (CPU), a micro processor unit (MPU), a micro controller unit (MCU), or a type widely known in the art to which the present invention pertains. In addition, the processor 10 may perform an operation for at least one application or program for performing the prognosis prediction method of atrial fibrillation according to the second embodiment of the present invention, and as will be described later, a deep learning function It should be said that it is desirable to be an artificial intelligence processor for deep learning.

The network interface 20 supports wired/wireless Internet communication of the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention, and may support other known communication methods. Accordingly, the network interface 20 may be configured to include a corresponding communication module.

The memory 30 stores various data, commands and/or information, and loads one or more computer programs 41 from the storage 40 in order to perform the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention. can do. Although RAM is illustrated as one of the memories 30 in FIG. 1 , it goes without saying that various storage media can be used as the memory 30 .

The storage 40 may non-temporarily store one or more computer programs 41 and a large amount of network data 42 . The storage 40 is a non-volatile memory such as a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a hard disk, a removable disk, or in the art to which the present invention pertains. It may be any one of widely known computer-readable recording media.

The computer program 41 is loaded into the memory 30, and is operated by the one or more processors 10 to (A) receive any one or more of non-invasive clinical information and electrocardiogram information for a patient, (B) the received An operation to perform pre-processing of any one or more of non-invasive clinical information and electrocardiogram information about a patient, and (C) input any one or more of non-invasive clinical information and electrocardiogram information about a patient who has performed the pre-processing to the predictive model An operation of outputting any one of a prediction result of an invasive clinical prognosis for the patient and a prediction result of the invasive clinical information may be executed.

The operation performed by the computer program 41 briefly mentioned so far can be viewed as a function of the computer program 41, and a more detailed description of the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention. It will be described later in the description.

The data bus 50 serves as a movement path for commands and/or information between the processor 10 , the network interface 20 , the memory 30 , and the storage 40 described above.

The apparatus 100 for predicting the prognosis of atrial fibrillation according to the first embodiment of the present invention described above may be a physically independent electronic device, but non-invasive clinical information about a patient from a server (not shown) operated by a medical institution such as a hospital. and electrocardiogram information, it may be implemented as a function of the server, in this case, the server may be a tangible physical server or a virtual cloud server. .

Hereinafter, a method for predicting the prognosis of atrial fibrillation according to a second embodiment of the present invention will be described with reference to FIGS. 2 to 18 .

2 is a flowchart illustrating representative steps of a method for predicting the prognosis of atrial fibrillation according to a second embodiment of the present invention.

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the atrial fibrillation prognosis prediction apparatus 100 receives any one or more of non-invasive clinical information and electrocardiogram information about the patient (S210).

Here, non-invasive clinical information about the patient includes the patient's age, sex, weight, height, BMI index, systolic blood pressure, diastolic blood pressure, size of the atrium, and how long before atrial fibrillation started Information that can be easily acquired through As the information recorded in the form of electric potential on the surface of the body, it is information that can be obtained through a simple electrocardiogram examination.

For the non-invasive clinical information and electrocardiogram information received in step S210, it is sufficient just to selectively receive one or more of the two types of information. There are cases where both types of information are needed for each prediction result of the invasive clinical information or the prediction result of invasive clinical information. It would be desirable to receive them all.

Such non-invasive clinical information and electrocardiogram information will generally be received from a server (not shown) operated by a medical institution such as a hospital, and as mentioned above, the prognosis predicting apparatus of atrial fibrillation according to the first embodiment of the present invention ( 100) is implemented as a function of the server (not shown), reception means loading, and when a medical institution separately stores non-invasive clinical information and ECG information, reception means input You can have up to

On the other hand, when the driving purpose of the atrial fibrillation prognosis predicting apparatus 100 according to the first embodiment of the present invention is learning, any one or more of the non-invasive clinical information and the electrocardiogram information received in relation to the patient is the maximum number. It would be desirable to receive any one or more of non-invasive clinical information and electrocardiogram information about the patient, and if the driving purpose is to output any one of the prediction result of the invasive clinical prognosis and the prediction result of the invasive clinical information for a specific patient, It would be desirable to receive only one or more of non-invasive clinical information and electrocardiogram information about the patient.

Hereinafter, on the premise that any one or more of the non-invasive clinical information and the electrocardiogram information for the maximum number of patients has already been received and learned about it, the prediction result of the invasive clinical prognosis and the prediction result of the invasive clinical information for a specific patient A case in which the purpose of driving is to output any one is taken as an example to continue the description.

If any one or more of the non-invasive clinical information and the electrocardiogram information for the patient are received, the atrial fibrillation prognosis prediction apparatus 100 performs pre-processing on any one or more of the non-invasive granular information and the electrocardiogram information about the received patient. (S220).

Here, the pre-processing corresponds to a process to match the format, standard, etc. that can be input to the predictive model in step S230, which will be described later, and may be different depending on whether the information received in step S210 is non-invasive clinical information or electrocardiogram information. . It will be described below.

FIG. 3 is a flowchart detailing step S220 of performing pre-processing when the information received in step S210 is non-invasive clinical information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the apparatus 100 for predicting the prognosis of atrial fibrillation determines whether a missing value exists among the non-invasive clinical information about the patient received in step S210 (S220-1).

4, the non-invasive clinical information received for a specific patient is shown in a table format, and it can be seen that the “systolic blood pressure” item among various non-invasive clinical information for patient “Hong Gil-dong” is empty. A missing value is a value that does not exist for a specific item included in the clinical information, and a plurality of missing values may exist for non-invasive clinical information for a specific patient. because it may be different.

As a result of the determination in step S220-1, if there is a missing value, the atrial fibrillation prognosis prediction apparatus 100 estimates and writes the missing value based on the learning data (S220-2).

Previously, it was assumed that any one or more of the non-invasive clinical information and the electrocardiogram information for the maximum number of patients have already been received and learned. List the priorities of other patients with the same or similar values of other items included in the non-invasive clinical information for Any of a variety of known methods for estimating may be used.

On the other hand, if learning is not completed, for example, in the first driving state of the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention, non-invasive clinical information about a plurality of patients is received in step S210. If you do, you will be able to estimate the missing value based on this, and if you received only non-invasive clinical information about one patient in the initial driving state or step S210, the missing value corresponding to objective information such as the patient's age, height, weight, BMI index, etc. The average value of may be entered arbitrarily.

Referring to FIG. 5 , it can be seen that the “systolic blood pressure” item, which was empty in FIG. 4 , is written as 163 by estimating from blood pressure values of patients with similar items to patient Hong Gil-dong.

If the missing values are estimated and written, the prognosis predicting apparatus 100 for atrial fibrillation performs at least one of recursive analysis and correction on the estimated and written missing values (S220-3).

Here, the recursive analysis can be viewed as an estimation by the clinical factors associated with the item indicated by the missing value. In this case, the apparatus 100 for predicting the prognosis of atrial fibrillation according to the first embodiment of the present invention is a specific information included in the non-invasive clinical information. Among the items, the clinical factors should already have information about the items related to each other.

As a result of the recursive analysis on missing values, if the value of the missing value entered in step S220-2 is different from the recursive analysis result, the value of the missing value is corrected with the recursive analysis result, or between the recursive analysis result and the value of the missing value entered in step S220-2 It will be possible to correct any one of the values of , and the correction itself may not be performed at all. In the latter case, step S220-3 does not necessarily need to be performed. That is, whether correction is performed in step S220-3, whether to select a value, and more broadly, the execution of step S220-3 itself is determined by the user of the apparatus 100 for predicting the prognosis of atrial fibrillation according to the first embodiment of the present invention. can be seen as freely configurable.

Referring to FIG. 6 , it can be seen that the “systolic blood pressure” item, in which a value of 163 is written in FIG. 5, is changed to 157 as the recursive analysis and correction are performed.

After step S220-3, the apparatus 100 for predicting the prognosis of atrial fibrillation estimates the missing value and performs normalization on the entered non-invasive clinical information (S220-4) and the normalized non-invasive clinical information A step of encoding (S220-5) may be performed, which corresponds to a known technology commonly performed in the pre-processing process for information, and detailed description is omitted, and through this, non-invasive treatment for the patient is performed. Clinical information will be converted into a machine-readable state.

On the other hand, if there is no missing value as a result of the determination in step S220-1 above, step S220-4 may be directly performed without going through steps S220-2 and S220-3. In this case, step S220-4 includes estimation of missing values and It can be seen as a step that simply performs normalization regardless of writing.

This time, a pre-processing process performed when the information received in step S210 is electrocardiogram information will be described.

7 is a flowchart detailing step S220 of performing pre-processing when the information received in step S210 is electrocardiogram information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the apparatus 100 for predicting the prognosis of atrial fibrillation extracts single rhythm electrocardiogram information from among the electrocardiogram information about the patient received in step S210 (S220-1').

The ECG information obtained through the ECG test is mostly 12-lead ECG information, measured at 10-second intervals (Normal Sinus Rhythm, where sinus rhythm means that the ECG signals are evenly spaced). , this is also the case with the electrocardiogram information received in step S210.

On the other hand, single-rhythm ECG information can be calculated by applying a PQRST wave that appears repeatedly in the entire ECG information (500Hz) to a signal processing algorithm, but in the present invention, it is parsed from sinus rhythm 12-lead ECG information measured at 10-second intervals ( Parcing) to extract it.

Referring to FIG. 8 , single-rhythm ECG information (bottom) extracted from sinus rhythm 12-lead ECG information (top) measured at 10-second intervals can be checked, and ECG Feature that can be extracted from single-rhythm ECG information are shown together as examples.

If the single-rhythm ECG has been extracted, one or more ECG indicators are extracted from the single-rhythm ECG information extracted by the prognostic device 100 for atrial fibrillation (S220-2').

Here, the electrocardiogram index includes not only the heart rate, PR interval, QRS complex, QT interval, P onset, P offset, Q onset, Q offset, T onset, T offset, P width and T width, but also the previously exemplarily shown in FIG. 8 . Various indices may exist, which are exemplarily shown together with their values in the table shown in FIG. 9 .

If the ECG index is also extracted, one or more peak-specific information is extracted from the single rhythm ECG information extracted in step S220-1' using one or more ECG indexes extracted by the prognosis prediction apparatus 100 for atrial fibrillation (S220). -3')

Here, the one or more peak-specific information may be either P-wave or PR-Wave, and FIG. Extracting PR-wave information using P onset and Q onset is shown as an example.

To explain this in more detail, P-wave or PR-Wave, information for each peak, can be extracted using each axis information. Information about -wave or PR-Wave can be extracted.

On the other hand, the electrocardiogram information for the patient can be pre-processed through another process as well as the steps S220-1′ to S220-3′ described above. Instead of using single-rhythm ECG information, it can be viewed as whole-rhythm ECG information using all sinus rhythm 12-lead ECG information.

More specifically, the prognosis prediction apparatus 100 of atrial fibrillation may receive any one or more selections of a high-frequency filter and a low-frequency filter from the user (S220-4 ′), Pre-processing may be performed by receiving the cutoff frequency from the user (S220-5'). Since this is a known pre-processing process used in a general signal processing field, a detailed description thereof will be omitted, and only the pre-processing result is illustrated in FIG. 12 .

Let us return to the description of FIG. 2 again.

If pre-processing has been performed, the apparatus 100 for predicting the prognosis of atrial fibrillation inputs any one or more of non-invasive clinical information and electrocardiogram information about the patient who has performed the pre-treatment into the predictive model to predict the result of the invasive clinical prognosis for the patient and the invasive clinical Any one of the prediction results of the information is output (S230).

Here, the invasive clinical prognosis and invasive clinical information for a patient can be viewed as a kind of predictive variable for outputting predictive results through the present invention (hereinafter, Li will be used in parallel with “predictive variable”), and specific invasive clinical Depending on the type of prognosis and the type of specific invasive clinical information, the information to be input into the prediction model, for example, whether the output of the prediction result is possible only with the non-invasive clinical information, whether the output of the prediction result is possible only with the ECG information, the non-invasive clinical information and the Whether or not all of the electrocardiogram information is required may vary, and if non-invasive clinical information is required, even the detailed items included in the non-invasive clinical information may vary. Therefore, since it is difficult to match all of these detailed items and predictive variables in the present specification, the operation of the predictive model that can be conceptually explained by encompassing all of them will be described.

The predictive model is a CNN (Convolution Neural Network)-based neural network model, whether the information received in step S210 is non-invasive clinical information about the patient, electrocardiogram information, all of these information, and the predictor variables that are the prediction results to finally output The structure may vary depending on which one it is, and the apparatus 100 for predicting the prognosis of atrial fibrillation according to the first embodiment of the present invention outputs a predictive model optimized for these information and predictive variables in real time to output a predictive result. can be created with

13 is a flowchart of generating a predictive model in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention.

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the apparatus 100 for predicting the prognosis of atrial fibrillation receives one or more selected information from the user among non-invasive clinical information and electrocardiogram information (S1310), and any of the invasive clinical prognosis and invasive clinical information to be predicted from the selected information One predictor variable is selected (S1320).

This can be seen as a step in the user interface (UI, User Interface) side that allows the user to select what the prediction result to be finally output is as described above and what information is currently received to calculate it, step S1320 You can even choose whether the objective is classification (Softmax) or regression (MSE).

Thereafter, the prognosis predicting apparatus 100 of atrial fibrillation receives a range of hyperparameters from the user (S1330), and searches for the hyperparameters that have been assigned a range through the Grid Search Manager (S1340) A predictive model optimized for the predictor variable selected by the user may be generated.

Here, the hyperparameters specified by the user are, for example, the size of the convolution kernel (layer, layer) C _k , the number of convolution kernels C _N , the size of the pooling layer P _s , and the size of the fully connected layer. It can be FC _N which is the number of neurons, learning rate which is learning rate, dropout rate, etc. It goes without saying that a range for various hyperparameters can be specified, and predictive model using Grid Search Manager Since it corresponds to known technology, a detailed description thereof will be omitted.

If the prediction model is generated by selecting what the prediction result to be finally output is and what information is currently received to calculate it, the prediction result can be output by inputting the information. First, a case in which the information received in step S210 is non-invasive clinical information about a patient or a case in which the information selected by the user is only non-invasive clinical information will be described.

14 is a diagram illustrating a case in which, in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention, the information received in step S210 is non-invasive clinical information about the patient or the information selected by the user is only non-invasive clinical information It is a flowchart showing the process of outputting the prediction result to .

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the atrial fibrillation prognosis prediction apparatus 100 inputs non-invasive clinical information about the patient who has performed the pre-processing in step S220 into the prediction model and expands it into a first format for a convolution operation (S230-1). ).

For example, if the number of non-invasive clinical information received in step S210 or non-invasive clinical information selected by the user is N _n , the predictive variable is Y, and the number of patients is M, the preprocessed non-invasive clinical information is convolutional. It is to extend the dimension to [M × N _n × 1], which is the first form for .

Thereafter, the atrial fibrillation prognosis prediction apparatus 100 normalizes the non-invasive clinical information about the patient expanded to the first format based on the number of patients (S230-2).

Since the number of patients is M, [M × N _n × 1], which is non-invasive clinical information about patients expanded to the first format, is normalized based on M patients, and this can also be seen as a kind of preprocessing process.

Thereafter, a convolution layer is applied to the non-invasive clinical information about the patient normalized by the prognosis prediction apparatus 100 of atrial fibrillation (2D convolution), batch normalization is performed, and a Leaky ReLU activation function is applied. to expand to the second format (S230-3).

More specifically, this can be seen as being applied to a convolutional neural network including a 2D convolutional layer having the form [C _k × ₁ ]. In the second form, the dimension is extended to [M × N _n × C _N-1 ].

Thereafter, the apparatus 100 for predicting the prognosis of atrial fibrillation flattens the non-invasive clinical information about the patient expanded to the second format for neural network operation (S230-4).

This means that all dimensional elements of non-invasive clinical information about the patient are flattened in the form of [M × C _FC ], C _FC = N _n * C _N-1 for neural network computation.

After that, a fully connected layer is applied to the non-invasive clinical information about the patient flattened by the prognosis prediction device 100 of atrial fibrillation, batch normalization is performed, the ReLU activation function is applied, and the drop A dropout layer is applied (S230-7).

Through the fully connected layer, iterative matrix operation and batch normalization can be performed up to the FC _N ^N- 1th in the form of [M×FC] ㆍ [C _FC × FC _N ⁱ ] with the neurons of the FC _N list, and the ReLU activation function [M × FC _N ^N ] of neural network operation can be output through application and application of dropout layer.

Finally, the apparatus 100 for predicting the prognosis of atrial fibrillation applies the output layer to the non-invasive clinical information about the patient to which the dropout layer is applied, and the prediction result of the invasive clinical prognosis for the patient and the prediction result of the invasive clinical information Either one is output (S230-8).

The output layer is [M × FC _N ^N ] ㆍ [FC _N N × Y] Finally, [M × Y] can be output as either the prediction result of the invasive clinical prognosis for the patient or the prediction result of the invasive clinical information through the [FC N ^N × Y] neural network operation. If the predictor variable Y is a categorical variable, the final output [M × Y] of the neural network will be Y ^predict = Softmax([M × Y]) after the Softmax function is applied.

Steps S230-1 to S230-4 described above, steps S230-7 to S230-8 are steps made inside the prediction model, and when viewed from the outside, they will be treated as black box processing, but The description has been made, and when the actual steps are performed, the size C _k of the convolution kernel (layer, layer), which is a hyperparameter searched for in generating the predictive model, and the number of convolution kernels C _N are used. something to do.

This time, a case in which the information received in step S210 is electrocardiogram information about a patient or a case in which the information selected by the user is only electrocardiogram information will be described.

15 is a diagram illustrating a prediction result when the information received in step S210 is electrocardiogram information about a patient or when the information selected by the user is only electrocardiogram information in the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention. A flowchart showing the printing process.

This is only a preferred embodiment in achieving the object of the present invention, and some steps may be added or deleted as necessary, and furthermore, any one step may be included in another step.

Meanwhile, it is assumed that all steps are performed by the atrial fibrillation prognosis prediction apparatus 100 according to the first embodiment of the present invention.

First, the apparatus 100 for predicting the prognosis of atrial fibrillation inputs the electrocardiogram information about the patient who has performed the pre-processing in step S220 into the prediction model and expands it into a third format for convolution operation (S230-1′). .

For example, if the length of the ECG information received in step S210 or the non-invasive clinical information selected by the user is E _n , the number of ECG leads is L _n , the predictive variable is Y, and the number of patients is M, It is to extend the dimension of clinical information to [M × E _n × L _n × 1], which is the first form for the convolution operation.

Thereafter, the atrial fibrillation prognosis prediction apparatus 100 normalizes the ECG information for the patient expanded to the third format based on the number of patients and the number of leads (S230-2′)

Since the number of patients is M, [M × E _n × L _n × 1], which is the ECG information for the patient expanded to the third format, is normalized based on M patients and the number of ECG leads L _n . It can be seen as a pre-processing process, and the difference from the previous step S230-2 is that the information is electrocardiogram information, so the number of electrocardiogram leads is additionally normalized as a reference.

Thereafter, a convolution layer is applied to the electrocardiogram information of the patient normalized by the prognosis prediction apparatus 100 of atrial fibrillation, batch normalization is performed, and the Leaky ReLU activation function is applied to extend to the fourth form. Do (S230-3-1')

More specifically, this can be seen as being applied to a convolutional neural network including a 2D convolutional layer having the form [C _k × C _k ], batch normalization is performed up to the C _N- 1th layer, and the Leaky ReLU activation function is The applied fourth form is to extend the dimension to [M × E _n × L _n × C _N-1 ].

Thereafter, the atrial fibrillation prognosis prediction apparatus 100 applies a pooling layer to the electrocardiogram information for the patient expanded in the fourth format and expands to the fifth format (S230-3-2′).

This step S230-3-2′ was not previously performed when non-invasive clinical information about the patient was processed. According to the PS, the size of the pooling layer, the fifth form is [M × (E _n /P _s ) × (L _n /P _s ) × C _N-1 ].

Thereafter, the atrial fibrillation prognosis prediction apparatus 100 flattens the electrocardiogram information for the patient expanded to the fifth form for neural network operation (S230-4′).

It flattens all dimensional elements of the electrocardiogram information about the patient in the form [M × E _FC ], E _FC = (E _n /P _s )*(L _n /P _s )*C _N-1 ] for neural network computation. means to do

After the above step S230-4′, the fully connected layer is applied to the electrocardiogram information for the patient flattened by the atrial fibrillation prognosis prediction device 100, batch normalization, and the ReLU activation function are applied. and applying the dropout layer (S230-7′) and applying the output layer to the electrocardiogram information for the patient to which the dropout layer is applied to predict the result of the invasive clinical prognosis for the patient and Although the step (S230-8') of outputting any one of the prediction results of the invasive clinical information is performed, it is the same as the steps S230-7 and S230-8 described above, so a detailed description will be omitted to prevent duplicate description.

So far, when the information received in step S210 with reference to FIG. 14 is non-invasive clinical information about the patient or when the information selected by the user is only non-invasive clinical information, the information received in step S210 with reference to FIG. 15 is the patient The process of outputting a prediction result in the case of ECG information or when the user-selected information is only ECG information has been described. In this case, in the preceding description, when the information received in step S210 is both non-invasive clinical information and ECG information about the patient, or when the information selected by the user is both non-invasive clinical information and ECG information, it is necessary to explain the processing. It will be described below.

When the information received in step S210 is both non-invasive clinical information and electrocardiogram information about the patient, or when the information selected by the user is both non-invasive clinical information and electrocardiogram information, a basic description is also given in the previous description of FIG. 14 and FIG. 15 is the same as the description. That is, the processing of the non-invasive clinical information and the ECG information is performed separately. However, if [M × C _FC ] that has undergone step S230-4 and [M × E _FC ] that has undergone step S230-4′ are applied to the fully connected layer and undergo subsequent processes, two prediction results are finally output. Therefore, the process of merging them into one is essential.

In this case, as shown in FIG. 16, as a result of the determination of step S230-5 and step S230-5 of determining whether there are a plurality of types of information input to the prediction model in step S230, if there are a plurality of types of input information, [M × C _FC ] and S230 of step S230-4 by further including the step (S230-6) of applying a concatenate layer by inputting non-invasive clinical information and electrocardiogram for the flattened patient into the predictive model [M × E _FC ] in step -4′ to [M × (C _FC + E _FC )] It can be expanded, and accordingly, one merged planarization layer may be output, and after performing steps S230-5 and S230-6, steps S230-7 and S230-8, or steps S230-7' and It will be said that any of the steps S230-8' may be performed.

17 is a schematic diagram showing the prediction model. The upper left side of the schematic diagram is according to non-invasive clinical information about the patient, and the right side is according to the electrocardiogram for the patient. Since it proceeds to step -7 (or the same S230-7'), if they are connected, explanations will be possible with the flowchart shown in FIG. 14 and the flowchart shown in FIG. 15, respectively.

On the other hand, since the above-described prediction model is a deep learning-based model, learning can be performed. Accordingly, after step S230, a step (S240) of learning steps S210 to S230 may be performed, in this case S240 The step is that the atrial fibrillation prognosis prediction device 100 includes a patient ID including any one or more of non-invasive clinical information and electrocardiogram information according to the risk of predictive variables for any one of invasive clinical prognosis and invasive clinical information for the patient. Classifying into class (S240-1), encoding the classified class (S240-2), batch (Batch) unit for any one or more of non-invasive clinical information and electrocardiogram information 1:1 The step of configuring (S240-3) and applying the gradient descent method (Adam Optimizer) to any one or more of the non-invasive clinical information and the electrocardiogram information in which the batch unit is 1:1 (S240-4) to perform learning may include In this case, the encoding in step S240-2 may be one-hot-encoding, and the batch unit in step S240-3 may be made in a stratified sampling method in consideration of class imbalance, in step S240-4 Cosine Annealing companies can be additionally applied to overcome the problem of convergence to the local minimum, and Early Stopping is applied to prevent overfitting, but the termination condition is 10 or more errors (Logit Error). It may be set to end learning when .

So far, a method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention has been described. According to the present invention, if the user selects the type of information received and the predictor variables related to the invasive clinical prognosis and invasive clinical information, which are the prognosis of atrial fibrillation that the user wants to predict, the prognosis prediction apparatus 100 for atrial fibrillation has very high accuracy. to quickly output the prediction results. In addition, both non-invasive clinical information and electrocardiogram information about the patient, which are necessary information, are information that can be easily obtained through general outpatient treatment or simple examination, minimizing the patient's physical and economic burden in predicting the prognosis of atrial fibrillation. can do. In addition, since the prediction process and prediction result can be learned while using the atrial fibrillation prognosis prediction device, the accuracy of the prediction result output can be improved as the use is repeated.

18 exemplarily shows the prediction result of the invasive clinical prognosis and the prediction result of the invasive clinical information that are finally output. More specifically, from the top, the left atrial wall stress, an invasive clinical prognosis, through the non-invasive clinical information of the patient: Type of AF, Hypertension, Disbetes mellitus, Vascular disease, Heart failure, LVEF, E/EM. Prediction results of the left atrial hypotension, which is an invasive clinical prognosis through the non-invasive clinical information Age, Sex, type of AF, CHARDsVASc score, LA size, E/Em, Hb, PR interval, and invasive clinical information through electrocardiogram information The prediction result of left atrial voltage and electrocardiogram information show the prediction results of the type, gender, and age of atrial fibrillation, which are invasive clinical information, by way of example. It goes without saying that you can contribute to that.

On the other hand, although not described in detail to prevent overlapping description, the apparatus 100 for predicting the prognosis of atrial fibrillation according to the first embodiment of the present invention and the method for predicting the prognosis of atrial fibrillation according to the second embodiment of the present invention are the same. It can be implemented as a computer program stored in the medium according to the third embodiment of the present invention including technical features. In this case, the computer program stored in the medium is combined with a computing device, (AA) receiving any one or more of non-invasive clinical information and electrocardiogram information about the patient, (BB) the received non-invasive clinical information about the patient and performing pre-processing on any one or more of the electrocardiogram information and (CC) inputting any one or more of non-invasive clinical information and electrocardiogram information for the patient who performed the pre-processing into the predictive model to determine the invasive clinical prognosis for the patient The step of outputting any one of the prediction result and the prediction result of the invasive clinical information may be executed.

Although the embodiments of the present invention have been described above with reference to the accompanying drawings, those of ordinary skill in the art to which the present invention pertains can realize that the present invention can be implemented in other specific forms without changing the technical spirit or essential features. you will be able to understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

10: Processor
20: network interface
30: memory
40: storage
41: computer program
50: data bus
100: Atrial fibrillation prognosis prediction device

Claims (18)

(a) receiving, by the apparatus for predicting the prognosis of atrial fibrillation, any one or more of non-invasive clinical information and electrocardiogram information about the patient;
(b) performing, by the atrial fibrillation prognosis prediction device, pre-processing on any one or more of the received non-invasive clinical information and electrocardiogram information about the patient; and
(c) The apparatus for predicting the prognosis of atrial fibrillation inputs any one or more of non-invasive clinical information and electrocardiogram information about the patient who has performed the pre-processing into the predictive model to predict the invasive clinical prognosis for the patient and invasive clinical information outputting any one of the prediction results of ;
A prognostic prediction method of atrial fibrillation comprising a. According to claim 1,
When the information received in step (a) is non-invasive clinical information about the patient, step (b) comprises:
(b-1) determining whether a missing value exists among the received non-invasive clinical information about the patient; and
(b-2) if a missing value exists as a result of the determination in step (b-1), estimating and writing the missing value based on training data;
A prognostic prediction method of atrial fibrillation comprising a. 3. The method of claim 2,
After step (b-2),
(b-3) performing any one or more of recursive analysis and correction on the estimated and entered missing values;
A prognostic prediction method of atrial fibrillation further comprising. 3. The method of claim 2,
After step (b-2),
(b-4) performing normalization on the entered non-invasive clinical information by estimating the missing value; and
(b-5) encoding the normalized non-invasive clinical information;
A prognostic prediction method of atrial fibrillation further comprising. According to claim 1,
When the information received in step (a) is the electrocardiogram information for the patient, and the electrocardiogram information for the patient is Normal Sinus Rhythm 12-lead electrocardiogram information measured at 10-second intervals, the (b) step,
(b-1′) extracting single rhythm electrocardiogram information from among the received electrocardiogram information about the patient;
(b-2') extracting one or more ECG indicators from the extracted single rhythm ECG information; and
(b-3') extracting one or more peak-specific information from the extracted single rhythm ECG information using the one or more extracted ECG indicators;
A prognostic prediction method of atrial fibrillation comprising a. 6. The method of claim 5,
The information for each peak is,
Either P-wave or PR-Wave,
A method for predicting the prognosis of atrial fibrillation. According to claim 1,
When the information received in step (a) is the electrocardiogram information for the patient, and the electrocardiogram information for the patient is Normal Sinus Rhythm 12-lead electrocardiogram information measured at 10-second intervals, the (b) step,
(b-4′) receiving a selection of one or more of a high-frequency filter and a low-frequency filter from a user; and
(b-5') receiving a cutoff frequency for at least one of the selected high-frequency filter and low-frequency filter;
A prognostic prediction method of atrial fibrillation comprising a. According to claim 1,
The predictive model is
(M-1) receiving one or more selected information received from the non-invasive clinical information and the electrocardiogram information from the user;
(M-2) receiving, from the user, a predictive variable for any one of an invasive clinical prognosis to be predicted and invasive clinical information to be predicted through the selected information;
(M-3) receiving designation of a range of hyperparameters from the user; and
(M-4) performing a search for the hyperparameter with the specified range through the Grid Search Manager;
A method for predicting the prognosis of atrial fibrillation generated through According to claim 1,
When the information received in step (a) is non-invasive clinical information about the patient, step (c) comprises:
(c-1) inputting non-invasive clinical information about a patient who has performed the pre-processing into the predictive model and expanding it into a first format for a convolution operation;
(c-2) normalizing the non-invasive clinical information for patients expanded to the first format based on the number of patients;
(c-3) applying a convolution layer to the non-invasive clinical information about the normalized patient, performing batch normalization, and applying a Leaky ReLU activation function to extend to a second form; and
(c-4) flattening the non-invasive clinical information about the patient expanded to the second format for neural network computation;
A prognostic prediction method of atrial fibrillation comprising a. 11. The method of claim 10,
If the number of patients is M, the number of non-invasive clinical information is N _n, and the number of kernels applied to the predictive model is C _N ,
The first form is
[M × N _n × 1],
The second form is
[M × N _n × C _N-1 ],
A method for predicting the prognosis of atrial fibrillation. 11. The method of claim 10,
(c-7) Apply a fully connected layer to the non-invasive clinical information about the flattened patient, batch normalization, apply the ReLU activation function, and dropout layer applying; and
(c-8) applying an output layer to the non-invasive clinical information about the patient to which the dropout layer is applied, and outputting any one of the prediction result of the invasive clinical prognosis for the patient and the prediction result of the invasive clinical information step;
A prognostic prediction method of atrial fibrillation further comprising. The method of claim 1,
When the information received in step (a) is electrocardiogram information about the patient, step (c) includes:
(c-1') inputting electrocardiogram information about the patient who has performed the pre-processing into the predictive model and expanding it into a third format for a convolution operation;
(c-2') normalizing the ECG information for the patient expanded to the third format based on the number of patients and the number of leads;
(c-3′-1) applying a convolution layer to the normalized electrocardiogram information for the patient, performing batch normalization, and applying a Leaky ReLU activation function to expand it to a fourth form;
(c-3′-2) extending to a fifth format by applying a pooling layer to the ECG information about the patient expanded in the fourth format; and
(c-4′) flattening the ECG information for the patient expanded to the fifth format for neural network calculation;
A prognostic prediction method of atrial fibrillation comprising a. 13. The method of claim 12,
If the number of patients is M, the length of the electrocardiogram information is E _n , the number of electrocardiogram leads is L _n , the number of kernels applied to the predictive model is C _{N , and} the size of the pooling layer is P _s ,
The third form is
[M × E _n × L _n × 1],
The fourth form is
[M × E _n × L _n × C _N-1 ],
The fifth form is
[M × (E _n /P _s ) × (L _n /P _s ) × C _N-1 ],
A method for predicting the prognosis of atrial fibrillation. 13. The method of claim 12,
(c-7′) Applying a Fully Connected Layer to the electrocardiogram information for the flattened patient, performing batch normalization, applying a ReLU activation function, and applying a dropout layer applying; and
(c-8′) applying an output layer to the electrocardiogram information for the patient to which the dropout layer is applied, and outputting any one of the prediction result of the invasive clinical prognosis for the patient and the prediction result of the invasive clinical information ;
A prognostic prediction method of atrial fibrillation further comprising. According to claim 1,
When the information received in step (a) is non-invasive clinical information and electrocardiogram information about the patient, step (c) comprises:
(c-5) determining whether there are a plurality of types of information input to the predictive model; and
(c-6) As a result of the determination in step (d-1), if there are a plurality of types of input information, the non-invasive clinical information and the electrocardiogram for the flattened patient are input to the predictive model to form a concatenate layer applying;
A prognostic prediction method of atrial fibrillation comprising a. According to claim 1,
(d) learning the steps (a) to (c);
further comprising,
Step (d) is,
(d-1) The patient ID including any one or more of the non-invasive clinical information and the electrocardiogram information is classified as a class according to the risk of the predictor variable for any one of the invasive clinical prognosis and the invasive clinical information for the patient classifying;
(d-2) encoding the classified class;
(d-3) configuring a batch unit for any one or more of the non-invasive clinical information and the electrocardiogram information in a 1:1 ratio; and
(d-4) performing learning by applying a gradient descent method (Adam Optimizer) to any one or more of the non-invasive clinical information and the electrocardiogram information in which the batch unit is 1:1;
A prognostic prediction method of atrial fibrillation comprising a. one or more processors;
network interface;
a memory for loading a computer program executed by the processor; and
A storage for storing large-capacity network data and the computer program,
The computer program is executed by the one or more processors,
(A) receiving any one or more of non-invasive clinical information and electrocardiogram information about the patient;
(B) an operation of pre-processing any one or more of the received non-invasive clinical information and electrocardiogram information about the patient; and
(C) Input any one or more of non-invasive clinical information and electrocardiogram information about the patient who has performed the pre-processing into the prediction model, and output any one of the prediction result of the invasive clinical prognosis for the patient and the prediction result of the invasive clinical information operation to do;
A prognostic predictor of atrial fibrillation. In combination with a computing device,
(AA) receiving any one or more of non-invasive clinical information and electrocardiogram information about the patient;
(BB) performing pre-processing on any one or more of the received non-invasive clinical information and electrocardiogram information about the patient; and
(CC) Input any one or more of non-invasive clinical information and electrocardiogram information about the patient who has performed the pre-treatment into the prediction model, and output any one of the prediction result of the invasive clinical prognosis and the prediction result of the invasive clinical information for the patient to do;
to run
A computer program stored on a medium.

Images