RU2742429C1 - Method for rapid assessment of changes in lung tissue with covid-19 without using computer tomography of thorax organs - Google Patents

Abstract

FIELD: medicine.
SUBSTANCE: invention relates to medicine, in particular to pulmonology and therapy, and can be used for the rapid assessment of the condition in patients with COVID-19, which makes it possible, inter alia, to determine the necessity of hospitalization of a patient in a hospital. The method for the rapid assessment of changes in lung tissue with COVID-19 involves determining a set of diagnostically significant indicators, for which purpose the level of biomarkers in the sample of biological fluids, blood and urine, obtained in the subject is measured; physical examination; collection of anamnesis, obtaining data on diagnostically significant indicators: severity of the patient’s condition, respiratory rate, respiratory or respiratory difficulties, body temperature, presence of a sense of insensitivity in the thorax, weakness and/or a sense of bone ache, cough, type of cough, sex, age of test, ischaemic heart disease, risk group: chronic disease, and/or pregnancy, and/or age 65 and over, as well as quantitative SARS-CoV-2 antibodies IgG and IgM, C-reactive protein, absolute number of lymphocytes, absolute number of granulocytes, absolute number of normablasts, ESR under Westergren, hematocrit, p24 HIV-1 antigen and/or antibody HIV-1/2, urobilinogens, nitrites; all the data obtained are processed using at least one classification model, which has been trained to determine the absence of lesions of pulmonary tissue, light, medium-heavy or severe pulmonary tissue.

EFFECT: technical result achieved when using the developed method is possibility of rapidly assessing degree of change of lung tissue with COVID-19 without using instrumental methods of investigation.
3 cl, 5 tbl, 2 dwg

Description

The technical field to which the invention relates

The invention relates to the field of medicine, namely to pulmonology and therapy, and can be used for the rapid assessment of the severity of the condition of patients with COVID-19, allowing, among other things, to determine the need for hospitalization of a patient in a hospital.

On March 11, 2020, the World Health Organization announced a pandemic for the COVID-19 disease caused by the SARS-CoV-2 virus. In the context of a coronavirus pandemic, computed tomography of the chest organs (CT scan) takes an important place in the diagnosis of the disease. CT data can serve as predictors of the need for hospitalization and the likelihood of an unfavorable outcome in the intensive care unit. At the same time, the significant disadvantages of the CT WGC procedure - low specific resource intensity, high economic costs, problems of radiation safety of patients and medical personnel - require the search for new approaches to determine the severity of pneumonia in COVID-19.

State of the art

There is a known method for assessing changes in the lungs with COVID-19 (Morozov S.P., Protsenko D.N., Smetanina S.V. and others. Radiation diagnosis of coronavirus disease (COVID-19): organization, methodology, interpretation of results: preprint No. CDT - 2020 - II. Version 2 of 17.04.2020 // Series "The best practices of radiation and instrumental diagnostics." - Issue 65. - M .: GBUZ "NPKTS DT DZM", 2020.). According to this publication, the main method for diagnosing, confirming and assessing the dynamics of COVID-19 (taking into account clinical and laboratory data) is computed tomography (CT) of the chest in high resolution. At the same time, their use is not indicated for screening coronavirus infection in the absence of symptoms of an acute respiratory viral disease. Typical manifestations of COVID-19 on CT scan of the chest organs: numerous seals of the pulmonary parenchyma of the "ground glass" type, predominantly round in shape, of varying length with / without consolidation; thickening of the interlobular interstitium of the "cobblestone" type; peripheral, multilobar localization. In general, according to radiation methods, there are zero, mild, medium-severe, severe and critical degrees of severity of the disease. The zero severity of the disease means either a normal picture of the lungs, or signs of any other pathological conditions (inflammatory, oncological, etc.). The severity of the disease is assessed based on the percentage of involvement of the lung parenchyma in the pathological process (the state of the lung with the greatest damage is taken into account).

In the Russian Federation, according to the Interim Guidelines of the Russian Society of Radiologists and Radiologists and the Russian Association of Specialists in Ultrasound Diagnostics in Medicine (Sinitsyn V.E., Tyurin I.E., Mitkov V.V. Temporary conciliatory guidelines of the Russian Society of Radiologists and Radiologists ( POPR) and the Russian Association of Specialists in Ultrasound Diagnostics in Medicine (RASUDM) "Methods of Radiological Diagnosis of Pneumonia in New Coronavirus Infection COVID-19" (version 2). Bulletin of Radiology and Radiology. 2020; 101 (2): 72-89), in conditions a large flow of patients for the rapid assessment of changes in the lungs detected by CT, the so-called "empirical" visual scale is recommended. It is based on a visual assessment of the approximate volume of compacted lung tissue. This scale has 5 gradations, it starts from 0, and then - division at intervals of 25%. The Moscow Department of Health uses guidelines according to which it is necessary to assess changes in lung tissue in COVID-19 according to CT scan of the OGK, based on determining the percentage of lung damage. The percentage of damage is assessed separately for each lung. The category of changes is determined by the lung with the greatest lesion.

However, the massive use of CT OGK to assess changes in the lung parenchyma in COVID-19 has a number of limitations and disadvantages:

• The risk of creating artificial epidemic foci (impossibility of full sterilization of the CT machine, air in the room between procedures, the need to violate the self-isolation regime for patients who did not require hospitalization).

• Irrational work of ambulance teams (work with patients of outpatient CT centers - the team is busy from 3 hours per patient).

• Economic costs for CT scan.

• Problems related to the radiation safety of patients and doctors (especially young people).

As an alternative method for determining the severity of lung damage in COVID-19, the specialists of the A.V. Vishnevsky developed an ultrasound method for determining the area of lesions of the lung parenchyma (RF patent 2729368, 06.08.2020 - prototype). According to this method, differentiation of moderate or pronounced interstitial changes is carried out, when pronounced interstitial changes are detected in the study areas, the severity of pneumonia in COVID-19 caused by the SARS-CoV-2 virus is determined by the formula: S = 4n / N, where S is the severity , characterized by the area of damage to the lung tissue, n is the number of zones with identified changes, N is the total number of zones for which the ultrasound study was carried out. When receiving the value of the severity 0 ≤ S <1, it is concluded that the area of the lesion of the lung parenchyma is up to 25%. When a value of 1 ≤ S <2 is obtained, it is concluded that the area of the lesion of the lung parenchyma is from 25 to 50%. When a value of 2 ≤ S <3 is obtained, it is concluded that the area of the lesion of the lung parenchyma is from 50 to 75%. When a value of S ≥ 3 is obtained, it is concluded that the area of the lesion of the lung parenchyma is 75% or more.

However, the known method does not provide the necessary diagnostic accuracy, since it does not take into account the overall clinical picture and often requires the use of other diagnostic imaging options.

The technical problem to be solved was the development of an express assessment of changes in lung tissue in COVID-19, which does not require the use of radiation diagnostic methods, but provides high diagnostic accuracy.

Disclosure of the essence of the invention

The technical result achieved when using the developed method is the ability to assess the degree of changes in lung tissue in COVID-19 in an express mode without using instrumental research methods.

This result is achieved due to the following set of essential features:

a set of diagnostically significant indicators is determined, for which the level of biomarkers is measured in a sample of biological fluids - blood and urine obtained from the subject; physical examination; collection of anamnesis;

receive data on diagnostically significant indicators: the severity of the patient's condition, respiratory rate, shortness of breath or shortness of breath, body temperature, the presence of a feeling of congestion in the chest, the presence of weakness and / or a feeling of aches, the presence of a cough, type of cough, gender of the subject, age of the subject, coronary heart disease, assignment of the subject to a risk group: the presence of chronic diseases, and / or the presence of pregnancy, and / or age 65 years and older, as well as quantitative indicators of the presence of IgG and IgM antibodies to SARS-CoV-2 IgG, IgM, C- reactive protein, absolute lymphocyte count, absolute granulocyte count, absolute normoblast count, Westergren ESR, RDW, hematocrit, HIV-1 p24 antigen and / or HIV-1/2 antibody, urobilinogen, nitrites;

the set of the obtained data is processed using at least one classification model trained to determine: the absence of damage to the lung tissue, mild, moderate or severe degree of damage to the lung tissue.

It is advisable to use the "random forest" method as classification models.

If it is possible to study the characteristics of blood coagulation, then the D-dimer, the upper reference limit for the D-dimer (taking into account the test system used), and glucose are additionally determined.

We analyzed the relationship of various clinical and laboratory parameters (186 variables were initially investigated) with the volume of lung tissue lesions according to CT data of WGC in 430,780 patients. Based on this analysis, a set of indicators was determined that are most significant in predicting changes in lung tissue in COVID-19. The parameters with the greatest significance were included as predictive features in the training model built using machine learning algorithms (random forest and neural network). When validating the results obtained using the classification model (0-1 CT versus 2-4 CT), the percentage of correct answers (accuracy) was 90%.

In addition, thanks to the developed method, the need for the use of CT OGK to assess the category of lung tissue changes in COVID-19 is minimized. Determination of the degree of changes in the lung tissue in the express mode allows timely hospitalization of patients with a high probability of severe pneumonia due to the reduction of the diagnostic stage of the outpatient CT center / minimization of the waiting time for the CT scan with adequate use of the resources of the outpatient CT center.

A patient with a confirmed COVID-19 infection undergoes a complex of clinical examination:

1. collecting anamnesis;

2. physical examination;

3.laboratory research:

3.1. clinical blood test with a leukocyte formula;

3.2. clinical analysis of urine.

The listed indicators are estimated either in absolute values or in arbitrary units (Table 1).

Table 1. Assessment of indicators.

Indicator Assessment Physical examination The severity of the patient's condition No symptoms "0 conv units"
Light severity "1 conv unit"
Average severity "2 conv units"
Severe severity "3 conv units" Breathing rate Respiratory movements per minute Shortness of breath or shortness of breath Absence of "0"
Availability "1" Body temperature Degrees C Having a feeling of congestion in the chest Absence of "0"
Availability "1" Having weakness and / or feeling aching Absence of "0"
Availability "1" Having a cough No cough "0 conv unit"
Cough is present "1 conv unit" Type of cough Dry cough "0 conv unit"
Cough with sputum "1 conv unit" Anamnesis Sex of the examinee male "0 conv unit"
female "1 conv unit" Patient's age Years Coronary heart disease Absence of "0 conv unit"
Availability of "1 conv unit" Assignment of the subject to the risk group No "0 conv unit"
The presence of chronic diseases "1 conv unit"
Pregnancy "2 conv units"
Age 65 years older than "3 conv units" Clinical blood test IgG to SARS-CoV-2 coeff. Positive IgM to SARS-CoV-2 coeff. Positive C-reactive protein mg / l Absolute lymphocyte count thousand / μl Absolute granulocyte count (neutrophils + basophils + eosinophils) thousand / μl The absolute number of normoblasts cells / 100 watering can. ESR according to Westergren mm / h D-dimer * ng / ml Reference limit on top of the D-dimer (depending on the test system used) * upper limit of the reference interval Glucose value * mmol / l RDW value % Hematocrit value % P24 antigen HIV-1
and / or
Antibodies HIV-1/2 absent "0 conv unit"
there are "1 conv unit" Clinical analysis of urine Urobilinogen μmol / l Nitrite NIT value:
0- absent
1- weakly positive result +
2- positive ++

* data are taken into account additionally, while the accuracy of the estimate increases.

Based on the markup of the data of all patients, a model was built based on machine learning algorithms ("random forest", a neural network) and validated.

A random forest is a machine learning algorithm that uses an ensemble of decision trees. All trees are built independently according to the following scheme:

• A subsample of the training sample is selected - a tree is built from it (for each tree - its own subsample).

• To construct each split in the tree, look through random features (for each new split - its own random features).

• The best trait and splitting according to it (according to a predetermined criterion) are selected. The tree is built until the sample is exhausted (until the leaves contain only representatives of one class).

Neural networks are trained in two stages:

• Forward propagation of the error;

• Backpropagation of the error.

During forward propagation of the error, a prediction of the response is made. Backpropagation minimizes the error between the actual response and the predicted response.

The survey parameters are used to calculate the classification model, resulting in the following indicators for assessing the severity of pneumonia associated with COVID-19:

• Probability of mild severity (CT 0-1) - the patient does not require hospitalization and can be treated at home.

• Likelihood of moderate to severe pneumonia (CT 3-4) - the patient should be immediately admitted to a hospital for intensive care.

At the same time, CT2 does not apply to either mild or moderate to severe form. Patients with CT2 can be treated at home, and can also be hospitalized during follow-up.

The main objective of the developed method was to identify patients with CT 0-1 (not requiring CT) and patients with CT 3-4 (requiring urgent hospitalization without preliminary examination at the CT center).

• Estimated degree of CT: establishing the exact severity of CT0 / CT1 / CT2 / CT3 / CT4

Each degree denotes an assessment of the involvement of lung tissue (Morozov S.P., Protsenko D.N., Smetanina S.V. et al. Radiation diagnostics of coronavirus disease (COVID-19): organization, methodology, interpretation of results: preprint No. CDT - 2020 - II. Version 2 of 04/17/2020 // Series "Best practices of radiation and instrumental diagnostics." - Issue 65. - M .: GBUZ "NPKTs DiT DZM", 2020.):

<5% - CT0

5-25% - KT1

26-49% - KT2

50-75% KT3

> 75% CT4

We used ROC analysis to assess the quality of the models (Fig. 1, a and b, - ROC curves on test samples for the binary problem (a, b)). The graphs (Fig. 1: a and b) show the models with the best predictive power for KT3-4 and KT0-1, in this case a random forest and a neural network.

Table 2 demonstrates in what percentage of cases each of the predictive classes (0,1,2,3+) was recognized as a real class according to the CT data of the OGK. Thus, KT0 in 60% of cases was recognized as KT0, in 24.6% as KT1, in 5.9% as KT2 and in 3.4% as KT3 +.

Example 1. Patient A, 47 years old. Diagnosis: COVID-19-associated pneumonia. CT 3 (U07.2)

Table 3.

Method CT0 CT1 KT2 KT3 + Classifier 01-24 0.045 0.995 Classifier 02-34 0.2 0.8 Classifier for all CT 0.01 0.09 0.24 0.66

According to the data presented in Table 3, with a probability of 99.5%, the patient has a degree of damage not lower than CT2. The classifiers KT02-KT3-4 and for all grades of CT indicate the highest probability of damage not lower than KT3.

Sign Value Physician severity examination Heavy Breathing rate 22 Have shortness of breath or shortness of breath Present Body temperature 37.5 Chest congestion Present IGG 7.06 IGM 0.72

C-reactive protein 65.1

Floor Male Weakness or aches Present Age 47.0 PCR Not detected Cough Present Type of cough Dry Lymphocyte count 1.3

Granulocyte count 3.4

The number of normoblasts 0.0 ESR according to Westergren 12.0 D-dimer 0.66 Glucose 6.1 Coronary heart disease Not RDW 13.6 Hematocrit 47.0 Risk group Chronicle Antibody / antigen testing HIV 0.0 URO 3.2 NIT 0

To confirm the conclusion obtained with the help of our developed method, the patient underwent CT scan. Conclusion according to CT data of the OGK: COVID-19-associated pneumonia. CT 3. Thus, the conclusions made on the basis of the developed method and the standard - coincided.

Example 2. Patient B, 67 years old. Diagnosis: COVID-19-associated pneumonia. CT 2 (U07.2)

Table 4

Method CT0 CT1 KT2 KT3 + Classifier 01-24 0.026 0.74 Classifier 02-34 0.96 0.04 Classifier for all CT 0.06 0.27 0.58 0.09

According to the data presented in Table 4, with a probability of 74%, the patient has a degree of lesion not lower than CT2. Classifiers KT02-KT3-4 and for all grades of CT indicate the highest likelihood of lung damage corresponding to CT2.

Sign Value Physician severity examination Average Breathing rate eighteen Have shortness of breath or shortness of breath Present Body temperature 36.8 Chest congestion Present IGG 0.65 IGM 0.57

C-reactive protein 65.1

Floor Female Weakness or aches Present Age 67.0 PCR Not detected Cough Present Type of cough Dry Lymphocyte count 1.3

Granulocyte count 3.4

The number of normoblasts 0.0 ESR according to Westergren 12.0 D-dimer - Glucose - Coronary heart disease Not RDW 13.6 Hematocrit 47.0 Risk group Chronicle Antibody / antigen testing HIV 0.0 URO 3.2 NIT 0

To confirm the conclusion obtained with the help of our developed method, the patient underwent CT scan. Conclusion according to CT data of the OGK: COVID-19-associated pneumonia. CT 2. Thus, the conclusions made on the basis of the developed method and the standard - coincided.

Example 3. Patient S., 41 years old. Diagnosis: COVID-19-associated pneumonia. CT 1 (U07.1)

Table 5

Method CT0 CT1 KT2 KT3 + Classifier 01-24 0.97 0.03 Classifier 02-34 1.0 0.0 Classifier for all CT 0.2 0.78 0.02 0.0

According to the data presented in Table 5, with a 97% probability, the patient has a degree of damage not higher than CT1. Classifiers KT02-KT3-4 and for all grades of CT indicate the highest likelihood of lung tissue damage corresponding to CT1.

Sign Value Physician severity examination Easy Breathing rate 19 Have shortness of breath or shortness of breath Absent Body temperature 37.0 Chest congestion Present IGG 1.18 IGM 0.36

C-reactive protein 1.59

Floor Male Weakness or aches Present Age 41.0 PCR Discovered Cough Present Type of cough Dry Lymphocyte count 1.3

Granulocyte count 3.4

The number of normoblasts 0.0 ESR according to Westergren 12.0 D-dimer - Glucose - Coronary heart disease Not RDW 13.6 Hematocrit 47.0 Risk group Chronicle Antibody / antigen testing HIV 0.0 URO 3.2 NIT 0

To confirm the conclusion obtained with the help of our developed method, the patient underwent CT scan. Conclusion according to CT data of the OGK: COVID-19-associated pneumonia. CT 1. Thus, the conclusions made on the basis of the developed method and the standard - coincided.

Claims (3)

1. A method for rapid assessment of changes in lung tissue in COVID-19, including the determination of a set of diagnostically significant indicators, for which the level of biomarkers is measured in a sample of biological fluids - blood and urine obtained from a subject; physical examination; collecting anamnesis, obtaining data on diagnostically significant indicators: the severity of the patient's condition, respiratory rate, shortness of breath or shortness of breath, body temperature, the presence of a feeling of congestion in the chest, the presence of weakness and / or a feeling of aches, the presence of cough, type of cough, gender of the subject, age of the subject, coronary heart disease, assignment of the subject to a risk group: the presence of chronic diseases and / or the presence of pregnancy and / or age 65 years and older, as well as quantitative indicators of IgG and IgM antibodies to SARS-CoV-2, C-reactive protein, absolute lymphocyte count, absolute granulocyte count, absolute normoblast count, Westergren ESR, RDW, hematocrit, HIV-1 p24 antigen and / or HIV-1/2 antibody, urobilinogen, nitrites; the set of the obtained data is processed using at least one classification model trained to determine: the absence of damage to the lung tissue, mild, moderate or severe degree of damage to the lung tissue. 2. The method according to claim 1, characterized in that the “random forest” method is used as the classification models. 3. The method according to claim 1, characterized in that the D-dimer, the upper reference limit for the D-dimer, and glucose are additionally determined.

Images