Method for monitoring functional state of body during treatment and rehabilitation procedures

Abstract

FIELD: medical diagnostics; rehabilitation.SUBSTANCE: methods and systems for managing the rehabilitation of people with disabilities. It can also be used to screen the functional state of the cardiovascular system in individuals at high risk of cardiovascular complications. A method for assessing the control of the functional state of human organs and systems is proposed, which consists of obtaining a cardiac signal at an a priori specified observation aperture, selecting multi-scale windows in it corresponding to the moments of signal observation, transforming signals in these windows and using the transformed signals as descriptors for a functional state classifier; when forming descriptors, half of the signal observation aperture is divided into equal intervals, the beginnings of which are considered to be signal observation points; at each observation point, multi-scale windows are formed and the signal energy density is calculated in each multi-scale window. As a result of calculating the energy density in the windows, an energy density matrix is obtained, the number of rows of which is equal to the number of window scales, and the number of columns is equal to the number of signal observation points. Each column of the energy density matrix is considered as a variation series, the Kolmogorov entropies of these variation series are calculated, a histogram of Kolmogorov entropies is determined, and its readings are used as descriptors for the trained classifier of the patient's functional state. To classify the functional state of patients during therapy or rehabilitation procedures, individual patient classifiers are used, built on fully connected neural networks with linear activation functions in the output layer. To train each patient classifier, two training samples are used; the first sample uses the Kolmogorov entropy histograms obtained when the patient is in a satisfactory condition, and the second sample uses the Kolmogorov entropy histograms obtained when the patient is in an unsatisfactory condition. As a result of training the patient classifier, a two-dimensional cluster space is obtained, formed by the outputs of neurons of the output layer of a fully connected neural network, corresponding to the probabilities of the patient falling into the cluster of satisfactory and the cluster of unsatisfactory functional states, and the current assessment of the effectiveness of therapy or rehabilitation procedure is assessed based on the dynamics of the movement of the current functional state in the two-dimensional cluster space.EFFECT: liability of therapeutic and rehabilitation procedures.2 cl, 8 dwg

Description

[0001] This technical solution relates to the field of medical rehabilitation, specifically to methods and systems for managing the rehabilitation of people with disabilities. It can also be used for screening the functional state (FS) of the cardiovascular system in individuals at high risk of cardiovascular complications.

[0002] To monitor treatment effectiveness, diagnostics should be systematic and include not only clinical observations and symptom assessments but also objective methods for measuring the patient's functional capacity. Such methods may include physical tests, questionnaires, and the use of modern medical devices and technologies.

[0003] Evaluation and prediction of the functional state of the whole organism based on the results of studying the cardiovascular system is based on the fact that hemodynamic changes in various organs and systems occur earlier than the corresponding functional disorders, and the study of the processes of temporal organization, coordination and synchronization of information, energy and hemodynamic processes in the cardiovascular system allows us to identify the earliest changes in the control link of the whole organism. The cardiovascular system with its regulatory apparatus is considered as an indicator of the adaptive reactions of the whole organism, its regulation reflects all levels of control of physiological functions [Baevsky R.M. Theoretical and applied aspects of assessing and predicting the functional state of the body under the influence of long-term space flight factors / / Remarks at the meeting of the Scientific Council of the Russian Federation - IBMP RAS. - Moscow, October 2005].

[0004] Currently, the concept of cardiovascular risk (CVR) assessment is based on the use of scales. The most well-known of these are SCORE, SCORE2, Framingham, etc. [Belyalov F.I. Forecasting and scales in medicine. 3rd ed., revised and enlarged. Moscow: MEDpress-inform, 2020. 248 p.: ill. ISBN 978-5-00030-761-8]. Scales allow one to assess the risk of developing fatal and non-fatal cardiovascular events and clearly evaluate the effect and benefit of preventive therapy by comparing the specific risk in each individual and risk indicators under ideal parameters. However, many researchers agree that scales have a number of limitations that can lead to errors in identifying individuals with high CVR.Thus, patients who do not need preventive measures may be exposed to them, and, conversely, those who really need them may not receive the necessary preventive measures [Prediction of cardiovascular events using proportional hazards models and machine learning models: a systematic review / I. A. Mishkin, A. V. Kontsevaya, A. V. Gusev, O. M. Drapkina / / Scientific and practical peer-reviewed journal "Current problems of health care and medical statistics" 2023, No. 2. - pp. 804-829. DOI 10.24412 / 2312-2935-2023-2-804-829].

[0005] Russian Patent No. 2651708, "Predicting the Risk of a Cardiovascular Event and Its Application," proposes a method for predicting a cardiovascular event over the next five years. Risk is assessed using a set of biomarkers, the values ​​of which are determined through in vitro analysis. Protein biomarkers and their values ​​can be detected and measured using various mass spectrometry methods or the proximity ligation method. The method involves determining the values ​​of at least N biomarkers in a patient's biological sample, the analysis of which indicates whether the individual has an increased risk of developing a cardiovascular event within a five-year period.

[0006] There are many classification methods that can be used to construct a diagnostic classifier from a set of biomarker values. In general, classification methods are most easily implemented using machine learning (ML) concepts, which involves accumulating a dataset using samples obtained from individuals within two (or more, for multi-alternative classification) different groups to be distinguished. Since the class (group or population) to which each sample belongs is known in advance, a classification method can be obtained using various ML models.

[0007] The main drawback of methods based on biomarker analysis is the complete inability to predict cardiovascular complications using screening methods. This is a significant drawback, as it prevents monitoring cardiovascular risk dynamics during treatment and rehabilitation. Therefore, prognostic systems based on time series analysis of electrophysiological signals related to cardiovascular function are preferred over biomarker-based systems.

[0008] The task of time series analysis related to their forecasting is one of the main ones. Its goal is to predict future values ​​of the measured characteristics of the object under study based on observation data. Currently, many methods for predicting the behavior of a dynamic object have been developed. They are divided into local and global methods. Global methods use autoregression, moving average, singular spectrum analysis, etc. Local methods are based on local approximation and do not require a priori information about the system that generated the time series, therefore they have an advantage over others in problems related to forecasting the behavior of irregular series [Loskutov A. / Mathematical Foundations of Chaotic Dynamic Systems. Lecture Course / / Advances in Physical Sciences, 2007, Vol. 177, No. 8. - P. 989-1015].

[0009] A variation series of RR intervals is often used as a time series. Russian Patent No. 2039523 proposes a method for early diagnosis of regular processes in the body's functional systems, including the identification of RR intervals and the analysis of their envelope spectrum in the cardiac, pulmonary, vascular, and metabolic functional zones. The sample contains at least 360 RR intervals, and the spectrum is constructed on a logarithmic frequency scale. Spectral density minima are determined, and the cardiac, pulmonary, vascular, and metabolic functional systems are distinguished using these minima. The peak zone of each system is determined by at least 50 points. The amount of information required to organize this spectrum is calculated, the amount of information is calculated, and the state of regulatory process feedback is judged based on this amount of information.The disadvantage of this method is that, firstly, spectral analysis is carried out without a functional test, which reduces the accuracy of the assessment of the functional state; secondly, when analyzing the functional state, the synchronicity of the body's regulatory systems is not taken into account, that is, spectral analysis ignores the phase characteristics of the spectrum.

[0010] The method presented in [Baevsky R.M., Baranov V.M., Bersenev E.Yu., Funtova I.I., Semenov Yu.N., Grigoriev A.I., Prilutsky D.A. Method for Determining Functional Reserves of Regulation of the Human Cardiorespiratory System / / Patent RU 2240035, priority dated 20.11.2004], makes it possible to assess the functional reserves of the body based on determining the RF of the human cardiorespiratory system by recording the heart rhythm using an ECG and the capillary blood flow velocity of a finger, while simultaneously recording the air flow velocity during breathing. The analysis of the obtained parameters includes the duration of cardiac intervals, pulse wave velocity, and the duration of the respiratory cycle. The degree of synchronization of the parameters of cardiac activity and respiration and the modulus of the average coefficient of mutual correlation are calculated based on the degree and coefficients of their mutual correlation. The study is carried out in three stages.The first is under resting conditions, the second is at a set breathing rate of 6 cycles per minute, and the third is during tests with maximum breath-holding during inhalation and exhalation. The functional capacity of the cardiorespiratory system's regulatory mechanisms is determined by the response to functional tests with different breathing patterns. A disadvantage of this method is its limited applicability to individuals with disabilities.

[0011] The "Method for Assessing the Functional Reserves of the Human Body" is known [Bobrovnitsky I.P., Pavlichenko S.A., Yakovleva M.Yu., Ashmarin E.G., Lebedeva O.D. Method for Assessing the Functional Reserves of the Body (RU Patent No. 2464935, published October 27, 2012)], which allows for obtaining a quantitative integrated assessment of a person's functional reserves based on data from a study of the cardiovascular and respiratory systems at rest and during the Martine stress test, the results of psychophysiological tests, anthropometric measurements, assessment of glucose and cholesterol levels in the blood, and questionnaire data from subjects. The obtained information is processed using a set of algorithms and the results are converted into a 10-point four-level scale, according to which the quantitative assessment of functional reserves is interpreted. Based on this method, a diagnostic technology and the “Health Reserves” automated control system were developed, including a software module for assessing functional reserves [Yakovlev M.Yu., Bobrovnitsky I.P., Lebedeva O.D. Application of a diagnostic software module for monitoring the body's functional reserves to assess the effectiveness of health and rehabilitation programs. / / Issues of balneology, physiotherapy and therapeutic physical education. - 2012. - No. 2. - P. 7-9].

[0012] The closest approach to the proposed method, which was chosen as a prototype, is the method for classifying muscle fatigue described in Russian Patent 2766764. This method is based on the analysis of the variation series obtained from the sampling of an electromyosignal (sEMG). The electromyosignal is collected from muscles participating in a functional test or technological operation. It is then segmented using threshold processing. This results in a time sequence of segments and their wavelet transforms. Descriptors for the ML model are obtained by studying the energy evolution of the wavelet spectra.To stratify muscle fatigue during the current period of the functional test, the wavelet plane of the sEMG segment corresponding to the motor activity interval during the current period of the functional test is analyzed. Based on this analysis, two descriptor vectors are generated. The first is determined by calculating the global entropy of wavelets in the rows (scales) of the wavelet plane, and the second is determined by calculating the global entropy of wavelets in the columns of the wavelet plane. The second descriptor vector includes only components that exceed a threshold value set such that it takes the maximum threshold value without skipping indices within the remaining sequence of vector components. After generating the two descriptor vectors, they are used in the machine learning model to classify the degree of muscle fatigue during the current period of the functional test.The disadvantage of this method, as well as all the above, is that the functional status scales are constructed on the basis of statistical data for a group of patients and do not take into account the individual characteristics of a particular patient, which can lead to significant health problems during treatment or rehabilitation procedures.

[0013] The technical objective of the proposed method is to increase the efficiency and quality of monitoring the functional state of the body during treatment or rehabilitation procedures by taking into account the individual characteristics of the patient.

[0014] The stated objective is achieved by using a known method for assessing the functional state of body organs and systems. This method involves obtaining a cardiac signal at an a priori specified observation aperture, identifying multi-scale windows corresponding to signal observation moments, transforming the signals in these windows, and using the transformed signals as descriptors for a functional state classifier. When forming descriptors, the first half of the signal observation aperture is divided into equal intervals, the beginnings of which are considered signal observation points. Multi-scale windows are formed at each observation point, and the signal energy density is calculated in each multi-scale window. Calculating the energy density in the windows yields an energy density matrix, the number of rows of which is equal to the number of window scales, and the number of columns equal to the number of signal observation points in the observation aperture.Each column of the energy density matrix is ​​considered as a variation series, the Kolmogorov entropies of these variation series are calculated, a histogram of Kolmogorov entropies is determined, and its readings are used as descriptors for the trained classifier of the patient's functional state.

[0015] In order to classify the functional state of patients during therapy or rehabilitation procedures, individual classifiers are used, built on fully connected neural networks with linear activation functions in the output layer. Two training samples are used to train each classifier. The first sample uses histograms of Kolmogorov entropies obtained when the patient is in a satisfactory condition, and the second sample uses histograms of Kolmogorov entropies obtained when the patient is in an unsatisfactory condition.As a result of the classifier training, a two-dimensional cluster space is obtained, formed by the outputs of neurons of the output layer of a fully connected neural network, corresponding to the probabilities of the patient falling into the cluster of satisfactory and the cluster of unsatisfactory functional states, and the current assessment of the effectiveness of therapy or rehabilitation procedure is assessed based on the dynamics of the movement of the current functional state in the two-dimensional cluster space.

[0016] Fig. 1 shows a structural diagram of a device for implementing the method.

[0017] Fig. 2 shows a diagram of the energy density matrix in windows.

[0018] Fig. 3 shows the column structure of the energy density matrix in windows.

[0019] Fig. 4 shows the algorithm diagram of the descriptor calculation module.

[0020] Fig. 5 shows the histograms of the Kolmogorov entropies of a patient in different functional states.

[0021] Fig. 6 shows a two-dimensional cluster space in a square with diagonal coordinates (0,1); (1,0).

[0022] Figure 7 shows Table 1 with an example of the rehabilitation results at some control points for a specific patient in the form of the D and dD indicators, which determine the distance from the patient’s FS image to the image of an unsatisfactory FS in the cluster space.

[0023] Fig. 8 shows Table 2 with the results of assessing the patient’s functional state using the proposed method and the results obtained based on risk stratification in the GRACE scale.

[0024] The device for implementing the method (Fig. 1) contains a patient 1, a cardiac signal sensor 2, a pre-processing unit 3, an observation aperture former 4, an ADC 5, and a memory unit 6, connected in series, with a second input connected to the second output of the observation aperture former 4. Units 3, 4, 5 and 6 are implemented in a microcontroller 6, which is connected to a computer 8 via a wireless communication channel. To implement the method, the computer 8 contains sequentially executed modules: a cardiac signal segmentation module 9, a module for calculating the energy density in windows 10, a module for calculating descriptors 11, a functional state classifier 12 and a module for calculating the trajectories of the functional state 13.

[0025] The method for monitoring the functional state of the body during treatment and rehabilitation procedures is carried out using a device, the structural diagram of which is shown in Fig. 1. The device picks up a cardiac signal from patient 1 via a cardiac signal sensor 2 connected to it, pre-processes the cardiac signal in filter unit 3, forms an observation aperture for the signal via unit 4 and digitizes it using an analog-to-digital converter (ADC) 5. The cardiac signal digitized at the observation aperture is stored in memory unit 6. After digitizing the cardiac signal at a given observation aperture, microcontroller 7 transmits it for further analysis to computer 8.

[0026] The method is implemented as follows. A cardiac signal sensor 2 is connected to patient 1. An electrocardiograph or a pulse oximeter in standard mode can be used as a cardiac signal sensor. The pulse oximeter is used in standard mode, and the photoplethysm signal is read from it. ADC 5 digitizes the cardiac signal processed in pre-processing unit 3, and its readings are sent to memory unit 6. The recording process to memory unit 6 is carried out under the control of microcontroller 7. Microcontroller 7 generates the "Start monitoring" signal, as a result of which monitoring aperture former 4 allows the cardiac signal access to the analog input of ADC 5, and a signal appears at the digital output of monitoring aperture former 4, allowing the recording of digitized signals from the output of ADC 5 to memory 6.

[0027] After digitalizing the cardiac signal at the observation aperture, the microcontroller 7 transfers the data from the memory unit 6 to the computer 8. The segmentation module 9 selects M observation points at the signal observation aperture, located with a uniform step θ along the cardiac signal observation aperture, and L multi-scale windows at each cardiac signal observation point.

[0028] The input information to module 9 is:

[0029] N- number of digital readings on the observation aperture;

[0030] δ - the increment value of the window width w;

[0031] θ - the increment value of the coordinate of the window of width w.

[0032] The output information of this module is:

[0033] number of window scales L, determined by the formula

[0034] , (1)

[0035] where tranc is the operator for selecting the integer part of a fractional number;

[0036] window sizes determined by the formula

[0037] , (2)

[0038] where - window scale index;

[0039] number of window sizes on the signal aperture

[0040] ; (3)

[0041] window coordinate numbers .

[0042] The output data of the segmentation module 9 are the input for the module 10 for calculating the energy density in the windows of the cardiac signal. The output information of the module for calculating the energy density in the windows of the cardiac signal is a matrix of size L×M, each element of which is calculated using the formula

[0043] , (4)

[0044] where - digital readout of cardiac signal with index .

[0045] Figure 2 shows a diagram of the energy density matrix in windows, and Figure 3 shows the structure of a column of the energy density matrix. The diagram shows that the signal aperture spans each window size at M observation points. Therefore, the aperture is divided in half, and observation points are formed only in the first half of the observation aperture.

[0046] The descriptor calculation module operates according to the algorithm shown in Fig. 4. The input data for this module is the energy density matrix, the elements of which are calculated according to formula (4). This matrix is ​​loaded into the module by block 14. The Kolmogorov entropy is calculated in block 17 according to the formula [Korolev O.L. Application of entropy in modeling decision-making processes in economics: monograph / O.L. Korolev, M.Yu. Kussy, A.V. Sigal; edited by prof. A.V. Sigal. - Moscow: INFRA-M, 2022. - 202 p. - (Scientific thought). - DOI 10.12737 / 1865188]

[0047] , (5)

[0048] where - the entropy obtained from the variation series composed of the elements of the column with number τ, calculated in block 16 using the formula

[0049] , (6)

[0050] where - the probability of energy density with value x i in the variation series , obtained from the energy density in column τ, the energy density matrix.

[0051] The Kolmogorov entropy (K-entropy) value defines an estimate of the rate of information loss and can be interpreted as a measure of the system’s “memory,” or a measure of the rate of “forgetting” of the initial conditions [Bobrovnitsky I.P., Lebedeva O.D., Yakovlev M.Yu. Evaluation of the body’s functional reserves and identification of individuals at risk for common diseases / / Issues of balneology, physiotherapy, and therapeutic physical education. - 2011. - No. 6. - P. 40-43].

[0052] To calculate probabilities In block 15, histograms are calculated in each column τ, the energy density matrix according to the formula

[0053] , (7)

[0054] where histogram is an operator that returns a matrix of size bin×2, the first column of which contains the values ​​of the bin intervals of the partition of the energy density range in column τ, and the second column contains the frequency of occurrence of energy density values ​​in these intervals; - column with number τ of the energy density matrix R.

[0055] To move to the probabilities corresponding to the operands in formula (6), it is necessary to divide the elements of the second column of the matrix GR1τ by the total number of elements in the column, that is, by the value L:

[0056] , (8)

[0057] where - the second column of the matrix (7).

[0058] According to formula (7), L matrices (7) are determined, therefore, at the output of block 16 we have L histograms (8). The abscissa axis of the histograms D1 is determined by the formula

[0059] . (9)

[0060] To normalize it, we use the formula

[0061] . (10)

[0062] In block 17, using formulas (5) and (6), we calculate the L-1 Kolmogorov entropies. In block 18, we calculate the histogram of Kolmogorov entropies.

[0063] , (11)

[0064] where K1 is a vector determined by formula (5).

[0065] The abscissa axis of the histogram (11)D2 is determined similarly to formula (10). However, given that (11) can take both positive and negative values, the abscissa axis is normalized using the formula

[0066] , (12)

[0067] where , - operator for determining the absolute value of each element of vector D2.

[0068] Figure 5 shows two examples of such histograms corresponding to two different FS.

[0069] The FS 12 classifier is created for each patient and is based on a fully connected neural network architecture with an input layer containing bin neurons and an output layer containing two neurons. The ReLU function is used as the hidden layer activation function, and the output layer uses linear activation functions. The classifier is trained using two groups of data sets. The first group includes cardiac monitoring signal recordings obtained when the patient's condition is satisfactory, while the second group includes cardiac monitoring signals obtained when the patient's condition is unsatisfactory.

[0070] After training the neural network on the training sample consisting of these two groups of data sets, we have a two-dimensional cluster space, each element of which represents a vector in a two-dimensional space, the coordinates of which are determined by the numbers at the output of the classifier. If their outputs are normalized, we obtain a two-dimensional cluster space in a square with diagonal coordinates (0,1); (1,0), shown in Fig. 6. Considering that the coordinates of a healthy patient should be close to the point (0,1), and the coordinates with unsatisfactory FS to the point (1,0), the effectiveness of treatment or rehabilitation at the control point (CP) can be estimated by the change in the current state of the patient to the point (0,1) and to the point with coordinates (1,0). The cluster space displays module 13. If the parameters v1 and v2 are not weighted, the effectiveness of treatment is estimated by the patient's coordinate in the space of these features. In Fig.6 shows, as an example, possible trajectories of the FS of the same patient in this cluster space at control points (v11,v21) and (v12,v22).

[0071] During treatment or rehabilitation procedures, the classifier outputs change, which determines the dynamics of the functional state vector in the cluster space. Repeatedly determining the distance of the diagnosed point from the points corresponding to the cluster centers is used to assess the effectiveness of the treatment or rehabilitation. The movement of the diagnosed patient's point toward the point corresponding to the image of healthy patients, that is, the point with coordinates (0,1), indicates the effectiveness of the treatment, and vice versa. Different treatment methods will correspond to different speeds of movement of the diagnosed patient's point toward the image of healthy individuals, that is, different trajectories in the cluster space. Having specified the CT, in block 12, the "trajectory" of the patient's FS in the cluster space is calculated in relation to the treatment or rehabilitation plan [Butusov, A.V.Algorithms for monitoring the effectiveness of therapeutic and rehabilitation procedures based on clinical blood test parameters in a medical decision support system / A.V. Butusov, A.V. Kiselev, E.V. Petrunina, R.I. Safronov, V.V. Pesok, A.E. Pshenichny / / Bulletin of the Southwestern State University. Series: Control, Computer Science, Informatics. Medical Instrumentation. - 2023. - Vol. 13, No. 1. - Pp. 170-190]. For example, this trajectory can be calculated in the MATLAB environment. The calculation listing is as follows:.

[0072] B=zscore(A);

[0073] C=[dist(B(1,:),B(3,:)'),dist(B(2,:),B(3,:)')];

[0074] dD=C(2)-C(1),

[0075] where A and B are matrices of input normalized data;

[0076] C(2) - the length of the segment connecting in the cluster space the image of the unsatisfactory FS - coordinate (1,0) and the image of the FS of the monitored patient;

[0077] C(1) - the length of the segment connecting in the cluster space the image of a satisfactory FS - coordinate (0,1) and the current image of the FS of the monitored patient.

[0078] Matrix A consists of three rows. The first contains the image of a satisfactory functional state, the second contains the image of an unsatisfactory functional state, and the third contains the image of the patient's current functional state.

[0079] Repeatedly determining the distance of a diagnosed point from points in the corresponding clusters can be used to assess the effectiveness of treatment or rehabilitation. The movement of a point in the current FS of a rehabilitated patient toward a point corresponding to the FS pattern of healthy patients indicates the effectiveness of treatment. Different treatment plans or rehabilitation programs will also correspond to different rates of movement of the point of the diagnosed patient toward the FS pattern of healthy individuals. By specifying control points, we can calculate the "trajectory" of the patient's FS in cluster space in relation to the disease.

[0080] To experimentally study the effectiveness of a method for monitoring the functional state of the body using Kolmogorov entropy indices as descriptors, we examined 90 men admitted for rehabilitation treatment at the Avicenna Clinical Scientific Medical Center 4 or more months after myocardial infarction. The study included patients in functional classes (FC) I, II, and III according to the New York Heart Association (NYHA) classification. The majority of patients were in FC I and II (NYHA). The average age of the patients was 59.7±0.6 years. Patients received comprehensive treatment, including physical training according to an individualized program. Upon admission to the department and upon discharge from it, all patients underwent clinical, laboratory and instrumental examination according to generally accepted methods, including: electrocardiography (ECG), bicycle ergometry (BEM), Holter monitoring (HM) ECG.

[0081] A team of highly qualified experts classified patients into FS clusters. Additional indicators were also used. Based on the results of examination of patients in the experimental group and the retrospective analysis data, 90 neural network classifiers of FS were trained.

[0082] When developing the physical training program, both clinical data and the results of instrumental examinations, such as bicycle ergometry and ECG, were taken into account. The training load was calculated based on each patient's individual threshold power (ITP) and ranged from 30% of ITP at the initial stage of training to 50% of ITP at the final stage. Accordingly, the increase in heart rate (HR) during training should not exceed 30-50% of the increase in HR at the ITP altitude, as determined by bicycle ergometry upon admission to the center.

[0083] An example of the rehabilitation results at some control points for a specific patient in the rehabilitation process from the experimental group is presented in Table 1 in Fig. 7 in the form of the D and dD indicators, which determine the distance from the patient's FS image to the images of an unsatisfactory FS and a satisfactory FS in the cluster space.

[0084] Table 1 shows that the dD indicator significantly decreased during the rehabilitation period, which is due to the inadequate load prescribed by the treating physician. The IPM was then adapted to the patient's physical condition, and the increase in dD indicator indicates the effectiveness of the chosen rehabilitation program.

[0085] The proposed FS monitoring during rehabilitation will allow the physician to more objectively assess the effectiveness of treatment and make a diagnosis based on simpler tests. This suggests that using the resulting system for monitoring the treatment and rehabilitation process is appropriate.

[0086] Using the various indicators obtained from the output of the trained classifier, characterizing the patient's FS, as scales for constructing histograms of the distribution of classes ω1 and ω2, classification thresholds were determined that made it possible to solve the problem of ROC analysis.

[0087] The results of the obtained FS model were compared with the results of the proposed classifier and the results obtained based on risk stratification in the GRACE scale (Table 2, Fig. 8). The classifier model on the control sample demonstrated very good predictive ability (AUC = 0.88), with a sensitivity of 75%; specificity of 72%.

[0088] The classification quality indicators of the synthesized classifier allow us to recommend it for intelligent support of monitoring the FS of patients during treatment and rehabilitation procedures with exogenous influences.

Claims

1. A method for monitoring the functional state of an organism during treatment and rehabilitation procedures, which consists of obtaining a cardiac signal at an a priori specified observation aperture, selecting multi-scale windows on it, transforming the signals in the windows at specified observation points and using the transformed signals as descriptors for a functional state classifier, characterized in that when forming descriptors, half of the signal observation aperture is divided into equal intervals, the beginnings of which are considered to be signal observation points, multi-scale windows are formed at each observation point and the signal energy density is calculated in each multi-scale window, an energy density matrix is ​​obtained, the number of rows of which is equal to the number of window scales, and the number of columns is equal to the number of signal observation points, each column of the energy density matrix is ​​considered as a variation series, the Kolmogorov entropies of these variation series are calculated,The Kolmogorov entropy histogram is determined, and its readings are used as descriptors for the trained classifier of the patient's functional state.

2. The method according to paragraph 1, characterized in that for classifying the functional state of patients during the course of therapy or rehabilitation procedure, individual classifiers are used, built on fully connected neural networks containing as many neurons in the input layer as the number of intervals of partitioning the dynamic range has a histogram of the Kolmogorov entropy, and two neurons in the output layer, having linear activation functions, for training each classifier, two training samples are used, the elements of the first sample are the histograms of the Kolmogorov entropies obtained when the patient is in a satisfactory condition, and the second training sample is the histograms of the Kolmogorov entropies obtained when the patient is in an unsatisfactory condition, as a result of training the classifier, a two-dimensional cluster space is obtained, formed by the outputs of the neurons of the output layer of the fully connected neural network,corresponding to the certainty of the patient's inclusion in the cluster of satisfactory and the cluster of unsatisfactory functional states, and the current assessment of the effectiveness of therapy or rehabilitation procedures is assessed based on the dynamics of the movement of the current functional state in a two-dimensional cluster space.

Images