RU2839398C1 - Method for classification of muscular fatigue based on analysis of wavelet transformation of segments of motor activity of electromyosignal - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: present technical solution relates to the field of medical rehabilitation devices and robotics, in particular to methods and systems for controlling an exoskeleton in the rehabilitation of people with diseases leading to disorders of the musculoskeletal system and musculoskeletal system, as well as for use as human-machine interfaces of industrial exoskeletons. In a known method for assessing muscle fatigue, which consists in obtaining a sEMG signal from a muscle participating in a test functional test or a technological operation, its segmentation according to a certain rule, which enables to form a time sequence of segments, determination of the obtained segments for subsequent study of their energy evolution and constructing descriptors for a muscle fatigue classifier, for muscle fatigue stratification in the current period of the functional test, the wavelet plane of the sEMG segment corresponding to the motor activity interval in the current period of the functional test, and based on its analysis, two vectors of descriptors are formed, the first of which is determined by calculating the global entropy of wavelets in rows (scales) of the wavelet plane, and the second is determined by calculating the global entropy of wavelets in columns of the wavelet plane, wherein the second vector of descriptors includes only components that exceed a threshold value, which is set so that it takes the maximum value of the threshold, which does not lead to omission of indices within the remaining sequence of vector components.

EFFECT: reduction of error in classification of muscular fatigue levels by using machine learning models as classifiers.

3 cl, 10 dwg

Description

This technical solution relates to the field of medical rehabilitation devices and robotics, in particular to methods and systems for controlling an exoskeleton during the rehabilitation of people with diseases leading to disorders of the musculoskeletal system and musculoskeletal system, as well as for use as human-machine interfaces for industrial exoskeletons.

Muscle fatigue is defined as “the inability to maintain the required or expected force.” It is a complex phenomenon observed in everyday life that has generated great interest in the fields of sports, medicine, robotics, and ergonomics. For many patients with neuromuscular disorders, taking muscle fatigue into account is critical in developing effective rehabilitation programs, and fatigue assessment can provide important information about skeletal muscle function and the need for assistive devices in industrial and medical exoskeletons.

Currently, in clinical practice, muscle fatigue is assessed using qualitative rating scales such as the 6-minute walk test (6MWT) [McDonald CM, Henricson EK, Han JJ, Abresch RT, Nicorici A, Elfring GL, Atkinson L, Reha A, Hirawat S, Miller LL. The 6-minute walk test as a new outcome measure in Duchenne muscular dystrophy. Muscle Nerve. 2010; 41(4):500-10. https://doi.org/10.1002/mus.21544] or by subjective questionnaires administered to the patient, such as the Multidimensional Fatigue Inventory (MFI), Fatigue Severity Scale (FSS), and Visual Analogue Scale (VAS) [Lu JS, Weiss MD, Carter G. Assessment and treatment of fatigue in neuromuscular diseases. Am J Hosp Palliat Med. 2010; 27(2):145-57. https://doi.org/10.1177/1049909109358420]. The difficulty in using this and similar scales is due to the significant range of muscle strength values between scores of 4 and 5.

In biotechnical rehabilitation systems, surface electromyography (sEMG) is a non-invasive and widely used method for assessing muscle fatigue. Some characteristics of the sEMG signal can be indicators of muscle fatigue. For example, during submaximal tasks, muscle fatigue will be manifested by a decrease in the velocity and frequency of muscle fibers and an increase in the amplitude of the sEMG signal [Cifrek M, Medved V, Tonkovich S, Ostojic S. Assessment of muscle fatigue in biomechanics based on surface EMG. Clin Biomech. 2009; 24(4):327-40. https://doi.org/10.1016/j.clinbiomech.2009.01.010 ]. The trend and rate of change will depend on the intensity of the task: as a rule, the sEMG amplitude increases during submaximal efforts and decreases during maximal efforts; Moreover, a significantly greater reduction in the frequency content of the signal is observed during maximal efforts compared to submaximal ones [Carr JC, Beck TW, Ye X, Wages NP. Intensity-dependent sEMG response of the biceps brachii during sustained maximal and submaximal isometric contractions. Eur J Appl Physiol. 2016; 116(9):1747–55. https://doi.org/10.1007/s00421-016-3435-6 ]. Accordingly, spectral (e.g., mean frequency) and amplitude parameters (e.g., root mean square deviation (RMS)) of the signals can be used to measure muscle fatigue [Kahl L, Hofmann UG. Comparison of algorithms for quantitative assessment of muscle fatigue in upper limb muscles based on sEMG signals. Med Eng Phys. 2016; 38(11):1260-9. https://doi.org/10.1016/j.medengphy.2016.09.009 ; González-Izal M, Malanda A, Navarro-Amézqueta I, Gorostiaga EM, Mallor F, Ibañez J, Izquierdo M. EMG spectral indices and muscle force fatigue during dynamic contractions. J Electromyogr Kinesiol. 2010; 20(2):233-40. https://doi.org/10.1016/j.jelekin.2009.03.011 ], however, the context of the contraction type and intensity must be specified for correct interpretation. A significant problem with most existing muscle fatigue assessment protocols is that they rely on quantifying the maximal loss of voluntary force, the maximal voluntary muscle contraction (MVC) [Kahl L, Hofmann UG. Comparison of EMG-based muscle fatigue quantification algorithms in upper limb muscles. Med Eng Phys. 2016; 38(11):1260-9. https://doi.org/10.1016/j.medengphy.2016.09.009 ; González-Izal M, Malanda A, Navarro-Amézqueta I, Gorostiaga EM, Mallor F, Ibañez J, Izquierdo M. EMG spectral indices and muscle force fatigue during dynamic contractions. J Electromyogr Kinesiol. 2010; 20(2):233-40. https://doi.org/10.1016/j.jelekin.2009.03.011] or highly fatiguing dynamic tasks [Oliveira ASC, Gonçalves M, Cardozo AC, Barbosa FSS. Electromyographic fatigue threshold of the biceps brachii during dynamic contraction. Electromyogr Clin Neurophysiol. 2005; 45(3):167-75] that cannot be reliably performed in clinical practice, especially in pediatric patients.

During motor activity, a number of muscles are involved in the movement being performed. These muscles are called synergists. Synergists are a group of muscles working in one direction. Muscle synergy is a coordinated local and temporary activity of many muscles that are connected to each other to maintain high performance of a test movement or technological operation. One of the fundamental principles of a dynamic system is self-organization. In other words, when a system combines individual parts of a control process into a whole, its elements behave in an orderly manner. This system can achieve coordinated actions without the need for instructions from a higher center. As muscle fatigue increases, the synchronicity of the synergist muscles is disrupted and synergy decreases. On this basis, a number of technical solutions have been proposed for an objective assessment of muscle fatigue based on the results of multichannel sEMG analysis.

Thus, the essence of the method for monitoring muscle fatigue based on the analysis of synergy patterns [Clinical Model to the Analysis of Synergy Pattern Changes of Back Muscles and its Relationship with the Occurrence of Fatigue/Armin HakKak Moghaddam Torbati, Ehsan Tahami and Hamid Reza Kobravi // The Open Bioinformatics Journal, 2018, 11, 53-60. DOI: 10.2174/1875036201811010052] is that, based on the analysis of sEMG taken from various synergist muscles supporting the performance of the same test movement, descriptors were formed for a trainable classifier that discriminated muscle fatigue at three levels. In the training samples formed on the basis of these descriptors, three clusters of muscle fatigue were identified in a 15-minute test task. Muscle fatigue levels were determined based on experimental studies that recorded changes in sEMG patterns over time. It was experimentally established that muscle sEMG patterns suddenly changed at the ninetieth and six hundredth second of the test muscle load, which were the boundaries of the discriminated clusters.

The main disadvantage of this method is that a time-based approach is used to form muscle fatigue clusters, whereas in rehabilitation biotechnical systems and industrial robots it is necessary to form fatigue clusters based on the ratio of the functional capabilities of the operator or patient and the magnitude of the current muscle load.

Another method based on the principle of synergistic muscle interaction, and which can also be considered as an analogue, is presented in [Method for assessing muscle fatigue based on monitoring synergy patterns and a device for its implementation / Filist S.A., Trifonov A.A., Kuzmin A.A., Safronov R.I., Petrunina E.V. Patent for invention 2766764 C1, 03/15/2022. Application No. 2021105609 dated 03/04/2021]. Muscle synergy patterns in this method, in contrast to the method described above, are formed by forming frequency and amplitude subchannels in each sEMG channel and determining the correlation index of signals in the frequency subchannels of all sEMG channels and the correlation index of signals in the amplitude subchannels of all sEMG channels. The classification of the obtained synergy patterns is carried out by the fuzzy logical inference block, based on the results of which a decision is made to assess muscle fatigue.

The main disadvantage of this method is that it uses multichannel sEMG analysis, whereas in a number of diagnostic and rehabilitation tasks it is necessary to monitor the functional state of a specific muscle.

A method for classifying muscle fatigue based on monitoring the evolution of single-channel sEMG indicators was chosen as a prototype. The implementation of the method includes the following procedures:

1. Data collection (single-channel sEMG recording from the corresponding muscle).

2. Calculate the continuous wavelet transform ( CWT ) of sEMG every 25 seconds. Calculating the CWT every 25 seconds is proposed to meet the minimum requirements for the amount of data that can be handled and the medical requirements that protect the patient's health.

3. Calculate the total wavelet entropy ( TWE ) for the wavelet coefficients at each interval. The definition of TWE is based on the principle of Shannon entropy, defined as

,

where p _i is the probability distribution or relative wavelet energy ( RWE ) at a given scale:

,

where k is the sample number of the wavelet coefficient on the scale with number i .

To quantify the energy gain, the total (global) wavelet entropy ( TWE) was chosen because it performed better than other wavelet energy classifiers. TWE is defined as:

,

where N is the maximum scale used in wavelet analysis.

4. Calculation of the percentage of TWE growth in 25-second intervals (relative to the first 25 seconds).

5. Muscle fatigue levels are determined based on the percentage increase in TWE : an increase of 1%-136% is the first level of fatigue, an increase of 137%-199% is the second level of fatigue, an increase of 200%-280% is the third level of fatigue, and above 281% is the fourth level of fatigue.

This method has been used for biocontrol of muscle electrical stimulation in biotechnical rehabilitation systems, for example, electrical stimulation should be stopped at a muscle fatigue level equal to two [Victoria A. Salazar Herrera, J. Franklin Andrade Romero, Mauricio Amestegui Moreno. Algorithm of detection and alert of muscle fatigue in paraplegic patients, by Digital Signal Proccesing of Surface Electromyogram // IWSSIP 2010 - 17th International Conference on Systems, Signals and Image Processing. Pp. 530-533].

The disadvantages of this method include the lack of the ability to adapt the muscle fatigue scale to the individual physical capabilities of the patient or operator, which, in particular, complicates the formation of data sets for machine learning systems based on sEMG indicators, as well as the fact that this method takes into account only the evolution of wavelets, either only in time or global evolution, i.e. when constructing descriptors, the evolution of wavelets by frequency and by time in the corresponding twenty-five second windows is not differentiated.

The technical objective of the proposed method is to reduce the error in classifying muscle fatigue levels by using machine learning models as classifiers.

The stated task is achieved by the fact that in the known method of assessing muscle fatigue, which consists in obtaining an sEMG signal from a muscle participating in a test functional test or technological operation, segmenting it according to a certain rule that allows forming a time sequence of segments, determining these obtained segments for subsequent study of their energy evolution and constructing descriptors for a muscle fatigue classifier, for stratification of muscle fatigue in the current period of the functional test, the wavelet plane of the sEMG segment corresponding to the interval of motor activity in the current period of the functional test is studied, and based on its analysis, two vectors of descriptors are formed, the first of which 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, while the second vector of descriptors includes only components that exceed the threshold a value set so that it takes on the maximum threshold value that does not result in skipping indices within the remaining sequence of vector components.

In order to form a training sample for training the muscle fatigue classifier from the data set obtained based on the wavelet analysis of sEMG segments, an individual muscle fatigue scale is determined for each patient from the experimental group by means of a two-stage experiment with a test load, at the first stage of which the maximum muscle fatigue is established, determined by the load-time indicator, with the same load on the tested muscle for all patients, and at the second stage, which is carried out after all patients in the experimental group have completed the first stage, the average time to reach the maximum muscle fatigue in the experimental group is determined, after which the test load is established for each patient in such a way that muscle fatigue for each patient reaches its maximum value within the established average time.

Fig. 1 shows a structural diagram of a device implementing the proposed method.

Fig. 2 shows a structural diagram of a myoelectronic reading device using one EMG channel as an example.

Fig. 3 shows the diagrams of the electromyosignal. Fig. 3a is a recording of several periods of motor activity, and Fig. 3b is a segment of the electromyosignal corresponding to the motor activity of the muscle being studied.

Fig. 4 shows the wavelet plane of the electromyosignal, the graphical representation of which is shown in Figure 3b.

Fig. 5 shows timing diagrams illustrating the process of strobe formation.

Fig. 6 shows a diagram of the algorithm for generating two descriptor vectors.

Fig. 7 shows the diagram of the algorithm for reducing the dimensionality of the second vector of descriptors.

Fig. 8 shows the structure of the muscle fatigue classifier built according to the multi-agent ideology.

Fig. 9 shows a general view of the hand with a dynamometer during an experiment to construct a universal scale of muscle fatigue.

Fig. 10 illustrates the process of forming a universal muscle fatigue scale from individual muscle fatigue scales.

The device in Fig. 1 consists of a series-connected myoelectronic reading device (MERD) 1, an analog-to-digital converter (ADC) 2, an RMS calculation unit 3, a strobe generator (SG) 4, a segment formation unit (SFU) 5, a wavelet segment transformation unit (WST) 6, a descriptor formation unit (DFU) 7 and a classifier MU 8. In this case, units 2, 3, 4 and 5 are implemented on the basis of a microcontroller 9, and units 6, 7 and 8 are implemented on the basis of a PC 10.

The myoelectronic reading device Fig. 2 consists of a series-connected block of electrodes 11, a biopotential amplifier 12 and a band-pass filter 13.

The essence of the method is to obtain a wavelet transform ( CWT ) of an sEMG segment that corresponds to the motor activity of the muscle under study in the time interval, followed by obtaining descriptors for use in the muscle fatigue classifier. Figure 3a shows the sEMG signal of a muscle performing a periodic test load. The motor activity interval is highlighted by two vertical red lines. To form descriptors, we use the wavelet transform of this segment, the diagram of which is shown in Figure 3b. The wavelet plane of this segment is shown in Figure 4. The wavelet plane is constructed for the lower frequency of 20 Hz.

Taking into account the logic of the classifier operation, muscle fatigue should be monitored at each period of motor activity of the muscle under study, i.e. in each segment. Three blocks are used to select a segment: RMS 3, FS 4 and BFS 5. BFS selects from the sEMG readings the readings corresponding to the motor activity of the muscles. It has two inputs and one output. The Strobe with FS enters its first input, and the sEMG readings enter its second input. The principle of its operation is illustrated by the sEMG diagrams shown in Fig. 3. The principle of Strobe formation is illustrated in Fig. 5. Fig. 5a shows the RMS signal, according to which U _por. is determined at the level of 0.9RMS(max), and the process of Strobe formation according to RMS and U _por. is shown in Fig. 5b.

To obtain the first group (vector) of descriptors, we determine the values of wavelet entropy at all levels of the segment's wavelet decomposition. That is, one and a half to two thousand readings at each level of decomposition are folded into one component of the descriptor vector using the formula:

, (1)

where j is the decomposition level number or the component index in the descriptor vector,

, (2)

, (3)

CWN ( j , k ) is the amplitude of the wavelet coefficient in the j -th row and k -th column of the wavelet plane, K _j is the number of wavelet coefficients at the j -th level,

. (4)

Thus, the dimension of the first vector of descriptors is determined by the number of detail levels N or, in other words, the number of lines of the wavelet plane. The maximum number of detail levels is determined by the number of samples in the signal K , that is, N _max = K/ 2. N _min or N is directly related to the sampling frequency ( f _S ) of the analyzed signal. In the case of sEMG decomposition sampled with a frequency of f _S = 640 Hz with the lower analysis frequency equal to 20 Hz, the number of decomposition levels N = 200 [DWT analysis of numerical and experimental data for the diagnosis of dynamic eccentricities in induction motors / J. Antonino-Daviu [et al.] // Mechanical Systems and Signal Processing. 2007. Vol. 21. No. 6. P. 2575 - 2589]. Thus, the first vector will have two hundred descriptors determined by formula (1).

To determine the components of the second vector, we use a formula similar to (1), but we determine the energy of the wavelet coefficients by the columns of the wavelet plane:

, (5)

where k is the sample number at the level of detail of the wavelet plane,

, (6)

. (7)

In general, the sEMG segment takes about 1.5...2 seconds and can contain over one and a half thousand readings. However, if we look at the wavelet plane in Fig. 4, we can see that the maximum energy of the spectrum is concentrated in its center. Therefore, having set the threshold value for the energies in the columns E _thr. , we can use only those columns whose energy is higher than the threshold to calculate the components of the second vector.

Fig. 6 shows the algorithm for generating two descriptor vectors for the sEMG segment. In block 14, the wavelet coefficients of the sEMG segment are input. In blocks 15…18, the total energy of the wavelet plane is determined, and in blocks 19…21, the components of the first descriptor vector are determined. In block 22, these components are output as a set .

To determine the second vector of descriptors, blocks 23…28 are used. Blocks 23 and 24 perform calculations according to formula (7). To reduce the dimensionality of the vector of descriptors, block 25 is used, in which the transition from the set is carried out to the set Its algorithm diagram is shown in Fig. 7.

In block 37, the decision maker (DM) specifies the value of E _out , which at the beginning of the algorithm execution can be taken as 0.01 E _out . In blocks 31…34, the left boundary of the set is viewed. and determine the tuple of its elements that do not exceed the threshold value. In blocks 35 and 36, the elements of this tuple are excluded from the set , forming from it its multitude .

In blocks 37…40, the right boundary of the set is viewed. and determine the tuple of its elements that do not exceed the threshold value. In blocks 41 and 42, the elements of this tuple are excluded from the set , forming a new set .

In block 43, the decision maker carries out the selection result and, if necessary (block 44), changes the value of E _por . in block 30 and the selection process is started again.

In block 26 the resulting set is viewed. , based on the elements of which the second vector of descriptors is calculated in blocks 27 and 28. In block 29, it is output.

The muscle fatigue classifier is built on a multi-agent ideology. At the lower level are autonomous intelligent agents that form the space of informative features for autonomous intelligent agents located at the upper hierarchical level. At the upper hierarchical level of the classifier, we use an aggregator that aggregates the decisions of the medical risk classification modules located at the lower hierarchical level. Its structure is shown in Fig. 8.

It includes two independent muscle fatigue classifier modules 47 and 48, which analyze the descriptor vectors (45 and 46). A fuzzy neural network was used as an aggregator (meta-classifier), the tuning algorithms of which are described in detail in [Hybrid method for monitoring muscle fatigue in a robotic system / A.A. Kuzmin, R.A. Tomakova, E.V. Petrunina, D.A. Ermakov, S. Kadyrova // Bulletin of the South-West State University. Series: Control, computing, informatics. Medical instrumentation. 2023. Vol. 13, No. 3. Pp. 64-81. https://doi.org/10.21869/2223-1536-2023-13-2-64-81], at the output of which the final value of the muscle fatigue indicator is formed.

To form a training sample for training the muscle fatigue classifier from the data set obtained based on the wavelet analysis of sEMG segments, an individual muscle fatigue scale is determined for each patient from the experimental group by a two-stage experiment with a test load, at the first stage of which the maximum muscle fatigue is established, determined by the load-time indicator, with the same load on the tested muscle for all patients, and at the second stage, which is carried out after all patients in the experimental group have completed the first stage, the average time to reach the maximum muscle fatigue in the experimental group is determined, after which an individual test load is set for each patient so that muscle fatigue for the patient reaches its maximum value in the established average time. Using this load, a data set is formed for the experimental group, from which training and control samples are formed.

When training the muscle fatigue classifier, the data must be labeled. This means that the corresponding muscle fatigue must be determined for each sEMG segment. Let us give an example of constructing a universal muscle fatigue scale. To study the nature of the sEMG dependence on muscle fatigue, we will record sEMG in the forearm area, where such muscles as the superficial flexor of the fingers, the long palmar muscle, the radial flexor of the wrist, etc. are located. One of the main functions of these muscles is flexion of the hand. To control the magnitude of the load on the hand, an electronic medical dynamometer of the DMER-120 type was used. The general view of the hand with a dynamometer and attached electrodes is shown in Fig. 9.

At the first stage, the subject was given the task of squeezing the dynamometer with such force that certain values were displayed on the screen and held for 5 seconds: 5, 10, 15, etc. dan (*10 Newton). In this way, the biological feedback loop was closed, which allows the registered sEMG parameter to be linked to the force of the dynamometer. The process of increasing the load continued until the subject experienced an “inability to maintain the required or expected force.”

At the conclusion of the first stage, the time of the experiment by the i -th subject T _i was determined according to the formula

, (8)

where F _i is the ultimate holding force of the i -th subject, t is the holding time of a fixed value of the dynamometer reading, - the step of force change on the dynamometer scale.

Then similar experiments were conducted on all members of the experimental group. The average time of maximum muscle fatigue was determined:

, (9)

where N is the number of subjects in the experimental group.

After determining the average time to the onset of maximum muscle fatigue, we can establish an individual muscle fatigue scale for each subject by determining the individual step of force change on the dynamometer scale.

, (10)

or individual holding time of a fixed value of the dynamometer reading

. (11)

Using formulas (10) or (11), we obtain individual scales for muscle fatigue for each subject. To obtain a universal scale of muscle fatigue, the readings on which can be used as Targets in the data set for the machine learning model, the readings on the scales in holding force (formula (10)) and in time readings (formula (11)) must be marked with dimensionless marks. The number of marks on the individual scales is the same for all subjects. The first mark will be 30% of muscle fatigue on the individual scale, the second - 50%, the third 85%.

At the second stage, a data set is obtained for the machine learning model for classifying muscle fatigue by matching the independent variables, represented as a pair of vectors, to the corresponding labels on the individual scales, which correspond to the labels on the universal scale. Fig. 10 illustrates the process of matching the labels on the individual scales and the labels on the universal fatigue scale. As the Goal function, we use the dislocation of muscle fatigue on the individual scale and assign a value from zero to three to the Goal function, depending on which labels on the individual scale the muscle fatigue indicator is located between. As a result, we obtain a table of experimental data, the rows of which store information about each patient of the experimental group, and the columns contain the corresponding independent variables and the Goal function. The machine learning model is trained and tested using the experimental data table obtained in this way.

When implementing the rehabilitation program in a continuous mode with a step of 25 seconds, descriptor vectors are formed and the muscle fatigue class is determined. If the muscle fatigue level corresponds to the level of 1 ... 30% on the universal scale, then it is considered that this is class "0", and a decision on this class is not made, that is, the rehabilitation procedure is carried out in accordance with the rehabilitation program. If the fatigue indicator is within the first and second marks, then the descriptor vectors will correspond to the first class of the fatigue level, which corresponds to the level of increased attention to the patient. If the muscle fatigue indicator is between the second and third marks (class "2"), then electrical stimulation should be stopped, and in the case of using an exoskeleton, servomotors should be turned on to create the corresponding assisting moments or the assisting moments should be increased. If the muscle fatigue indicator is beyond the third mark (class "3"), then the test load on the patient should be removed.

The algorithm implementing rehabilitation monitoring by means of the proposed method has been tested on the Matlab® platform taking into account several tests and various initial conditions representing the most complex cases for managing the rehabilitation process. All the obtained results correctly return the three classes of muscle fatigue.

Thus, a method for classifying muscle fatigue is proposed, based on the analysis of segments of the electromyosignal corresponding to the period of muscle activity of the subject, allowing monitoring the functional capabilities of his muscles in the current state. The method allows forming training and control samples of descriptors, samples in which were obtained from subjects with different values of maximum muscle fatigue based on individual muscle fatigue scales, and creating machine learning models that allow forming universal muscle fatigue scales invariant to the physical capabilities of the subject, and managing rehabilitation procedures.

Claims (3)

1. A method for classifying muscle fatigue based on the analysis of the wavelet transform of segments of the motor activity of an electromyosignal, which consists of obtaining an electromyosignal from muscles participating in a functional test or technological operation, segmenting it based on threshold processing, whereby a time sequence of segments is obtained, determining the wavelet transform of these segments for subsequent study of their energy evolution and constructing descriptors for machine learning models designed to classify muscle fatigue, characterized in that, based on the wavelet transform of an electromyosignal segment corresponding to an interval of motor activity in the current period of the functional test, two vectors of descriptors are formed, the first of which is determined by calculating the total entropy of wavelets in the rows (scales) of the wavelet plane, and the second is determined by calculating the total entropy of wavelets in the columns of the wavelet plane, while the second vector of descriptors includes only the components that exceed a threshold value set in such a way that it does not lead to skipping indices within the remaining sequence of vector components, and after the formation of two descriptor vectors, they are used in a machine learning model to classify the degree of muscle fatigue at the current moment of the functional test. 2. The method according to claim 1, characterized in that in order to form a data set for the machine learning model, an individual muscle fatigue scale is determined for each subject in the experimental group by means of a two-stage experiment with a test load, at the first stage of which the maximum muscle fatigue is established, determined by the load-time indicator, with an equal step of force change on the dynamometer scale and an equal time of maintaining a fixed value of the dynamometer reading for all patients, and at the second stage, which is carried out after all patients in the experimental group have completed the first stage, the average time to reach the maximum muscle fatigue in the experimental group is determined, after which the test load is established for each patient in such a way that muscle fatigue for each patient reaches its maximum value in the established average time, and the divisions obtained on the individual scale are used as an indicator invariant to the physical capabilities of the subjects, the Goal in the training sample for constructing a machine learning model that forms a universal muscle fatigue scale. 3. The method according to paragraphs. 1 and 2, characterized in that a universal muscle fatigue scale is used to control the rehabilitation procedures for motor activity disorders, containing three marks corresponding to four classes of muscle fatigue, wherein if the muscle fatigue level corresponds to the level of 1 ... 30% on the universal scale, then it is assumed that this is class "0", and a decision on this class is not made, and the rehabilitation procedure is carried out in accordance with the rehabilitation program, if the muscle fatigue indicator corresponds to the level of 31 ... 50% on the universal scale, which corresponds to the interval between the first and second marks, then the descriptor vectors will correspond to the "1" class of the fatigue level, which requires increased attention to the patient, if the muscle fatigue indicator corresponds to the level of 51 ... 85%, that is, is between the second and third marks (class "2"), then electrical stimulation should be stopped, and in the case of using an exoskeleton, servomotors should be turned on to create the corresponding assisting moments or the assisting moments should be increased, if the fatigue level has exceeded value of 85% of the universal scale, then the rehabilitation procedure should be terminated.