A muscle tone rating and abnormal muscle dynamic positioning system and method

Abstract

The application discloses a muscle tension rating and abnormal muscle dynamic positioning system and method, and the muscle tension rating and abnormal muscle dynamic positioning system comprises an information acquisition module, an information processing module and a man-machine interaction module, the information acquisition module is used for collecting personal information and surface electromyogram signals of a detected person, the information processing module is used for muscle tension rating, determination of abnormal movement types and positioning of abnormal muscles, the man-machine interaction module is used for information management, evaluation result display, supervision and tracking and input of diagnosis information, and the like. The application can automatically identify muscle coupling abnormalities of a patient in various movement states, analyze internal muscle factors causing movement symptoms of the patient, evaluate a muscle tension disorder level of the patient and assist in positioning of abnormal muscles.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of explainable artificial intelligence, in particular to a muscle tone rating and abnormal muscle dynamic positioning system and method. BACKGROUND

[0002] Muscle tone refers to the mutual traction force between cells inside the muscle, which is the tightness of the muscle motor unit and is the basis for maintaining body posture and movement. Muscle tone disorder is a movement disorder caused by abnormality of the nervous system, and its incidence rate is only second to Parkinson's disease. Patients often have task-specific persistent or intermittent movement or posture abnormalities, which seriously endanger their living standards and may pose a major challenge to families and society, requiring comprehensive medical and social support to alleviate its impact.

[0003] In the clinical diagnosis of muscle tone disorder, quantitative assessment of the movement symptoms of muscle tone disorder has important clinical application value for diagnosis, treatment and rehabilitation. However, current clinical diagnosis and treatment often rely on physician experience or use clinical scales for grading assessment. These methods can provide some quantitative reference for disease assessment, but are easily affected by subjective factors of the tester, and the assessment results are ambiguous and lack of details.

[0004] In recent years, with the development of movement assessment technology, more and more new technologies focus on applying sensing technology and movement analysis to the quantitative assessment of muscle tone disorder. For example, patent No. 201811534539.X discloses a muscle tone assessment method and device, which uses an ultrasonic diagnostic device in the module to measure the elastic modulus of skeletal muscle at different joint angles, and assesses the muscle tone level by analyzing the change trend of the joint range of motion and the elastic modulus. Patent No. 202111001080.9 discloses another muscle tone level assessment method and device, which uses a surface electromyography sensor in the device to obtain the electromyography time domain and frequency domain features of the muscles at the wearing position, and then inputs a preset assessment model to output the muscle tone level. In fact, task-specific involuntary movement or posture is a typical feature of the movement symptoms of muscle tone disorder, which is manifested at the muscle level as abnormal co-activation of agonist and antagonist muscles induced by specific tasks. Quantitative analysis of this can make the assessment directly to the essence. However, the above assessment methods have the problem of single test method or assessment means, and do not consider activating the task-specific movement symptoms of muscle tone disorder patients through multiple dynamic tasks, and do not analyze the internal muscle abnormal factors of the movement symptoms of patients, so there are defects in the comprehensiveness and pertinence of quantitative assessment.

[0005] In addition, there is a great space for improvement in the treatment of dystonia. For example, in addition to oral drugs and neurosurgical treatment, abnormal muscle tone can be inhibited by injecting botulinum toxin into the lesion site to relieve symptoms in patients, and this botulinum toxin A injection method has good effect on the treatment of focal or segmental dystonia. However, the positioning of abnormal muscles often depends on the clinical inquiry, palpation of the physician, or the use of invasive electromyographic sensors to assist positioning in a static state, the former is highly subjective, and the latter can cause invasive damage while limiting the patient to remain static. However, dystonia has task-specificity, that is, the abnormal symptoms of patients are usually triggered when performing special motor activities, which is particularly evident in the early and middle stages of the disease, so that the static diagnostic method has the risk of missed diagnosis or misdiagnosis. In summary, the existing method cannot locate the abnormal muscles of the patient during movement, and there is a lack of automatic identification method guided by a model. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a muscle tension rating and abnormal muscle dynamic positioning system and method, which can automatically identify the muscle coupling abnormalities of patients in various motion states, analyze the internal muscle factors causing the symptoms of patients in motion, and evaluate the level of muscle tension disorder and assist in positioning the abnormal muscles.

[0007] The technical scheme of the present application is:

[0008] A muscle tension rating and abnormal muscle dynamic positioning system, comprising an information acquisition module, an information processing module and a man-machine interaction module, the information acquisition module comprising a personal information acquisition and input unit and a surface electromyographic signal acquisition unit, the information processing module being arranged on a cloud server, the information processing module comprising a data storage unit, a signal preprocessing unit, a motion recognition and segmentation unit, a muscle tension rating unit and an abnormal muscle positioning unit, the man-machine interaction module comprising a test guidance unit, an information management unit, a result display unit, a supervision and tracking unit and a diagnosis information input unit.

[0009] The personal information acquisition and input unit sends the acquired and input personal information to the information management unit, the information management unit checks, operates and manages the personal information through the touch display screen, and the man-machine interaction module sends the personal information to the information processing module, and the data storage unit stores the personal information, and the information management unit calls and manages the personal information stored in the data storage unit.

[0010] The test guidance unit performs a guidance demonstration of the muscle tension test. During the test, the surface electromyographic signal acquisition unit sends the collected surface electromyographic signal to the information processing module, the signal preprocessing unit performs signal preprocessing on the surface electromyographic signal, the surface electromyographic signal after preprocessing is sent to the motion recognition and segmentation unit for motion recognition and segmentation, and then the segmented electromyographic time sequence signal is input to the muscle tension rating unit for motion performance scoring, muscle inhibition abnormality degree rating and muscle tension rating, to obtain the motion performance score, muscle inhibition abnormality degree rating result and muscle tension rating result, and then the abnormal muscle positioning unit performs abnormal muscle positioning and abnormal motion type determination to obtain the position of the muscle pair with abnormal co-activation and the abnormal motion type;

[0011] The information processing module sends the motion performance score, muscle inhibition abnormality degree rating result and muscle tension rating result obtained after processing by the muscle tension rating unit to the human-computer interaction module, which is displayed on the touch display screen through the result display unit;

[0012] The information processing module sends the position of the muscle pair with abnormal co-activation and the abnormal motion type obtained after processing by the abnormal muscle positioning unit to the human-computer interaction module, which is displayed on the touch display screen through the result display unit;

[0013] Meanwhile, the motion performance score, muscle inhibition abnormality degree rating result and muscle tension rating result obtained after processing by the muscle tension rating unit, and the position of the muscle pair with abnormal co-activation and the abnormal motion type obtained after processing by the abnormal muscle positioning unit are all sent to the data storage unit, so as to realize storage of the detection result information;

[0014] The diagnosis scheme issued and entered by the diagnosis information input unit is also sent to the data storage unit, so as to realize storage of the diagnosis information;

[0015] The supervision and tracking unit manages and displays the multiple evaluation results stored according to the timeline in the data storage unit through the touch display screen.

[0016] The surface electromyographic signal acquisition unit selects a surface electromyographic sensor or a wearable device embedded with a surface electromyographic sensor.

[0017] A muscle tension rating and abnormal muscle dynamic positioning method, specifically comprising the following steps:

[0018] (1) Information collection: first, the personal information of the detected person is entered, and then the surface electromyographic signal acquisition unit is connected to the skin surface of the detected person's limb, and the detected person completes the prescribed limb test task according to the requirements demonstrated by the test guidance unit, and when the corresponding action of the limb test task is completed, the surface electromyographic signal acquisition unit collects the surface electromyographic signal of the detected person and sends it to the information processing module;

[0019] (2), signal processing:

[0020] S21, the signal preprocessing unit of the information processing module first performs signal preprocessing on the received surface electromyography signal;

[0021] S22, the preprocessed surface electromyography signal is input to the motion recognition and segmentation unit, the motion recognition and segmentation unit first identifies the motion type using a human motion recognition model, and then uses a motion segmentation model to segment the time sequence window corresponding to different motions, so that each type of motion corresponds to a segment of electromyography time sequence signal. A segment of electromyography time sequence signal after segmentation is synchronously input to the muscle tension rating unit and the abnormal muscle positioning unit of the information processing module;

[0022] S23, the muscle tension rating unit first extracts signal time, frequency features and motion macro features from the segmented electromyography time sequence signal using a motion performance evaluation model, and then scores the motion performance according to the signal time, frequency features and motion macro features. Then the muscle tension rating unit uses a muscle inhibition abnormality degree grading evaluation model to score the muscle inhibition abnormality degree; Finally, the muscle tension rating unit inputs the motion performance score and the muscle inhibition abnormality degree score into the muscle tension fusion evaluation model for muscle tension rating. The muscle tension rating is from one to five, and the higher the rating, the more serious the muscle tension disorder;

[0023] S24, the abnormal muscle positioning unit uses an abnormal muscle positioning model to calculate the fusion index filtered out by the muscle inhibition abnormality degree grading evaluation model to obtain the main coupling characteristic index. The motion represented by the main coupling characteristic index is an abnormal motion type, and the abnormal parameter is determined according to the fusion index corresponding to the main coupling characteristic index. The muscle pair corresponding to the abnormal parameter is the abnormal muscle, so as to determine the position of the muscle pair with abnormal coactivation;

[0024] S25, the motion performance score, the muscle inhibition abnormality degree rating result, the muscle tension rating result, the muscle pair position with abnormal coactivation and the abnormal motion type data constitute the processing result, and the data storage unit of the information processing module stores the processing result;

[0025] (3), result display: the information processing module synchronously sends the processing result to the man-machine interaction module for display; the physician issues and enters the diagnosis scheme according to the processing result, and manages the multiple evaluation results stored according to the time line.

[0026] The surface electromyography signal acquisition unit is connected to the skin surface of the detected person's limbs, and the specific connection positions are the rectus femoris muscle, the semitendinosus muscle, the tibialis anterior muscle, the long peroneal muscle, the medial gastrocnemius muscle and the soleus muscle of the detected person; wherein there are eight pairs of agonist and antagonist muscle pairs, which are the tibialis anterior muscle-long peroneal muscle, the tibialis anterior muscle-medial gastrocnemius muscle, the tibialis anterior muscle-soleus muscle and the rectus femoris muscle-semitendinosus muscle of the bilateral legs.

[0027] The limb test task includes single joint test and multi-joint composite test; the single joint test requires the detected person to sit on a seat, and the bilateral ankle joints perform three isometric contractions of dorsiflexion, plantar flexion, inversion and eversion, respectively; the multi-joint composite test requires the detected person to perform three tests, each test starts in a sitting position, after receiving a start instruction, stands up from the seat, keeps standing still for a few seconds, then switches from the standing state to the walking state and starts straight walking, turns after walking a set distance and returns to the initial position, turns again and walks back and forth for multiple groups, turns at the starting position after walking and stands still for a few seconds, and then sits back on the seat.

[0028] The signal preprocessing is specifically band-pass filtering, de-meaning, high-pass filtering, filtering out power frequency interference and filtering out white noise on the collected surface electromyography signal, so as to obtain the preprocessed surface electromyography signal.

[0029] The processing process of the motion recognition and segmentation unit is specifically:

[0030] S221, the preprocessed surface electromyography signal is segmented into sliding window data segments according to a set time span;

[0031] S222, time domain features, frequency domain features and time-frequency domain features are extracted from the sliding window data segments; the time domain features include standard deviation, absolute mean, interquartile range, peak-to-peak value, skewness, kurtosis, Wilson amplitude, zero-crossing coefficient and regression coefficient signal energy; the frequency domain features are parameters obtained after Fourier transform of the surface electromyography signal, including average power frequency, median frequency, quartile frequency, top three values of spectral density and power spectrum entropy; the time-frequency domain features are parameters obtained after wavelet transform of the surface electromyography signal, i.e. the energy of each layer wavelet coefficient; then, the features that significantly affect the clinical scale score of any motion performance are selected from the time domain features, the frequency domain features and the time-frequency domain features, and a machine learning method is used to select a feature set that simultaneously satisfies a small number of features and good classification results from the selected features;

[0032] S223, the features of each sliding window data segment are input into the trained human motion recognition model, and the motion type represented by each sliding window data segment is output, wherein the human motion recognition model is learned and recognized by using a random forest machine learning algorithm;

[0033] S224, splicing the sliding window data segments in time sequence, extracting two sliding window data segments adjacent in time sequence in which the motion type changes, composing a transition data segment, reconstructing all the extracted transition data segments in time as a new signal data segment as a motion category transition data segment, inputting the motion category transition data segment into the trained motion segmentation model, outputting the time of motion type transition, taking the time of motion type transition as the start and end time of each electromyography time sequence signal, segmenting the preprocessed surface electromyography signal according to the motion type, so that each electromyography time sequence signal corresponds to one motion type.

[0034] The processing process of the muscle tension rating unit is specifically:

[0035] S231, first, extract signal time and frequency features and motion macro features from a segmented electromyography time sequence signal, the signal time and frequency features are new features extracted from time domain features, frequency domain features and time-frequency domain features after dimension reduction, and the motion macro features are the time of completing the sit-to-stand transition and the gait time features of the detected person; then input the signal time and frequency features and the motion macro features into the trained motion performance evaluation model, output the motion performance score, the motion performance score includes the ankle joint activity function score, the posture conversion function score, the standing stability function score and the walking function score of the detected person, each function score is from one to five, and the higher the rating result is, the more serious the dysfunction is;

[0036] S232, first, pair the segmented electromyography time sequence signal into two-channel electromyography data according to the pairing of agonist and antagonist muscles, to obtain eight pairs of paired data; then extract intermuscular coupling parameters from the eight pairs of paired data, including time domain coupling parameters and frequency domain coupling parameters, the time domain coupling parameter refers to the electromyography signal time domain co-activation area CL of the agonist and the antagonist muscle, the calculation formula is as follows:

[0037]

[0038] In formula (1), env1 and env2 represent the surface electromyography signal envelopes of the agonist and the antagonist muscle respectively, ∫min(env1, env2) represents the area of the overlapping part of the two surface electromyography signal envelopes, and ∫max(env1, env2) represents the area of the outer contour part of the two surface electromyography signal envelopes;

[0039] The calculation formula of the frequency domain coupling parameter Arr is as follows:

[0040] Arr=∫IMC XY |(IMC XY >σ IMC ) (2).

[0041] IMC IMC is expressed as summing all IMC XY greater than a significant coherence threshold σ XY , IMC IMC is the ratio of the cross-spectral density of agonist and antagonist signals and the auto-spectral density, σ IMC is the significant coherence threshold, and the formula for calculating σ

[0042]

[0043] In formula (3), N IMC is the number of sliding windows of each EMG time series signal, and α = 0.05 is the significance level.

[0044] Then, the fusion indicators are extracted from the intermuscular coupling parameters, including the normalized absolute mean, the normalized maximum, and the normalized minimum. First, the intermuscular coupling parameters of the control group converted from the healthy people are standardized, and the formula is shown in formula (4):

[0045]

[0046] In formula (4), IC represents the intermuscular coupling parameters after standardization, IC is the intermuscular coupling parameters to be standardized, and are the mean and standard deviation of the intermuscular coupling parameters of the healthy control group, respectively.

[0047] Then, on the basis of formula (4), the fusion indicators are extracted, and the fusion objects of the indicators are the intermuscular coupling parameters extracted from the eight pairs of paired data. For the intermuscular coupling parameters after standardization, the normalized absolute mean is the absolute mean of the intermuscular coupling parameters after standardization extracted from all eight pairs of paired data, the normalized maximum is the mean of the three largest data in the intermuscular coupling parameters after standardization extracted from all eight pairs of paired data, and the normalized minimum is the mean of the three smallest data in the intermuscular coupling parameters after standardization extracted from all eight pairs of paired data.

[0048] Finally, the screened fusion indicators are input into the pre-trained muscle inhibition abnormality degree grading evaluation model, and the muscle inhibition abnormality degree score of the detected person is output. The muscle inhibition abnormality degree score is from one to five, and the higher the evaluation result is, the more serious the muscle abnormality degree is.

[0049] S233, input the sports performance score and the muscle inhibition abnormality degree score into the trained muscle tension fusion evaluation model, and output the muscle tension evaluation result of the detected person. The muscle tension evaluation is from one to five, and the higher the evaluation result is, the more serious the muscle tension disorder degree is.

[0050] The training of the movement performance evaluation model, the muscle inhibition abnormality degree grading evaluation model, and the muscle tension fusion evaluation model is first to establish a sample data set, the label of the sample data is a movement performance clinical scale score or a muscle tension disorder clinical scale score, then, a training set is randomly selected from the sample data set and composed; resampling is performed on the samples in the training set, so that the number of samples of each label tends to be the same; the resampled training set is classified and the classification target is the label of the detected person corresponding to each sample, and the classification model obtained by training is the trained movement performance evaluation model, muscle inhibition abnormality degree grading evaluation model, or muscle tension fusion evaluation model.

[0051] The processing process of the abnormal muscle positioning unit is specifically:

[0052] S241, a model explanation technology is used to calculate a contribution degree parameter of the fusion index to the muscle inhibition abnormality degree grading evaluation model output result, the contribution degree parameter is a shapley value of an input model feature, for any feature x i , the shapley value is calculated according to the following formula (5):

[0053]

[0054] In formula (5), M S is a complete set of input features, #M S is a base of M S , Su S is a feature subset of M i not containing feature x S , #Su S is a base of Su S , v(Su S ) is a model output of Su S , v(Su S ∪{i}) is a model output after feature x i is added to Su S , and! represents a factorial; the model to be explained is a muscle inhibition abnormality degree grading evaluation model, the input model feature refers to a fusion index related to a muscle tension disorder clinical scale score selected by using a statistical method, and the complete set of features refers to a complete set of all input model features;

[0055] S242, the absolute value of the shapley value is calculated, and outliers in the absolute values are searched to locate the main coupling characteristic index: when there is an outlier, the index represented by the outlier is the main coupling characteristic index; if there is no outlier, the top three largest absolute values are selected as the main coupling characteristic index.

[0056] S243. The movement represented by the main coupling characteristic indicator is an abnormal movement type of the subject. When the fusion indicator corresponding to the main coupling characteristic indicator is the standardized absolute average value, then according to the three-fold variance principle, the parameter with an absolute value greater than 3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the muscle pair corresponding to the abnormal parameter is an abnormal muscle, that is, the position of the muscle pair with abnormal co-activation is known; when the fusion indicator corresponding to the main coupling characteristic indicator is the standardized maximum value, then according to the three-fold variance principle, the parameter with a value greater than 3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the muscle pair corresponding to the abnormal parameter is an abnormal muscle; when the fusion indicator corresponding to the main coupling characteristic indicator is the standardized minimum value, then according to the three-fold variance principle, the parameter with a value less than -3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the corresponding muscle pair is an abnormal muscle.

[0057] Advantages of the present invention:

[0058] (1) When the present invention is used, the patient, i.e. the person being tested, can undergo a muscle tension rating test under the guidance of the human-computer interaction module. The collected surface electromyography information is then analyzed in real time by the information processing module in the cloud server, and the results are displayed on the human-computer interaction module. The entire testing process is fast, and the doctor can retrieve multiple evaluation results of the patient stored according to the timeline from the information processing module in the cloud server, so that the doctor can timely grasp the current status and past progress of the patient's muscle tension rating and abnormal muscle information.

[0059] (2) The muscle tension rating unit of the present invention comprehensively evaluates the external motor symptoms of the patient's dystonia and the intrinsic muscle factors that cause these external motor symptoms, wherein the external motor symptoms are manifested as the motor performance score of the patient's main limb functions, and the intrinsic muscle factors are manifested as the rating results of the abnormal degree of muscle inhibition of the patient's corresponding limbs, so that doctors can quickly and accurately formulate treatment and rehabilitation plans.

[0060] (3) The abnormal muscle positioning unit of the present invention realizes automatic monitoring of the patient's abnormal muscles under dynamic conditions, so that the doctor can quickly and accurately grasp the abnormal muscle position that causes the patient's dystonia and the type of movement that induces dystonia, so that the doctor can quickly and accurately formulate a treatment and rehabilitation plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a principle block diagram of the muscle tension rating and abnormal muscle dynamic positioning system of the present invention.

[0062] Figure 2 It is a flow chart of the muscle tension rating and abnormal muscle dynamic positioning method of the present invention. DETAILED DESCRIPTION

[0063] With reference to the drawings and embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0064] See Figure 1 A muscle tension rating and abnormal muscle dynamic positioning system comprises an information acquisition module 1, an information processing module 2 and a man-machine interaction module 3. The information acquisition module 1 comprises a personal information acquisition and input unit 1111 and a surface electromyogram signal acquisition unit 12. The surface electromyogram signal acquisition unit 12 uses a surface electromyogram sensor or a wearable device embedded with a surface electromyogram sensor. The information processing module 2 is arranged on a cloud server. The information processing module 2 comprises a data storage unit 21, a signal preprocessing unit 22, a motion recognition and segmentation unit 23, a muscle tension rating unit 24 and an abnormal muscle positioning unit 25. The man-machine interaction module 3 comprises a test guidance unit 31, an information management unit 32, a result display unit 33, a supervision and tracking unit 34 and a diagnosis information input unit 35.

[0065] The personal information acquisition and input unit 11 sends the acquired and input personal information to the information management unit 32. The information management unit 32 checks, operates and manages the personal information through a touch display screen. Meanwhile, the man-machine interaction module 3 sends the personal information to the information processing module 2. The data storage unit 21 stores the personal information. Meanwhile, the information management unit 32 calls and manages the personal information stored in the data storage unit 21.

[0066] The test guidance unit 31 demonstrates the muscle tension test. During the test, the surface electromyogram signal acquisition unit 12 sends the acquired surface electromyogram signal to the information processing module 2. The signal preprocessing unit 22 performs signal preprocessing. The preprocessed surface electromyogram signal is sent to the motion recognition and segmentation unit 23 for motion recognition and segmentation. Then, the segmented electromyogram time sequence signal is input to the muscle tension rating unit 24. The muscle tension rating unit 24 performs motion performance scoring, muscle inhibition abnormality degree rating and muscle tension rating to obtain the motion performance score, muscle inhibition abnormality degree rating result and muscle tension rating result. Then, the abnormal muscle positioning unit 25 performs abnormal muscle positioning and abnormal motion type determination to obtain the position of the muscle pair with abnormal co-activation and the abnormal motion type.

[0067] The information processing module 2 sends the motion performance score, muscle inhibition abnormality degree rating result and muscle tension rating result obtained by the muscle tension rating unit 24 to the man-machine interaction module 3. The result display unit 33 displays the results on the touch display screen.

[0068] The information processing module 2 sends the abnormal muscle positioning unit 25 processed results of the existence of abnormal co-activation muscle pair position and abnormal movement type to the human-computer interaction module 3, and displays it on the touch display screen through the result display unit 33;

[0069] At the same time, the movement performance score processed by the muscle tension rating unit 24, the muscle inhibition abnormality degree rating result and the muscle tension rating result, and the existence of abnormal co-activation muscle pair position and abnormal movement type processed by the abnormal muscle positioning unit 25 are all sent to the data storage unit 21, so as to realize the storage of the detection result information;

[0070] The diagnosis information input unit 35 also sends the diagnosis scheme to the data storage unit 21, so as to realize the storage of the diagnosis information;

[0071] The supervision tracking unit 34 manages and displays the multiple evaluation results stored in the data storage unit 21 according to the time line through the touch display screen.

[0072] See Figure 2 A muscle tension rating and abnormal muscle dynamic positioning method, specifically comprising the following steps:

[0073] (1) Information collection: first, input the personal information of the detected person, and connect the surface electromyography signal acquisition unit 12 to the skin surface of the detected person's limbs, and the specific connection position is the rectus femoris muscle, the semitendinosus muscle, the tibialis anterior muscle, the peroneus longus muscle, the medial gastrocnemius muscle and the soleus muscle of the detected person; There are eight pairs of agonist and antagonist muscles, respectively, the tibialis anterior muscle-peroneus longus muscle, tibialis anterior muscle-medial gastrocnemius muscle, tibialis anterior muscle-soleus muscle and rectus femoris muscle-semi-tendinosus muscle of the bilateral legs, then the detected person completes the prescribed limb test task according to the requirements demonstrated by the test guidance unit 31, when the corresponding action of the limb test task is completed, the surface electromyography signal acquisition unit 12 collects the surface electromyography signal of the detected person and sends it to the information processing module 2;

[0074] The limb test task includes single-joint test and multi-joint composite test; The single-joint test requires the detected person to sit on a seat, and the bilateral ankle joints are respectively subjected to three isometric contractions of dorsiflexion, plantar flexion, inversion and eversion; The multi-joint composite test requires the detected person to perform three tests, each test starts with a sitting position, after receiving the start instruction, stands up from the seat, keeps standing still for a few seconds, then switches from the standing state to the walking state and starts straight walking, turns after walking a set distance and returns to the initial position, turns again and walks back and forth for multiple groups, turns and stands still at the starting position after walking, and then sits back on the seat;

[0075] (2) Signal processing:

[0076] S21, the signal preprocessing unit 22 of the information processing module 2 first performs signal preprocessing on the received surface electromyogram signal, specifically, band-pass filtering, de-meaning, high-pass filtering, filtering out power frequency interference and filtering out white noise, so as to obtain a preprocessed surface electromyogram signal;

[0077] S22, the preprocessed surface electromyogram signal is input to the motion recognition and segmentation unit 23, and the specific processing process is as follows:

[0078] S221, the preprocessed surface electromyogram signal is divided into sliding window data segments according to a set time span;

[0079] S222, time domain features, frequency domain features and time-frequency domain features are extracted from the sliding window data segments; the time domain features include standard deviation, absolute mean, interquartile range, peak-to-peak value, skewness, kurtosis, Wilson amplitude, zero-crossing coefficient and regression coefficient signal energy; the frequency domain features are parameters obtained after Fourier transform of the surface electromyogram signal, including average power frequency, median frequency, quartile frequency, top three values of spectral density and power spectrum entropy; the time-frequency domain features are parameters obtained after wavelet transform of the surface electromyogram signal, i.e. the energy of each layer wavelet coefficient; then, the features that significantly affect the clinical scale score of any motion performance are selected from the time domain features, frequency domain features and time-frequency domain features, and a machine learning method is used to select a feature set that simultaneously satisfies a small number of features and good classification results from the selected features;

[0080] S223, the features of each sliding window data segment are input into the trained human motion recognition model, and the motion type represented by each sliding window data segment is output; the human motion recognition model is learned and recognized by using a random forest machine learning algorithm, and the number of trees set in the random forest machine learning algorithm is 8;

[0081] S224, the sliding window data segments are spliced in time sequence, two adjacent sliding window data segments in which the motion type changes are extracted, and a transition data segment is formed; all the extracted transition data segments are reconstructed into a new signal data in time as a motion category transition data segment; the motion category transition data segment is input into the trained motion segmentation model, and the time of motion type transition is output; the time of motion type transition is taken as the start and end time of each electromyogram time sequence signal, and the preprocessed surface electromyogram signal is segmented according to the motion type, so that each electromyogram time sequence signal corresponds to one motion type;

[0082] The motion segmentation model for sitting-to-standing transition is as follows: first, a sliding window data segment before the transition data segment of the sitting-to-standing transition is extracted, and the mean value gyr of the bilateral thigh coronal axis angular velocity of the sliding window data segment is calculated SIT and the standard deviation σ SITthe mean value of the thigh vertical axis acceleration acc SIT ; extract the sliding window data segment after the sit-to-stand transition data segment and convert the time sequence of the converted data segment, and calculate the mean value of the thigh vertical axis acceleration acc STD ; and finally calculate the criterion for the start and end time of the sit-to-stand transition according to formula (6):

[0083]

[0084] extract the transition data segment of the sit-to-stand transition, wherein the time when the thigh coronal axis angular velocity first reaches is the start time of the sit-to-stand transition, and the time when the thigh vertical axis acceleration first reaches is the end time of the sit-to-stand transition;

[0085] The segmentation model of the gait cycle is that the start and end time of the gait cycle is defined as the foot strike moment, when the foot is in contact with the ground, it will be subjected to an instantaneous impact upward, which is transmitted to the lower leg and manifested as a rapid change in the lower leg vertical axis acceleration. The positioning of the start and end time of the gait cycle is realized through a peak detection algorithm;

[0086] The segmentation model of the stand-to-walk transition is that the mean value gyr std and the standard deviation σ STD of the bilateral lower leg coronal axis angular velocity in the standing phase are calculated first; the criterion threshold for the start time of the stand-to-walk transition is gyr std + 3 × σ STD ; the transition data segment of the stand-to-walk transition is extracted, and the time when the bilateral lower leg coronal axis angular velocity first exceeds the respective threshold is the time when the lower limb starts to walk; the earlier one of the two times is the start time of the stand-to-walk transition, and the lower limb corresponding to the time is the lower limb that takes the first step; finally, the time of the first foot contact event of the lower limb that takes the first step is determined by using the gait cycle segmentation method, which is the end time of the stand-to-walk transition;

[0087] S23, the muscle tone rating unit 24 performs muscle tone disorder assessment, and the specific processing process is:

[0088] S231, first extract the signal time, frequency characteristics and motion macro features from the segmented muscle electrical time sequence signal, then input the signal time, frequency characteristics and motion macro features into the trained motion performance evaluation model, and output the motion performance score, which includes the ankle joint activity function score, posture transition function score, standing stability function score and walking function score of the detected person. Each function score is from level one to level five, and the higher the rating result indicates the more serious the dysfunction;

[0089] For any kind of lower limb movement (including lower limb single joint test, standing, walking and sit-to-stand transfer), the training process of the movement performance evaluation model is: first, the signal time and frequency characteristics and movement macro characteristics are extracted from the surface electromyography signals of the detected person and the healthy person control group, and a sample data set is established, each sample is a data vector composed of all the above characteristics of the detected person, and the label of the sample data is the movement performance clinical scale score of each subject performing the lower limb movement (the label of the detected person in the healthy person control group is 0); then, the samples are randomly divided into five groups of data, and four groups of data are randomly selected to form a training set, and the samples are resampled on the training set to make the number of samples of each label tend to be the same; the resampled training set is classified and modeled, a one-versus-all support vector machine algorithm is used, the classification target is the label of each sample corresponding to the detected person, and the classification model obtained by training is the movement performance evaluation model of the lower limb movement, and the output of the movement performance evaluation model is the movement performance score of the detected person for the lower limb movement;

[0090] S232、First, the segmented electromyography time series signal is paired into two-channel electromyography data according to the agonist and antagonist muscles, and eight pairs of paired data are obtained; then, intermuscular coupling parameters are extracted from the eight pairs of paired data, including time domain coupling parameters and frequency domain coupling parameters, and the time domain coupling parameter refers to the electromyography time domain co-activation area CL of the agonist and antagonist muscles, and the calculation formula is as follows (1):

[0091]

[0092] In formula (1), env1 and env2 represent the surface electromyography envelopes of the agonist and antagonist muscles respectively, ∫min(env1, env2) represents the area of the overlapping part of the two surface electromyography envelopes, and ∫max(env1, env2) represents the area of the outer contour part of the two surface electromyography envelopes;

[0093] The calculation formula of the frequency domain coupling parameter Arr is as follows (2):

[0094] Arr=∫IMC XY |(IMC XY >σ IMC ) (2);

[0095] Formula (2) represents the sum of all IMC IMC greater than the significant coherence threshold σ XY , IMC XY is the ratio of the cross-spectral density of the agonist and antagonist muscles to the auto-spectral density, and σ IMC is the significant coherence threshold, and the calculation formula of σ IMC is as follows (3):

[0096]

[0097] In formula (3), N IMC is the number of sliding windows of each segment of electromyographic timing signal, and a = 0.05 is a significant level;

[0098] Then, a fusion index is extracted from the intermuscular coupling parameters, and the fusion index includes a standardized absolute mean value, a standardized maximum value, and a standardized minimum value. First, the intermuscular coupling parameters of the control group conversion data segment collected from healthy people are standardized, and the calculation formula is shown in formula (4) as follows:

[0099]

[0100] In formula (4), IC represents the intermuscular coupling parameters after standardization, IC is the intermuscular coupling parameters to be standardized, and are the mean and standard deviation of the intermuscular coupling parameters of the healthy person control group, respectively;

[0101] Then, on the basis of formula (4), the fusion index is extracted, and the fusion object of the index is the intermuscular coupling parameters extracted from the eight pairs of paired data. For the standardized intermuscular coupling parameters, the standardized absolute mean value is the absolute mean value of the standardized intermuscular coupling parameters extracted from all eight pairs of paired data, the standardized maximum value is the mean value of the three largest data in the standardized intermuscular coupling parameters extracted from all eight pairs of paired data, and the standardized minimum value is the mean value of the three smallest data in the standardized intermuscular coupling parameters extracted from all eight pairs of paired data;

[0102] Finally, the screened fusion index is input into the pre-trained muscle inhibition abnormality degree grading evaluation model, and the muscle inhibition abnormality degree score of the detected person is output. The muscle inhibition abnormality degree score is from one to five, and the higher the evaluation result is, the more serious the muscle abnormality degree is;

[0103] The training process of the muscle inhibition abnormality degree grading evaluation model is as follows: first, the screening fusion indicators are extracted from the surface electromyography signals of the detected person and the healthy person control group, a sample data set is established, each sample is a data vector composed of all the fusion indicators of the detected person, and the label of the sample data is the lower limb muscle tension disorder clinical scale score of each detected person (the label of the detected person in the healthy person control group is 0); then, the samples are randomly divided into five groups of data, and four groups of data are randomly selected to form a training set; the samples are resampled on the training set so that the number of samples of each label tends to be the same; the resampled training set is classified and modeled, a one-versus-all support vector machine algorithm is adopted, the classification target is the label of each sample corresponding to the detected person, and the classification model obtained by training is the trained muscle inhibition abnormality degree grading evaluation model.

[0104] S233, input the sports performance score and the muscle inhibition abnormality degree score into the trained muscle tension fusion evaluation model, output the muscle tension rating result of the detected person, and the higher the rating is, the more serious the muscle tension disorder is;

[0105] The muscle tension fusion evaluation model is a decision-level fusion model, and the training process thereof is as follows: first, the related data of the detected person and the healthy person control group are input into the sports performance evaluation model and the muscle inhibition abnormality degree grading evaluation model, the sports performance score and the muscle inhibition abnormality degree score of each kind of lower limb movement of each detected person are obtained, a sample data set is established, each sample is a data vector composed of all the scores of the detected person, and the label of the sample data is the lower limb muscle tension disorder clinical scale score of each detected person (the label of the detected person in the healthy person control group is 0); then, the samples are randomly divided into five groups of data, and four groups of data are randomly selected to form a training set; the samples are resampled on the training set so that the number of samples of each label tends to be the same; the resampled training set is classified and modeled, a one-versus-all support vector machine algorithm is adopted, the classification target is the label of each sample corresponding to the detected person, and the classification model obtained by training is the trained muscle tension fusion evaluation model;

[0106] S24, the abnormal muscle positioning unit 25 adopts an abnormal muscle positioning model to determine the abnormal movement type and locate the abnormal muscle, and the specific processing process is as follows:

[0107] S241, the contribution degree parameter of the fusion indicator to the output result of the muscle inhibition abnormality degree grading evaluation model is calculated by using a model explanation technology, the contribution degree parameter is the shapley value of the input model feature, and for any feature x i , the calculation formula of the shapley value is as follows:

[0108]

[0109] In formula (5), M S is the input feature set, #M S is the basis of M S , Su S is a feature subset of M i not containing feature x S , #Su S is the basis of Su S , v(Su S ) is the model output of Su S , v(Su S ∪{i}) is the model output after feature x i is added to Su S , and! represents factorial; the model to be explained is a muscle inhibition abnormality degree grading evaluation model, the input model features refer to fusion indexes related to the clinical scale score of muscle tone disorder selected by using a statistical method, and the feature set refers to a set of all input model features;

[0110] S242, absolute values of shapley values are calculated, and outliers in the absolute values are searched to locate a main coupling feature index: when there is an outlier, the index represented by the outlier is the main coupling feature index; if there is no outlier, the top three largest absolute values are selected as the main coupling feature index;

[0111] S243, the movement represented by the main coupling feature index is the abnormal movement type of the detected person, when the fusion index corresponding to the main coupling feature index is a standardized absolute mean value, according to the three times variance principle, the parameter with an absolute value greater than 3 among all intermuscular coupling parameters constituting the fusion index is an abnormal parameter, the muscle pair corresponding to the abnormal parameter is an abnormal muscle, that is, the position of the muscle pair with abnormal co-activation is known; when the fusion index corresponding to the main coupling feature index is a standardized maximum value, according to the three times variance principle, the parameter with a value greater than 3 among all intermuscular coupling parameters constituting the fusion index is an abnormal parameter, the muscle pair corresponding to the abnormal parameter is an abnormal muscle; when the fusion index corresponding to the main coupling feature index is a standardized minimum value, according to the three times variance principle, the parameter with a value less than -3 among all intermuscular coupling parameters constituting the fusion index is an abnormal parameter, and the corresponding muscle pair is an abnormal muscle;

[0112] S25, the movement performance score, the muscle inhibition abnormality degree grading result, the muscle tone grading result, the position of the muscle pair with abnormal co-activation, and the abnormal movement type data constitute a processing result, and the data storage unit 21 of the information processing module 2 stores the processing result;

[0113] (3), result display: the information processing module 2 synchronously sends the processing result to the human-computer interaction module 3 for display; the physician formulates and enters a diagnosis scheme according to the processing result, and manages the multiple evaluation results stored according to the time line.

[0114] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely divergences, modifications, replacements and variations of the embodiments, and the scope of the application is defined by the appended claims and their equivalents.

Claims

1. A muscle tension rating and abnormal muscle dynamic positioning system, characterized by: It includes an information collection module, an information processing module and a human-computer interaction module. The information collection module includes a personal information collection and entry unit and a surface electromyography signal acquisition unit. The information processing module is set up on a cloud server and includes a data storage unit, a signal preprocessing unit, a motion recognition and segmentation unit, a muscle tension rating unit and an abnormal muscle positioning unit. The human-computer interaction module includes a test guidance unit, an information management unit, a result display unit, a supervision and tracking unit and a diagnostic information entry unit. The personal information collection and entry unit sends the collected and entered personal information to the information management unit. The information management unit accesses, operates and manages the personal information through the touch screen. At the same time, the human-computer interaction module sends the personal information to the information processing module. The data storage unit stores the personal information. At the same time, the information management unit retrieves and manages the personal information stored in the data storage unit. After the surface electromyography signal acquisition unit is connected to the skin surface of the limb of the person being tested, the test guidance unit performs a guidance demonstration of the muscle tension test, and the person being tested completes the specified limb test task according to the requirements demonstrated by the test guidance unit. When the corresponding action of the limb test task is completed, the surface electromyography signal acquisition unit collects the surface electromyography signal of the person being tested and sends it to the information processing module. The signal preprocessing unit performs signal preprocessing, and the preprocessed surface electromyography signal is sent to the motion recognition and segmentation unit for motion recognition and segmentation. The motion recognition and segmentation unit first uses a human motion recognition model to identify the type of motion, and then uses a motion segmentation model to segment the time series windows corresponding to different motions, so that each type of motion corresponds to a segment of electromyography time series signal. The segmented segment of electromyography time series signal is synchronously input into the muscle tension rating unit and the abnormal muscle positioning unit; The processing process of the muscle tension rating unit is specifically as follows: S231. First, extract signal time and frequency features and motion macro features from the segmented electromyographic time series signal, where the signal time and frequency features are new features extracted from the time domain features, frequency domain features, and time-frequency domain features after dimensionality reduction, and the motion macro features are the time required to complete the sit-to-stand transition and the subject's gait time features. Then, input the signal time and frequency features and motion macro features into a trained sports performance evaluation model to output a sports performance score. The sports performance score includes the subject's ankle joint activity function score, posture transition function score, standing stability function score, and walking function score. Each function score is rated from level one to level five, with a higher rating indicating a more severe functional impairment. S232. First, the segmented EMG time series signal is paired into two-channel EMG data according to the agonist muscle and the antagonist muscle, obtaining eight pairs of paired data. Then, the intermuscular coupling parameters are extracted from the eight pairs of paired data, including the time domain coupling parameter and the frequency domain coupling parameter. The time domain coupling parameter refers to the time domain co-activation area CL of the EMG signals of the agonist muscle and the antagonist muscle, and the calculation formula is shown in the following formula (1): In formula (1), env1 and env2 represent the surface electromyographic signal envelopes of the agonist and antagonist muscles, respectively; ∫min(env1,env2) represents the area of ​​the overlapping part of the two surface electromyographic signal envelopes; ∫max(env1,env2) represents the area of ​​the outer contour of the two surface electromyographic signal envelopes; The calculation formula of the frequency domain coupling parameter Arr is shown in the following formula (2): Arr=∫IMC XY |(IMC XY >s IMC ) (2); Formula (2) represents the sum of all the significant coherence threshold σ IMC IMC XY , IMC XY is the ratio of the cross-spectral density to the autospectral density of the agonist and antagonist muscle signals, σ IMC is the significant coherence threshold, σ IMC The calculation formula is shown in the following formula (3): In formula (3), N IMC is the number of sliding windows for each segment of EMG time series signal, and α = 0.05 is the significance level; Then, fusion indices are extracted from the intermuscular coupling parameters. The fusion indices include the standardized absolute mean, the standardized maximum, and the standardized minimum. First, the intermuscular coupling parameters of the transformed data segments of the control group collected from healthy subjects are used for standardization. The calculation formula is shown in the following formula (4): In formula (4), represents the intermuscular coupling parameter after normalization, IC is the intermuscular coupling parameter to be normalized, and are the means and standard deviations of the intermuscular coupling parameters of the healthy control group; Then, based on formula (4), the fusion index is extracted. The fusion object of the index is the intermuscular coupling parameters extracted from the eight pairs of paired data. For the standardized intermuscular coupling parameters, the standardized absolute mean refers to the absolute mean of the standardized intermuscular coupling parameters extracted from all eight pairs of paired data, the standardized maximum refers to the mean of the three largest data among the standardized intermuscular coupling parameters extracted from all eight pairs of paired data, and the standardized minimum refers to the mean of the three smallest data among the standardized intermuscular coupling parameters extracted from all eight pairs of paired data. Finally, the selected fusion indicators are input into a pre-trained muscle inhibition abnormality grading assessment model to output the subject's muscle inhibition abnormality score. The muscle inhibition abnormality score ranges from level one to level five, with higher ratings indicating more severe muscle abnormality. S233. Inputting the motor performance score and the muscle inhibition abnormality score into the trained muscle tone fusion assessment model, and outputting a muscle tone rating result of the subject, where the muscle tone rating ranges from level one to level five, with a higher rating indicating a more severe degree of dystonia; The abnormal muscle positioning unit uses the abnormal muscle positioning model to calculate the fusion index screened out by the muscle inhibition abnormality grading assessment model to obtain the main coupling characteristic index. The movement represented by the main coupling characteristic index is the abnormal movement type. The abnormal parameter is determined based on the fusion index corresponding to the main coupling characteristic index. The muscle pair corresponding to the abnormal parameter is the abnormal muscle, thereby determining the position of the muscle pair with abnormal co-activation; The information processing module sends the sports performance score, muscle inhibition abnormality rating result, and muscle tension rating result obtained after processing by the muscle tension rating unit to the human-computer interaction module, and displays them on the touch screen through the result display unit; The information processing module sends the positions of the muscle pairs with abnormal co-activation and the abnormal movement types obtained by the abnormal muscle positioning unit to the human-computer interaction module, and displays them on the touch screen through the result display unit; At the same time, the sports performance score, muscle inhibition abnormality rating result and muscle tension rating result obtained after processing by the muscle tension rating unit, as well as the position of the muscle pair with abnormal co-activation and the abnormal movement type obtained after processing by the abnormal muscle positioning unit are all sent to the data storage unit to realize the storage of the detection result information; The diagnostic plan issued and entered by the diagnostic information entry unit is also sent to the data storage unit to achieve storage of the diagnostic information; The supervision and tracking unit manages and displays the multiple evaluation results stored in the data storage unit according to the time line through the touch display screen.

2. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 1, characterized in that: The surface electromyography signal acquisition unit is a surface electromyography sensor or a wearable device embedded with a surface electromyography sensor.

3. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 1, characterized in that: The surface electromyography signal acquisition unit is connected to the skin surface of the limbs of the subject, and the specific connection positions are the rectus femoris, semitendinosus, tibialis anterior, peroneus longus, medial gastrocnemius and soleus muscles of the subject; there are eight pairs of agonist and antagonist muscles, namely, tibialis anterior-peroneus longus, tibialis anterior-medial gastrocnemius, tibialis anterior-soleus and rectus femoris-semitendinosus of the bilateral legs.

4. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 3, characterized in that: The limb test tasks include single-joint tests and multi-joint compound tests; the single-joint test requires the subject to sit on a chair and perform three isometric contractions of dorsiflexion, plantar flexion, inversion and eversion of the ankle joints on both sides respectively; the multi-joint compound test requires the subject to perform three tests, each test starts with a sitting position. After receiving the start command, the subject stands up from the chair, stands still for a few seconds, then switches from the standing state to the walking state and starts walking straight. After walking a set distance, turn and return to the initial position, turn again and walk back and forth for multiple sets. After completing the walk, turn at the starting position and stand still for a few seconds before sitting back on the chair.

5. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 4, characterized in that: The signal preprocessing specifically includes performing band-pass filtering, removing the mean, high-pass filtering, filtering out power frequency interference, and filtering out white noise on the collected surface electromyography signal, thereby obtaining a preprocessed surface electromyography signal.

6. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 5, characterized in that: The processing process of the motion recognition and segmentation unit is specifically as follows: S221, dividing the pre-processed surface electromyography signal into sliding window data segments according to a set time span; S222, extracting time domain features, frequency domain features and time-frequency domain features from the sliding window data segment; The time domain features include standard deviation, absolute mean, interquartile range, peak-to-peak value, skewness, kurtosis, Wilson amplitude, zero-crossing coefficient and regression coefficient signal energy; the frequency domain features are parameters obtained after Fourier transform of surface electromyographic signals, including mean power frequency, median frequency, interquartile frequency, top three values ​​of spectral density and power spectrum entropy; the time-frequency domain features are parameters obtained after wavelet transform of surface electromyographic signals, i.e., the energy of wavelet coefficients of each layer; then, features that significantly affect the clinical scale score of any sports performance are selected from the time domain features, frequency domain features and time-frequency domain features, and then, using machine learning methods, a feature set that simultaneously satisfies the requirements of a small number of features and good classification results is selected from the selected features; S223, inputting the features of each sliding window data segment into a trained human motion recognition model, and outputting the motion type represented by each sliding window data segment, wherein the human motion recognition model uses a random forest machine learning algorithm for learning and recognition; S224. Splice the sliding window data segments in time sequence, extract two sliding window data segments with changed motion types that are adjacent in time sequence, form conversion data segments, reconstruct all the extracted conversion data segments into a new signal data segment in time as a motion category conversion data segment, input the motion category conversion data segment into the trained motion segmentation model, output the moment of motion type conversion, use the moment of motion type conversion as the start and end time of each segment of electromyographic timing signal, and segment the preprocessed surface electromyographic signal according to motion type, so that each segment of electromyographic timing signal corresponds to a motion type.

7. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 6, characterized in that: The training of the described sports performance evaluation model, muscle inhibition abnormality grading evaluation model, and muscle tone fusion evaluation model all begins by establishing a sample data set, wherein the label of the sample data is the sports performance clinical scale score or the muscle tone fusion evaluation model clinical scale score. Then, samples are randomly selected from the sample data set to form a training set; the samples in the training set are resampled so that the number of samples for each label converges; the classification model is trained on the resampled training set, wherein the classification target is the label of the subject corresponding to each sample. The classification model obtained by training is the trained sports performance evaluation model, muscle inhibition abnormality grading evaluation model, or muscle tone fusion evaluation model.

8. The muscle tension rating and abnormal muscle dynamic positioning system according to claim 7, characterized in that: The processing process of the abnormal muscle positioning unit is specifically as follows: S241. Use model interpretation technology to calculate the contribution parameter of the fusion index to the output of the muscle inhibition abnormality grading evaluation model. The contribution parameter is the Shapley value of the input model feature. For any feature x i , its Shapley value The calculation formula is shown in the following formula (5): In formula (5), M S is the full set of input features, #M S M S Deji, Su S does not contain feature x i M S The feature subset of #Su S Su S The base, v(Su S ) is Su S The model output, v(Su S ∪{i}) is the feature x i Join Su S The model output after ! represents factorial; the model to be explained is a grading assessment model for muscle inhibition abnormality, the input model features refer to the fusion indicators related to the dystonia clinical scale scores screened using statistical methods, and the feature set refers to the full set of all input model features; S242. Calculate the absolute values ​​of the Shapley values ​​and search for outliers among these absolute values ​​to locate the main coupling characteristic indicators: if an outlier exists, the indicator represented by the outlier is the main coupling characteristic indicator; if no outlier exists, select the indicators with the top three absolute values ​​as the main coupling characteristic indicators; S243. The movement represented by the main coupling characteristic indicator is an abnormal movement type of the subject. When the fusion indicator corresponding to the main coupling characteristic indicator is the standardized absolute average value, then according to the three-fold variance principle, the parameter with an absolute value greater than 3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the muscle pair corresponding to the abnormal parameter is an abnormal muscle, that is, the position of the muscle pair with abnormal co-activation is known; when the fusion indicator corresponding to the main coupling characteristic indicator is the standardized maximum value, then according to the three-fold variance principle, the parameter with a value greater than 3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the muscle pair corresponding to the abnormal parameter is an abnormal muscle; when the fusion indicator corresponding to the main coupling characteristic indicator is the standardized minimum value, then according to the three-fold variance principle, the parameter with a value less than -3 among all the intermuscle coupling parameters constituting the fusion indicator is an abnormal parameter, and the corresponding muscle pair is an abnormal muscle.

Images