HD-sEMG-based hand dyskinesia assessment method and system for cerebral apoplexy hemiplegia patient

Abstract

The invention belongs to the technical field of cerebral apoplexy rehabilitation medicine, and relates to evaluation of myoelectric medicine, in particular to an HD-sEMG-based evaluation method and evaluation system for hand dyskinesia of a cerebral apoplexy hemiplegia patient. According to the invention, the high-density surface myoelectricity data of the front arms of the healthy side and the affected side can be respectively collected when the stroke hemiplegia patient executes different hand actions, and the microscopic neuron characteristics are analyzed, so that the degree and progress of dyskinesia of the stroke hemiplegia patient can be accurately evaluated. Through statistical analysis of the number of motion units, the average discharge rate and the discharge mode, changes and rules of neurons in the rehabilitation process are revealed, the difference between the healthy side and the affected side is compared, and the motion function condition of the neurons is evaluated.

Description

Technical Field

[0001] The present invention belongs to the technical field of stroke rehabilitation medicine and relates to the evaluation of electromyography, and specifically to an evaluation method and an evaluation system for hand movement disorders in stroke hemiplegic patients based on HD-sEMG. Background Art

[0002] Stroke is one of the main causes of long-term disability in adults. Its sequelae often include hemiplegia and movement disorders, which seriously affect the patient's quality of life and daily functional recovery. With the development of medical technology, the methods for evaluating and treating movement disorders after stroke have gradually become more diverse and precise. At present, the diagnosis and condition assessment of hemiplegia after stroke mainly rely on clinical neurological examinations, imaging examinations (such as CT and MRI), scale assessments (such as Brunnstorm scale, Fugl-Meyer scale and Barthel index) and conventional surface electromyography (sEMG). Although these methods have certain value in evaluating the condition and rehabilitation progress of stroke patients, they also have some disadvantages.

[0003] Clinical examination and scale assessment rely on subjective judgment and lack objective quantitative data support; although imaging examinations can provide structural information, they cannot directly reflect functional recovery; conventional sEMG has low resolution and is difficult to capture subtle changes in neuromuscular activity. These limitations make it challenging to comprehensively and accurately evaluate the neurological function recovery and treatment effects of hemiplegic stroke patients.

[0004] High-density surface electromyography (HD-sEMG) technology has become one of the important tools for evaluating muscle and nerve function due to its non-invasive, high-resolution and real-time nature. HD-sEMG records the electrical activity on the muscle surface through a multi-channel electrode array, providing detailed muscle activation information, which helps to analyze and understand the complex functions and abnormal manifestations of the neuromuscular system. Using HD-sEMG technology, combined with advanced signal processing and motor unit decomposition algorithms, we can deeply explore the neural recovery mechanism of hemiplegic stroke patients, evaluate the effect of rehabilitation training, and provide a scientific basis for the formulation of personalized rehabilitation plans.

[0005] Therefore, the development of an evaluation system based on HD-sEMG technology is of great significance for improving the rehabilitation effect of patients with movement disorders after stroke.

[0006] At present, researchers have proposed a variety of methods to evaluate the motor function of hemiplegic stroke patients. The existing solutions that are similar to the solution proposed in the patent mainly include:

[0007] 1. The recovery of motor function was evaluated by low-frequency repetitive transcranial magnetic stimulation and intensive occupational therapy, and changes in motor cortical activity were studied based on magnetic resonance imaging (MRI) data. Thirty patients with upper limb paralysis after stroke were evaluated. They were hospitalized for 12 sessions of low-frequency repetitive transcranial magnetic stimulation (targeting the non-lesioned hemisphere) and daily occupational therapy for a total of 15 days. Hemispheric dominance was assessed using the laterality index using functional magnetic resonance imaging. Seed-based functional connectivity analysis was used to assess the functional connectivity between the affected side precentral gyrus and other areas. The results showed that low-frequency repetitive transcranial magnetic stimulation and occupational therapy intervention improved motor function, and this improvement was associated with an increase in functional connectivity between the affected side precentral gyrus and the healthy side precentral gyrus [1].

[0008] 2. Ultrasound measurement was used to evaluate the structural changes of the tibialis anterior muscle in stroke patients at rest and during maximum isometric contraction, and to explore the correlation between muscle structural parameters and ankle joint position and force generation function. The study recruited 7 stroke patients and 7 age- and sex-matched healthy subjects. Ultrasound was used to measure the muscle structural parameters of the tibialis anterior muscle at different ankle joint positions, including feather angle, thickness, and bundle length of the affected and healthy sides. At the same time, a handheld dynamic dynamometer was used to record muscle strength and ankle joint angle. The results showed that the feather angle and thickness of the tibialis anterior muscle on the affected side of stroke patients were lower than those on the healthy side. In addition, compared with the healthy side and healthy subjects, the three structural parameters and muscle strength of the affected side during maximum isometric contraction changed less. The conclusion pointed out that the muscle structure of the tibialis anterior muscle changed significantly after stroke, and these changes were related to the ankle joint angle and muscle strength, which may directly lead to impaired motor function in stroke survivors [2].

[0009] Disadvantages of the prior art:

[0010] 1. Low-frequency repetitive transcranial magnetic stimulation program.

[0011] (1) Insufficient individualization and predictability of treatment effects: There is currently a lack of effective biomarkers or predictive indicators, and it is impossible to accurately predict which patients will benefit from low-frequency repetitive transcranial magnetic stimulation therapy, so it is impossible to provide an individualized treatment plan for each patient. (2) Local effects and conduction limitations: The therapeutic effect of low-frequency repetitive transcranial magnetic stimulation is affected by factors such as head tissue structure, skull morphology, and depth of the cerebral cortex. These factors may lead to differences in treatment effects between different individuals and localized limitations. (3) Limitations of measurement methods: The use of magnetic resonance imaging for evaluation is expensive and requires equipment and professional technical support. It is not suitable for widespread use in daily rehabilitation treatment, especially in resource-scarce areas and medical environments.

[0012] 2. Ultrasonic measurement and evaluation scheme.

[0013] (1) Limitations: Ultrasonic measurements provide static muscle structure information, but are limited in assessing the dynamic changes of neuronal activity and motor units. (2) Limited motor quantification capabilities: Unable to provide accurate quantitative assessment of motor function, such as the specific characteristics of neuronal activity and dynamic changes during rehabilitation.

[0014] References[1]Ueda,RyoYamada,NaokiAbo,MasahiroRuwan,PradeepaWanniarachchiSenoo,Atsushi.MRI evaluation of motor function recovery by rTMSand intensive occupational therapy and changes in the activity of motorcortex[J].International Journal of Neuroscience,2020,130(1a6).

[0015] References [2] Peng L, Yanjun W, Yurong M. The muscle architectural parameters and strength changes of tibialis anterior in stroke survivors as measured by ultrasound [J]. Chinese Journal of Rehabilitation Medicine, 2012. DOI: 10.3969 / j.issn.1001-1242.2012.08.004. Summary of the invention

[0016] In view of the defects and limitations of the existing technical solutions, the present invention aims to propose a new evaluation method and evaluation system, which can more accurately and in real time evaluate the degree and progression of hand movement disorders in patients with hemiplegia due to stroke. Specifically, it includes:

[0017] 1. High-density surface electromyography signal acquisition and preprocessing module: By collecting 256 channels of surface electromyography data, it covers a wide range of muscle areas and provides more comprehensive neuronal activity information.

[0018] 2. Motor unit identification and characteristic analysis module: The fastICA algorithm is used to decompose high-density surface electromyography signals, which can accurately identify the activity patterns and changes of individual motor units.

[0019] 3. Dynamic assessment and feedback: Real-time data collection and analysis are realized to provide doctors with immediate auxiliary assessment and treatment feedback to help adjust rehabilitation plans and treatment programs.

[0020] Through these innovations, the present invention aims to address the limitations of existing technical solutions in dynamic assessment, precise quantification and real-time monitoring, and provide more effective support and guidance for the rehabilitation treatment of hemiplegic stroke patients.

[0021] The present invention discloses a method for evaluating hand movement disorders in hemiplegic stroke patients based on HD-sEMG, comprising the following steps:

[0022] Step 1: Obtain HD-sEMG signal data;

[0023] Step 2: Processing the HD-sEMG signal data obtained in step 1;

[0024] Step 3: Identify and decompose the motion units of the data processed in step 2;

[0025] Step 4: Perform statistical analysis on the motor unit data decomposed in step 3;

[0026] Step 5: Evaluate the statistical analysis results in step 4.

[0027] Preferably, the step 1 specifically includes:

[0028] Step 11: The electrode arrays are placed on the flexor muscle groups and the extensor muscle groups of the forearm respectively;

[0029] Step 12: Complete the gesture according to the guidance, and the electrode array collects the electromyographic signal data and uploads the data to the host computer;

[0030] Step 13: Repeat step 13 until all gestures are completed and the electromyographic signal data collection is completed.

[0031] Further preferably, in step 12, the long axis of the electrode array is parallel to the long axis of the subject's forearm.

[0032] Further preferably, in step 13, the gesture actions include wrist valgus, wrist eversion, internal rotation, external rotation, fist clenching, thumb and index finger two-finger pinching, and thumb, index finger and middle finger three-finger pinching.

[0033] Preferably, the step 2 specifically includes:

[0034] Step 21: Use a moving average filter to smooth the signal and remove short-term high-frequency noise;

[0035] Step 22: Use a notch filter with a dynamic adjustment function to perform directional filtering on power frequency interference and its harmonic components;

[0036] Step 23: Perform event-based segmentation and labeling on the collected data.

[0037] Preferably, the step 3 specifically includes:

[0038] Step 31: Splicing data for the same gesture action, intercepting a short section of resting state data before and after each sample as a transition interval for splicing;

[0039] Step 32: After completing data splicing in step 31, use the FastICA algorithm to identify motion units;

[0040] Step 33: Decompose the identified motion units.

[0041] Further preferably, the step 32 specifically includes:

[0042] Step 321: Use the delayed copy of the electromyographic signal to expand the electromyographic channel so that subsequent analysis can more accurately capture subtle changes and features of the signal.

[0043] Step 322: Subtract the mean value of each channel from the signal to ensure that the mean value of the signal is zero.

[0044] Step 323: performing linear transformation on the decentralized signal;

[0045] Step 324: Use the FastICA algorithm to perform reverse calculation on the signal data processed in step 323 to separate independent motion units.

[0046] Further preferably, the FastICA algorithm in step 32 includes a dynamic adjustment mechanism, and the dynamic adjustment mechanism specifically includes: dynamic adjustment of non-Gaussian parameters, adaptive learning rate adjustment, dynamic update of weight vectors, stability detection of signal separation, and performance feedback mechanism;

[0047] The dynamically adjusted non-Gaussian parameters include coefficients of kurtosis and negative entropy; according to the signal non-Gaussian value calculated in real time, the threshold range of kurtosis is adjusted between 0.5 and 1.5 to cope with the signal characteristics under different data conditions; for signals with a kurtosis threshold lower than 0.5, the weight of kurtosis is increased to increase the sensitivity to non-Gaussianity; for signals with a kurtosis threshold higher than 1.5, the weight of kurtosis is reduced to prevent over-emphasis on non-Gaussianity; the kurtosis parameter is adjusted in real time to ensure that the signal obtains the best separation effect under different data conditions;

[0048] The adjustment range of negative entropy is set to 0.1 to 0.5 to ensure that the algorithm can effectively separate signals under different noise levels; for signals with a negative entropy threshold lower than 0.1, the negative entropy coefficient is increased to improve the algorithm's ability to capture non-Gaussianity; for signals with a negative entropy threshold higher than 0.5, the negative entropy coefficient is reduced to avoid over-separation; the negative entropy parameters are dynamically adjusted to adapt to different noise environments and signal characteristics;

[0049] The adaptive learning rate adjustment is to dynamically adjust the learning rate according to the error feedback of each iteration, with the initial value set to 0.01 and the adjustment range from 0.001 to 0.1; calculate the error after each iteration; if the error decreases, gradually increase the learning rate, set the growth step to 0.005, until it reaches the upper limit of 0.1; if the error fluctuates or increases, reduce the learning rate, reduce the step to 0.001, until it reaches the lower limit of 0.001; adjust the learning rate according to the error feedback to improve the stability and convergence speed of the algorithm;

[0050] The dynamic update of the weight vector uses an adaptive weight adjustment strategy, in which the weight update pace is determined by the independence and non-Gaussianity of the current signal; in each iteration, the gradient of the weight vector is calculated; according to the independence and non-Gaussianity indicators of the signal, the weight update pace is dynamically adjusted to ensure that the weight vector optimization achieves the best effect;

[0051] The stability detection of the signal separation includes calculating the correlation coefficient between independent components. If the correlation coefficient is close to 0, it means that the independence is good; if the correlation coefficient is high, it is necessary to further adjust the algorithm parameters; according to the value of the correlation coefficient, the kurtosis and negative entropy parameters and the learning rate are automatically adjusted to improve the stability of the signal separation; and the algorithm is re-run until the separation effect is stable;

[0052] The performance feedback mechanism monitors the mean square error and signal independence index of the separation effect in real time. If the values ​​of the mean square error and independence index deviate from the expected range, the algorithm parameters, including kurtosis, negative entropy coefficient and learning rate, are dynamically adjusted; based on the performance feedback, the algorithm parameters are optimized to ensure that the signal separation quality reaches the optimal level.

[0053] Further preferably, in step 33, the identified motion units are decomposed, specifically, peak detection is performed on the decomposed independent components to identify the activities of each motion unit; k-means clustering analysis is performed on the detected peaks, and a class of peaks with larger values ​​are identified as discharge pulses of the motion units.

[0054] Preferably, the statistical analysis in step 4 is a statistical evaluation of the number of motor units, average discharge rate and discharge pattern of motor units on the healthy and affected sides.

[0055] The present invention also provides an evaluation system for a method for evaluating hand movement disorders in hemiplegic stroke patients based on HD-sEMG, specifically:

[0056] An HD-sEMG data acquisition system, wherein the HD-sEMG data acquisition system comprises an electromyography collector, wherein the sampling rate of the electromyography collector is 2048 Hz; the HD-sEMG data acquisition system is used to acquire HD-sEMG signal data;

[0057] A data processing module, which receives the HD-sEMG signal data sent by the HD-sEMG data acquisition system, processes the data, and transmits the result to the motor unit identification and decomposition module;

[0058] A motion unit identification and decomposition module, which receives the data processed by the data processing module to perform motion unit identification and decomposition, and transmits the result to the data statistical analysis module;

[0059] A data statistical analysis module, which receives the data processed by the motion unit identification and decomposition module for statistical analysis and transmits the result to the evaluation module;

[0060] An evaluation module receives and evaluates the data processed by the data statistical analysis module.

[0061] Compared with the prior art, this application has the following beneficial effects:

[0062] 1. High-density surface electromyography signal acquisition and preprocessing module:

[0063] 1. Comprehensive coverage of muscle areas: 256-channel surface electromyography data acquisition covers a wide range of muscle areas, providing more comprehensive neuronal activity information than traditional methods.

[0064] 2. High-resolution data: High-density acquisition ensures higher signal resolution, which can capture subtle changes in muscle activity and improve data accuracy.

[0065] 2. Motion unit identification and characteristic analysis module:

[0066] 1. Accurately identify motor units: By decomposing high-density surface electromyographic signals through the fastICA algorithm, the activity patterns and changes of individual motor units can be accurately identified.

[0067] 2. Refined characteristic analysis: Analysis of motor unit characteristics enables a deeper understanding of the functional status and abnormalities of the neuromuscular system and provides a more accurate quantitative assessment.

[0068] 3. Dynamic evaluation and feedback:

[0069] 1. Real-time data collection and analysis: The system can realize real-time data collection and analysis, providing doctors with immediate auxiliary evaluation and treatment feedback.

[0070] 2. Personalized treatment adjustment: Based on real-time feedback, doctors can quickly adjust rehabilitation plans and treatment options, provide personalized treatment, and improve rehabilitation outcomes.

[0071] Through these innovations, the invention has significant advantages in the following aspects:

[0072] 1. Dynamic evaluation: Compared with existing technologies, it can perform dynamic evaluation according to the motor recovery status of different Brunnstrom grades, providing a scientific basis for the adjustment and optimization of treatment plans.

[0073] 2. Accurate quantification: High-density acquisition and advanced signal processing algorithms make the quantification of motor function more accurate.

[0074] 3. Real-time monitoring: Real-time analysis and feedback mechanism ensure timely adjustment of rehabilitation treatment and improve treatment effect and efficiency.

[0075] Overall, combined with the Brunnstrom grading system, this invention significantly improves the accuracy and real-time assessment of the degree and progression of motor disorders in hemiplegic stroke patients, providing more effective support and guidance for personalized rehabilitation treatment.

[0076] Rehabilitation medicine (stroke rehabilitation), clinical diagnosis (diagnosis of neuromuscular diseases, assessment of muscle dysfunction). BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 High-density electromyography acquisition system for stroke patients.

[0078] Figure 2 Experiment guide interface.

[0079] Figure 3 Gesture task in stroke patients.

[0080] Figure 4 Comparison of EMG signals before and after whitening.

[0081] Figure 5 Comparison of the number of motor units between the healthy and affected hands in stroke patients.

[0082] Figure 6 Comparison of the number of motor units in the affected hand of stroke patients with different Brunnstorm grades.

[0083] Figure 7 Comparison of mean discharge rates between the healthy and affected hands in stroke patients.

[0084] Figure 8 Comparison of the average discharge rate of the affected hand in stroke patients with different Brunnstorm grades.

[0085] Fig. 9 Comparison of mean discharge patterns between the healthy and affected hands in stroke patients.

[0086] Fig.10 Comparison of discharge patterns of the affected hand in stroke patients with different Brunnstorm grades. DETAILED DESCRIPTION

[0087] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0088] Embodiment 1

[0089] In this embodiment, a total of 12 hemiplegic stroke patients (11 males and 1 female; age range 27–75 years old) were recruited. Each patient signed an informed consent form before the experiment. Before data collection, clinicians will evaluate the patient's ability to perform hand movements based on the Brunnstrom scale to quantify the patient's degree of motor function recovery. The higher the Brunnstrom rating, the higher the patient's hand flexibility and the stronger the autonomous control ability. The basic information of the patient is shown in Table 1. It can be seen that the Brunnstrom rating is distributed in levels 4, 5, and 6, indicating that the patient is in a stage of moderate to high functional recovery. These stages usually indicate that the patient has begun to regain independent movement ability and has the ability to complete more complex movements, thus providing rich movement data for the study. In patients at these three rating levels, the recovery of muscles and nervous systems is more complicated. Selecting these patients can help researchers gain an in-depth understanding of the muscle activity and movement coordination in different movement patterns in the later stages of rehabilitation, thereby providing a basis for evaluating the progress of functional recovery in patients. At the same time, patients at Brunnstrom levels 4 to 6 still have great potential for rehabilitation. Selecting these patients can help researchers observe changes in muscle activity patterns during continuous rehabilitation training, thereby providing data support for optimizing rehabilitation strategies.

[0090] Each patient needs to complete the experiment under the guidance of the experimental guidance interface and prompt sound. Figure 3The seven gesture tasks shown include wrist varus, wrist valgus, internal rotation, external rotation, fist clenching, thumb-index two-finger pinching, and thumb-index-middle three-finger pinching. These seven gesture tasks cover the basic functions of the forearm and hand, and can comprehensively evaluate the recovery of hand motor ability in stroke patients. Wrist varus and wrist valgus movements evaluate the ability of forearm rotation, internal rotation and external rotation movements involve shoulder joint movement, fist clenching movements evaluate the hand's grasping ability, and two-finger pinching and three-finger pinching movements focus on fine hand operations and finger coordination. At the same time, all gesture tasks are highly related to common hand activities in daily life. For example, wrist varus and wrist valgus movements are very important in daily tasks such as twisting towels or opening door locks, fist clenching movements are related to grasping objects, and pinching movements are crucial when picking up small objects, using cutlery, or writing. These movements activate different muscle groups in the forearm and hand, and can generate representative HD-sEMG signals, which can reflect the recovery of different muscle groups in the forearm and hand, and provide data support for further muscle function analysis and pattern recognition. Through the analysis of the electromyographic signals of these movements, the changes in muscle activity patterns of stroke patients under different movements can be captured.

[0091] Before data collection, patients can practice switching between different gestures until they can complete each gesture task as accurately as possible. The data collection system is installed for the subjects according to the technical solution of this application. During the data collection process, the subjects will sit in a comfortable chair and perform the required experimental tasks according to the prompts displayed on the computer screen in front of them. During the data collection process, each gesture is performed in sequence, and 2 groups are performed each. Each group includes 3 dynamic gesture tasks and 1 gesture holding task. During the dynamic gesture task, it is necessary to achieve a complete process from the resting state to the saturated gesture state and then return to the resting state; the gesture holding requires maintaining the force state of the gesture for at least 3 seconds. The patient has a 2s rest time between two trials in the group and a 5s rest time between two groups of trials to avoid muscle fatigue. If the subject misses or performs a trial incorrectly, they will notify the main tester and click the redo button on the guidance interface to redo the skipped or wrong trial.

[0092] Table 1 Information of stroke patients

[0093]

[0094] The evaluation method proposed in this embodiment includes the following steps:

[0095] Step 1: Collect HD-sEMG signal data through the electromyography acquisition system;

[0096] In this embodiment, the electromyography acquisition system uses the Quattrocento electromyography acquisition system produced by OTBioelectronica, Torino, Italy, and the sampling rate of the electromyography acquisition system is 2048Hz. The electromyography acquisition system uses a wired method to transmit signals, avoiding problems such as signal loss, signal interference or signal delay that may occur in wireless transmission. In order to minimize external interference and crosstalk, a dual power supply design of built-in battery and external mobile power supply is used during data acquisition, and the surrounding AC power supply is disconnected to reduce the risk of external power supply interference.

[0097] The bio-acquisition system was used to collect HD-sEMG signals from the forearms of patients with hemiplegia due to stroke. Before acquisition, the clinician evaluated the skin and muscle conditions of the patients to ensure that the skin and the electrode patch could fit firmly and that a stable HD-sEMG signal could be detected. For each stroke patient, the electromyographic signal data of the healthy side and the affected side were collected on the same day to ensure the consistency and comparability of the data. Before data acquisition, the forearm skin on the acquisition side was cleaned with scrub, rinsed with alcohol and wiped dry. Four 8×8 high-density electrode arrays (ELSCH064NM3 Pin Out, OTBioelettronica, Turin, Italy) were used to collect HD-sEMG signals from the forearms of patients. Each electrode array consisted of 64 elliptical electrodes, each with a long axis of 5 mm and a short axis of 2.8 mm. The center-to-center distance of adjacent elliptical electrodes was 10 mm. The four electrode arrays were placed on the flexor and extensor muscle groups, respectively, with the radial and ulnar sides of the arm, the ulnar joint of the humerus and the ulnar head as the boundaries of the electrode array placement area. On both sides of the forearm, two 8×8 electrode arrays on each side constitute a 16×8 electrode array. The long axes of the two 16×8 electrode arrays of the flexor and extensor muscles are parallel to the long axis of the subject's forearm. Before sampling, soak the reference electrode armband and wristband in clean water, wring out the water droplets, wrap the wristband around the ulnar head as the right leg driving electrode, and wrap the armband around the olecranon as the reference electrode.

[0098] In order to facilitate the subjects to successfully complete the experimental tasks, this embodiment has built a data collection and experimental guidance linkage collection system, such as Figure 1As shown. The acquisition system includes the Quattrocento electromyography acquisition system and a host computer, which transmit data via a network cable. The host computer has been installed with data acquisition software, which can display the currently collected electromyography signal data in real time during the data acquisition process, so that the experimenter can debug the equipment. The software has a built-in data recording function. When the recording is completed, the data will be stored in the storage device of the host computer. The experimental guidance part uses self-developed special acquisition software to facilitate data alignment during subsequent data processing and to facilitate the subjects to understand the progress of the experiment in real time. The experimental guidance interface is shown as follows Figure 2 As shown, pictures and sounds are used to guide the subjects to complete the current gesture task.

[0099] During the period when the subject needs to complete the gesture task, the guidance interface will display the corresponding picture and emit continuous beeps to remind the subject of the task progress. In addition, the user interface displays the progress of the experiment in real time in the upper right corner, including the current action sequence number, the current group number, and the next gesture prompt, so as to remind the patient of the next stage of the task in advance. Considering that stroke patients have weak motor ability and often have poor mental state, the guidance interface also sets pause, reset and error mark buttons. Clicking the mark button will pause in the next rest stage and pop up a window to ask whether it needs to be redone. After clicking the pause button, the subject can take a temporary rest for a while. After clicking the reset button, the interface will mark an error label after completing the current gesture and go back to the previous rest stage of the previous gesture. These functions can realize segmented data storage to avoid the situation where the electromyography acquisition device is low in power during the experiment. The guidance interface is connected to the level generator through the Bluetooth module. During each task, the level generator is controlled to output a high-level signal and input it into the auxiliary channel of the electromyography acquisition device for subsequent data cutting.

[0100] Step 2: Processing the HD-sEMG signal data collected in step 1;

[0101] In step 2, first step 21, use a moving average filter to smooth the signal and remove short-term high-frequency noise. This method reduces instantaneous interference fluctuations, smoothes data and improves the stability of the signal by calculating the average value within a certain range around each sample point in the signal. In the specific operation of this embodiment, a moving average window with a length of 50 samples is selected, and the filter calculates the mean of 25 samples before and after each sample point to smooth the data at this point. This window length is tuned to balance the removal of high-frequency noise and the retention of the main features of the signal. A longer window length helps to smooth the signal more effectively, but may cause blurring of the signal features. Therefore, in this embodiment, the window length of the preferred sample is 50, so as to retain the main waveform and features of the signal as much as possible while removing high-frequency noise.

[0102] In terms of further processing of power frequency interference, a notch filter with dynamic adjustment function is used to perform directional filtering on power frequency interference and its harmonic components. The filter first continuously monitors the spectral characteristics of the input signal through real-time spectrum analysis function. This process uses the fast Fourier transform (FFT) algorithm to dynamically detect the main interference frequency, such as 50Hz power frequency interference and its harmonic components (such as 100Hz, 150Hz, etc.). Based on the real-time spectrum data, the filter automatically adjusts the center frequency of the notch to accurately align with the detected interference frequency. In this way, the filter can effectively capture the power frequency and its harmonic components, and deeply suppress them to reduce their impact on the signal.

[0103] In order to cope with changes in interference frequency or expansion of bandwidth, the filter has an adaptive bandwidth adjustment function. When it is detected that the frequency component of the power frequency interference is strong or the bandwidth is wide, the filter can automatically adjust the bandwidth of its notch filter to ensure sufficient suppression of the interference signal. At the same time, when the interference signal is weak, the filter will narrow the bandwidth to reduce the impact on the useful signal. The filter is also equipped with a feedback mechanism to monitor the effect of the filtered signal in real time. The filter optimizes and adjusts the filtering parameters (such as notch depth, bandwidth, etc.) according to the feedback signal to ensure that the filtering effect is optimal. This mechanism can automatically adjust the parameters according to changes in the signal environment to maintain the stability and accuracy of signal processing.

[0104] In order to ensure the accuracy and reliability of the data, the collected data were cut and marked based on events. Since stroke patients may not be able to respond to the task prompts immediately when performing tasks, the data cut based on the marking level often contains a part of the resting state data. To overcome this challenge, a manual calibration method was used to check and adjust each data segment one by one to ensure that the quality of the final data set meets the analysis requirements.

[0105] Step 3: Motion unit identification and characteristic analysis

[0106] Considering that the muscle force pattern of the same subject is basically the same when performing the same gesture task, and the same gesture has multiple repeated samples. Therefore, we first splice the data of the same gesture (6 dynamic gestures and 2 hold gestures) before identifying the motor unit. During the execution of the gesture task, the muscle activity pattern is usually repeated and consistent. Splicing data can enhance the stability and consistency of the signal, thereby improving the accuracy of motor unit decomposition and the reliability of the results. This is very important for identifying and analyzing motor units, especially when the muscle activity pattern is relatively consistent. At the same time, splicing multiple samples can help reduce signal noise caused by individual differences, operation errors and other factors, and provide cleaner and more consistent data for analysis. However, directly splicing two dynamic data segments may cause the signal to appear abrupt or discontinuous at the splicing point. Therefore, in order to avoid the data mutation at the data splicing point affecting the motor unit decomposition, when splicing the data samples, a small section of resting state data (100 sampling points) is intercepted before and after each sample as a transition interval. Resting state data is usually the background noise and resting potential of the signal, which can act as a transition zone in the smooth transition of the signal. They can help maintain the continuity of the signal and avoid signal incoherence or outliers caused by splicing, thereby improving the accuracy of motion unit decomposition and identification. Assign weights to the data points in the transition interval, with the weight of the first segment gradually decreasing from 1 to 0, and the weight of the second segment gradually increasing from 0 to 1. Perform weighted averaging on the signal in the transition interval to obtain a smooth transition splicing signal. After data splicing, use the FastICA algorithm to identify motion units. The specific steps are as follows:

[0107] Extension. This step usually involves extending the EMG channel using a time-delayed copy of the EMG signal so that subsequent analysis can more accurately capture subtle changes and features in the signal.

[0108] Decenter. Subtract the mean of each channel from its signal to ensure that the mean of the signal is zero. This helps to remove the DC component in the signal, making subsequent analysis more accurate and stable.

[0109] Whitening. The decentralized signal is linearly transformed to make the covariance matrix of the signal a unit matrix, that is, the correlation between the various signal components is zero. This step further reduces the correlation between signals by projecting the signal into the feature space, which helps to extract more independent and accurate motion units. Figure 4The time domain signals and cross-correlation functions of the electromyographic signals of the two channels before and after whitening are shown. The left side shows the electromyographic time domain signals and their cross-correlation functions before whitening. Before whitening, the electromyographic time domain signals of channel 1 and channel 2 have a high degree of overlap, and the cross-correlation function has a peak near τ = 0, reaching the order of 106, which indicates that the signals of the two channels are highly correlated at this moment; the right side shows the electromyographic time domain signals and their cross-correlation functions after whitening. After whitening, the overlap of the electromyographic time domain signals of channel 1 and channel 2 is significantly reduced, and the value of the cross-correlation function is basically maintained in the range of [-0.5, 0.5], which indicates that the channels are independent of each other.

[0110] FastICA algorithm reversely solves motion units based on dynamic adjustment mechanism. After whitening, the preprocessed signal is reversely solved using FastICA algorithm to separate independent motion units. In FastICA algorithm, we introduced a dynamic adjustment mechanism to optimize the separation effect of motion units by adjusting the algorithm parameters in real time. This mechanism mainly includes the following aspects:

[0111] 1) Dynamically adjust non-Gaussianity parameters: In FastICA, non-Gaussianity is a key factor in evaluating signal independence. We dynamically adjust the weight parameters of non-Gaussianity measurement, including the coefficients of kurtosis and negentropy. According to the signal non-Gaussianity value calculated in real time, we adjust the threshold range of kurtosis between 0.5 and 1.5 to cope with the signal characteristics under different data conditions. Specifically, for signals with kurtosis values ​​below 0.5, the weight of kurtosis is increased to increase the sensitivity to non-Gaussianity. For signals with kurtosis values ​​above 1.5, the weight of kurtosis is reduced to prevent over-emphasis on non-Gaussianity. The kurtosis parameters are adjusted in real time to ensure that the signals are best separated under different data conditions. The adjustment range of negentropy is set from 0.1 to 0.5 to ensure that the algorithm can effectively separate the signals under different noise levels. For signals with negentropy values ​​below 0.1, the coefficient of negentropy is increased to improve the algorithm's ability to capture non-Gaussianity. For signals with negative entropy values ​​higher than 0.5, the negentropy coefficient is reduced to avoid over-separation. The negentropy parameter is dynamically adjusted to adapt to different noise environments and signal characteristics.

[0112] 2) Adaptive learning rate adjustment: During the iterative optimization process, the learning rate determines the size of each optimization step. We use an adaptive learning rate strategy to dynamically adjust the learning rate based on the error feedback of each iteration. The initial value is set to 0.01 and the adjustment range is 0.001 to 0.1. Calculate the error (such as mean square error, MSE) after each iteration. If the error decreases, gradually increase the learning rate and set the growth step to 0.005 until it reaches the upper limit of 0.1. If the error fluctuates or increases, reduce the learning rate and reduce the step to 0.001 until it reaches the lower limit of 0.001. Adjust the learning rate based on error feedback to improve the stability and convergence speed of the algorithm.

[0113] 3) Dynamically update the weight vector: In each iteration, the weight vector is dynamically adjusted based on the characteristics of the current signal. We use an adaptive weight adjustment strategy, where the weight update pace is determined by the independence and non-Gaussianity of the current signal. In each iteration, the gradient of the weight vector is calculated. Based on the independence and non-Gaussianity indicators of the signal, the weight update pace is dynamically adjusted to ensure that the weight vector optimization achieves the best effect.

[0114] 4) Stability test of signal separation: After each iteration, evaluate the stability of signal separation, including calculating the correlation coefficient between independent components. If the correlation coefficient is close to 0, it means that the independence is good; if the correlation coefficient is high, the algorithm parameters need to be further adjusted. According to the value of the correlation coefficient, the kurtosis and negentropy parameters and the learning rate are automatically adjusted to improve the stability of signal separation. Rerun the algorithm until the separation effect is stable.

[0115] 5) Performance feedback mechanism: During the operation of the algorithm, the mean square error (MSE) and signal independence index of the separation effect are monitored in real time. If the values ​​of the MSE and independence index deviate from the expected range, the algorithm parameters, including kurtosis, negentropy coefficient and learning rate, are dynamically adjusted. Based on the performance feedback, the algorithm parameters are optimized to ensure that the signal separation quality reaches the best level.

[0116] Identification and screening of motor units. Peak detection is performed on the decomposed independent components (source signals) to identify the activities of each motor unit. The purpose of peak detection is to determine the activation moment of each motor unit on the time axis, so as to analyze its activity pattern. K-means clustering analysis is performed on the detected peaks, and a class of peaks with larger values ​​are identified as the discharge pulses of motor units, because the discharge pulses of motor units usually have larger signal amplitudes.

[0117] Step 4: Statistical analysis of decomposed motion units

[0118] After decomposing the motor units, the number of motor units, average discharge rate and discharge pattern of the healthy and affected sides can be counted respectively. By statistically analyzing and comparing the differences between the healthy and affected sides, we can further understand the changes and patterns of neurons during the rehabilitation process and evaluate the recovery of motor function. Specifically, it includes:

[0119] 1) Number of motor units: Traditional assessment of stroke severity usually relies on clinical manifestations and imaging examinations. Using the number of motor units for assessment provides a more fine-grained physiological indicator that can accurately reflect the degree of damage to the nervous system from the level of electromyographic signals. This method can supplement existing assessment methods and improve the accuracy and comprehensiveness of stroke assessment. Figure 5 The number of motor units on the healthy and affected sides of each patient under different gestures is compared. Green is the number of motor units decomposed on the healthy side, and red is the number of motor units decomposed on the affected side. The results of the Wilcoxon rank sum test showed that, except for internal and external rotation movements, other movements showed statistically significant differences, that is, the number of motor units that can be decomposed on the affected side of stroke patients is greater than that on the healthy side. Comparison of the number of motor units on the affected side of stroke patients with different Brunnstorm grades Figure 6 As shown in the figure, except for the inward rotation gesture, which has little difference between patients at level 6 and level 5, for most gestures, as the level decreases (the more severe the motor disorder), the number of corresponding decomposed motor units will gradually increase.

[0120] The affected side typically shows lower levels of muscle contraction than the unaffected side when performing the same gesture. The unaffected side has stronger neuromuscular control and is able to effectively activate and coordinate muscle contractions. In contrast, nerve damage on the affected side affects the recruitment of motor units, thereby impairing their ability to effectively activate and control muscle movement. As stroke severity increases, more motor units are recruited to compensate for the loss of motor control and strength. This suggests that a higher number of motor units on the paretic side, especially in cases with more severe movement disorders, indicates a compensatory response to maintain motor function.

[0121] After a stroke, although the number of neurons in the damaged area may decrease, neurons in other areas may undergo redistribution or proliferation to compensate for the function of the damaged area. This remodeling may be manifested as a relative increase in the number of neurons, especially during the recovery period after a stroke. At the same time, brain tissue near the damaged area may promote the proliferation or migration of neurons through collateral circulation mechanisms to enhance functional recovery. After a stroke, neural stem cells may be activated and migrate to the damaged area to form new neurons. This regenerative mechanism may lead to a certain increase in the number of motor neurons in certain areas. The increase in the number of motor neurons after a stroke reveals a new mechanism for neural repair and functional recovery.

[0122] 2) Average discharge rate: The average discharge rate provides an overall view of neuronal activity by calculating the average discharge frequency of neurons over a period of time. Unlike the instantaneous discharge rate (the discharge frequency at a specific point in time), the average discharge rate reveals the long-term activity trend of neurons through the average value within a time window. This method can better capture the stability and regularity of neural activity and make up for the defect that the instantaneous discharge rate may be affected by instantaneous fluctuations and noise. The electromyographic signals of stroke patients may fluctuate greatly, so calculating the average discharge rate can provide a more stable and reliable assessment by smoothing the signal and reducing interference, thereby more accurately reflecting the long-term changes in neural function and muscle control status.

[0123] Assume that the number of sampling points between two adjacent discharge moments is n, and the sampling rate is f s , the average discharge rate R can be calculated as follows:

[0124]

[0125] Figure 7 The average discharge rate of the affected side of each patient under different gestures was compared. The results of the Wilcoxon rank sum test showed that the instantaneous discharge rates of the five actions of inversion, flexion, external rotation, two-finger pinching, and three-finger pinching were statistically different, but all gestures had a consistent trend of change, that is, the average discharge rate of the affected side of stroke patients was lower than that of the healthy side. Figure 8 As shown in the figure, it can be seen that among the different levels, the most serious level 4 patients always maintain the lowest motor unit discharge rate. Except for wrist valgus and external rotation, the overall trend of change still follows the trend that the lower the level, the lower the average discharge rate.

[0126] Brain damage caused by stroke can reduce the average discharge rate of motor units when performing certain actions or mobilizing certain muscle groups. The reduction in the average discharge rate indicates that the brain's ability to control muscles is reduced, which in turn weakens the speed and completion of the patient's actions. In addition, a decrease in the average discharge rate of motor units will lead to an increase in the discharge interval of the same motor unit. Therefore, the possibility of overlapping pulses from different motor units at the same time is reduced. The reduction in overlap can reduce interference in surface electromyographic signals, which in turn can lead to more motor units being decomposed and identified, ultimately resulting in more motor units on the affected side of stroke patients than on the healthy side. The average discharge rate provides a quantitative indicator of neurological function in stroke patients. By calculating the difference in the average discharge rate between the healthy and affected sides, the impact of stroke on motor neuron function can be more accurately assessed.

[0127] 3) Motor unit discharge pattern: According to the gesture label, the instantaneous discharge rate of each motor unit is classified and summarized. That is, for each gesture, the instantaneous discharge rate values ​​of all relevant motor units are collected. For each subject, a histogram is drawn based on the discharge rate data of different gesture labels, which is the discharge pattern of the motor unit. Fig. 9 The average discharge pattern of the healthy and affected sides of each patient under different gestures is compared. Each row represents a stroke patient, and each column represents a gesture. The corresponding labels are marked at the top and right of the figure. In each sub-figure, blue and orange are used to represent the distribution of discharge patterns on the healthy and affected sides, respectively. For most gestures, the instantaneous discharge rate of the motor unit on the affected side is mainly concentrated near the lower frequency, showing a more concentrated peak profile. On the healthy side of the same subject, the motor unit discharge frequency is more diffusely distributed in the range of 0-35pps, the aforementioned peak profile becomes flat, and the highest peak also moves to the right. This difference is particularly obvious in the two-finger pinch and three-finger pinch gestures. Fig.10 Comparison of discharge patterns of the affected hand of stroke patients with different Brunnstorm levels. The motor function of patients with level 6 is closest to that of healthy people, and their discharge pattern is also more gentle and uniform. The disease severity of patients with level 4 is the most serious among the three levels, and their discharge pattern also shows a relatively high and sharp peak. As the level decreases, the peak frequency of the instantaneous discharge rate also gradually shows a downward trend.

[0128] When stroke patients perform gesture tasks, at the micro-neuron level, the instantaneous discharge rate of the affected-side motor units shows a significant decrease and concentration trend compared with healthy subjects and their own healthy side. This difference is particularly significant when performing complex finger movements such as three-finger pinching. This is consistent with the above results. The previous section on the average discharge rate of motor units shows that brain damage caused by stroke will reduce the average discharge rate of motor units, and the analysis results in this section further prove that the reduction in discharge rate caused by this motor delay at the micro level is holistic, and the instantaneous discharge rate of almost all motor units will decrease, and the overall instantaneous discharge rate tends to be more unified. The analysis in this section further confirms the microscopic impact of brain damage caused by stroke on motor control, that is, the discharge rate of motor units is reduced as a whole and tends to be unified. This change in discharge pattern is one of the microscopic characteristics of stroke patients and may be an important cause of motor delay in stroke patients.

[0129] In summary, analyzing the technical issues of motor unit number, average discharge rate, and motor unit discharge pattern from the perspective of different Brunnstrom ratings can provide a more detailed understanding:

[0130] Analyzing the number of motor units from the perspective of different Brunnstrom grades can determine the motor unit loss in stroke patients. This solves the problem of how to quantify the impact of stroke on the number of motor units, thereby understanding the degree of functional damage to motor neurons. Patients with Brunnstrom grade 4 have begun to recover some voluntary control, and analysis of the number of motor units can help evaluate the recovery of patients' motor units at this stage. The reduced number of motor units may reflect the limitations of motor function recovery. Patients with Brunnstrom grade 5 have better recovery of motor function, but changes in the number of motor units still provide important information about further recovery and motor ability. Analysis of the number of motor units can help confirm whether the improvement in motor function is accompanied by the recovery of motor units. Patients with Brunnstrom grade 6 have close to normal motor function, and the stability or close to normal level of the number of motor units can verify the effect of rehabilitation and the overall recovery of motor function.

[0131] Calculating the average discharge rate of motor units can quantify the overall functional impairment of motor units. This solves the problem of how to accurately assess the functional impairment of motor units in stroke patients and provides a quantitative indicator of neurological damage. The motor control of patients with Brunnstrom grade 4 is still incomplete, and the reduction in the average discharge rate may indicate that neurological impairment still exists. By analyzing the discharge rate at this stage, the difficulties encountered by patients in the recovery process can be quantified. The motor function of patients with Brunnstrom grade 5 improves, and the change in the average discharge rate can reflect the degree of functional recovery. This analysis helps to understand the recovery of motor unit function and the improvement of motor ability. The discharge rate of patients with Brunnstrom grade 6 should be close to normal levels. Analyzing the average discharge rate at this stage can verify whether the expected level of motor control has been achieved and evaluate whether the rehabilitation effect meets the clinical goals.

[0132] Analysis of the discharge pattern of motor units reveals the microscopic effects of stroke on motor control. This addresses how to understand changes in motor unit discharge patterns and their impact on overall motor function, providing deep insights into the mechanisms of motor control. The discharge pattern of motor units in Brunnstrom grade 4 patients may still show significant abnormalities, and changes in discharge patterns can help understand the specific effects of brain damage on motor control and identify challenges in recovery. Improvements in the discharge pattern of Brunnstrom grade 5 patients can reflect improvements in the patient's motor control ability. Analyzing this pattern can help verify actual progress in the recovery process and optimize rehabilitation strategies. For Brunnstrom grade 6 patients, the discharge pattern of motor units should be close to normal levels when their motor function is close to normal. Analyzing this pattern can confirm the stability and consistency of the rehabilitation effect and verify whether the patient's motor control has returned to a near-normal level.

[0133] The evaluation system of the present invention can respectively collect high-density surface electromyography data of the healthy and affected forearms when hemiplegic stroke patients perform different hand movements, and analyze the micro-neuron characteristics, so as to accurately evaluate the degree and progression of motor disorders in hemiplegic stroke patients. The present invention can be used in the rehabilitation process of hemiplegic stroke patients to regularly evaluate the changes in the degree of hand motor disorders and provide quantitative evaluation indicators. According to the evaluation results of the system, doctors can adjust and optimize the treatment plan in time and select the most suitable rehabilitation training and treatment methods for patients. At the same time, the present invention can also be used for research in the field of neurorehabilitation to provide data support for understanding the neuroplasticity and recovery mechanism after stroke. The system of the present invention can realize real-time data acquisition and analysis, provide immediate auxiliary evaluation and feedback, and help to dynamically adjust the rehabilitation plan.

[0134] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0135] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for assessing hand movement disorders in hemiplegic stroke patients based on HD-sEMG, characterized by: The steps include: Step 1: Obtain HD-sEMG signal data; Step 2: Processing the HD-sEMG signal data obtained in step 1; Step 3: Identify and decompose the motion units of the data processed in step 2; Step 4: Perform statistical analysis on the motor unit data decomposed in step 3; Step 5: Evaluate the statistical analysis results in step 4; The step 1 specifically includes: Step 11: The electrode arrays are placed on the flexor muscle groups and the extensor muscle groups of the forearm respectively; Step 12: Complete the gesture according to the guidance, and the electrode array collects the electromyographic signal data and uploads the data to the host computer; Step 13: Repeat step 13 until all gestures are completed and the electromyographic signal data collection is completed.

2. The method for evaluating hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 1, characterized in that: In step 12, the long axis of the electrode array is parallel to the long axis of the subject's forearm.

3. The method for evaluating hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 1, characterized in that: In step 13, the gestures include wrist inversion, wrist eversion, internal rotation, external rotation, fist clenching, thumb and index finger pinching, and thumb, index finger and middle finger pinching.

4. The method for evaluating hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 1, characterized in that: The step 2 specifically includes: Step 21: Use a moving average filter to smooth the signal and remove short-term high-frequency noise; Step 22: Use a notch filter with a dynamic adjustment function to perform directional filtering on power frequency interference and its harmonic components; Step 23: Perform event-based segmentation and labeling on the collected data.

5. The method for evaluating hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 1, characterized in that: The step 3 specifically includes: Step 31: Splicing data of the same gesture action, intercepting a short section of resting state data before and after each sample as a transition interval for splicing; Step 32: After completing data splicing in step 31, use the FastICA algorithm to identify motion units; Step 33: Decompose the identified motion units.

6. The method for assessing hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 5, characterized in that: The step 32 specifically includes: Step 321: Use the delayed copy of the electromyographic signal to expand the electromyographic channel so that subsequent analysis can more accurately capture subtle changes and features of the signal. Step 322: Subtract the mean value of each channel from the signal to ensure that the mean value of the signal is zero. Step 323: performing a linear transformation on the decentralized signal; Step 324: Use the FastICA algorithm to perform reverse calculation on the signal data processed in step 323 to separate independent motion units.

7. The method for assessing hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 5, characterized in that: The FastICA algorithm in step 32 includes a dynamic adjustment mechanism, which specifically includes: dynamic adjustment of non-Gaussian parameters, adaptive learning rate adjustment, dynamic update of weight vectors, stability detection of signal separation, and performance feedback mechanism; The dynamically adjusted non-Gaussian parameters include coefficients of kurtosis and negative entropy; according to the signal non-Gaussian value calculated in real time, the threshold range of kurtosis is adjusted between 0.5 and 1.5 to cope with the signal characteristics under different data conditions; for signals with a kurtosis threshold lower than 0.5, the weight of kurtosis is increased to increase the sensitivity to non-Gaussianity; for signals with a kurtosis threshold higher than 1.5, the weight of kurtosis is reduced to prevent over-emphasis on non-Gaussianity; the kurtosis parameter is adjusted in real time to ensure that the signal obtains the best separation effect under different data conditions; The adjustment range of negative entropy is set to 0.1 to 0.5 to ensure that the algorithm can effectively separate signals under different noise levels; for signals with a negative entropy threshold lower than 0.1, the negative entropy coefficient is increased to improve the algorithm's ability to capture non-Gaussianity; for signals with a negative entropy threshold higher than 0.5, the negative entropy coefficient is reduced to avoid over-separation; the negative entropy parameters are dynamically adjusted to adapt to different noise environments and signal characteristics; The adaptive learning rate adjustment is to dynamically adjust the learning rate according to the error feedback of each iteration, with the initial value set to 0.01 and the adjustment range from 0.001 to 0.1; calculate the error after each iteration; if the error decreases, gradually increase the learning rate, set the growth step to 0.005, until it reaches the upper limit of 0.1; if the error fluctuates or increases, reduce the learning rate, reduce the step to 0.001, until it reaches the lower limit of 0.001; adjust the learning rate according to the error feedback to improve the stability and convergence speed of the algorithm; The dynamic update of the weight vector uses an adaptive weight adjustment strategy, in which the weight update pace is determined by the independence and non-Gaussianity of the current signal; in each iteration, the gradient of the weight vector is calculated; according to the independence and non-Gaussianity indicators of the signal, the weight update pace is dynamically adjusted to ensure that the weight vector optimization achieves the best effect; The stability detection of the signal separation includes calculating the correlation coefficient between independent components. If the correlation coefficient is close to 0, it means that the independence is good; if the correlation coefficient is high, it is necessary to further adjust the algorithm parameters; according to the value of the correlation coefficient, the kurtosis and negative entropy parameters and the learning rate are automatically adjusted to improve the stability of the signal separation; and the algorithm is re-run until the separation effect is stable; The performance feedback mechanism monitors the mean square error and signal independence index of the separation effect in real time. If the values ​​of the mean square error and independence index deviate from the expected range, the algorithm parameters, including kurtosis, negative entropy coefficient and learning rate, are dynamically adjusted; based on the performance feedback, the algorithm parameters are optimized to ensure that the signal separation quality reaches the optimal level.

8. The method for assessing hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 5, characterized in that: In the step 33, the identified motion units are decomposed, specifically, peak detection is performed on the decomposed independent components to identify the activities of each motion unit; k-means cluster analysis is performed on the detected peaks, and a class of peaks with larger values ​​are identified as discharge pulses of the motion units.

9. The method for assessing hand movement disorders in stroke hemiplegic patients based on HD-sEMG according to claim 1, characterized in that: The statistical analysis in step 4 is to statistically evaluate the number of motor units, average discharge rate and discharge pattern of motor units on the healthy and affected sides.

10. An evaluation system using the method for evaluating hand movement disorders in hemiplegic stroke patients based on HD-sEMG according to any one of claims 1 to 9, characterized in that: An HD-sEMG data acquisition system, wherein the HD-sEMG data acquisition system comprises an electromyography collector, wherein the sampling rate of the electromyography collector is 2048 Hz; the HD-sEMG data acquisition system is used to acquire HD-sEMG signal data; A data processing module, which receives the HD-sEMG signal data sent by the HD-sEMG data acquisition system, processes the data, and transmits the result to the motor unit identification and decomposition module; A motion unit identification and decomposition module, which receives the data processed by the data processing module to perform motion unit identification and decomposition, and transmits the result to the data statistical analysis module; A data statistical analysis module, which receives the data processed by the motion unit identification and decomposition module for statistical analysis and transmits the result to the evaluation module; An evaluation module receives and evaluates the data processed by the data statistical analysis module.

Images