Stroke hand function rehabilitation intelligent training health data evaluation system

Abstract

This invention relates to the field of stroke hand function rehabilitation technology, and more particularly to a smart training health data assessment system for stroke hand function rehabilitation. The system includes: a data acquisition and mapping module, a spatial filtering module, a spatiotemporal separation module, a template alignment module, a training adjustment module, and a safety monitoring module. The data acquisition and mapping module acquires multi-channel electromyographic signals of the hand and forearm target muscle groups of stroke patients and establishes a mapping between the acquisition area and the target muscle groups. The spatial filtering module performs spatial filtering on the acquired signals. The spatiotemporal separation module extracts spatiotemporal coordination patterns of the hand and adaptively adjusts the number of coordination patterns. The template alignment module spatially aligns the coordination patterns and calculates scores. The training adjustment module adjusts training parameters, and the safety monitoring module monitors safety indicators. In this invention, the system achieves a complete closed loop of health data assessment and training adjustment, improving the problems of disconnect between assessment and training needs and insufficient adaptability caused by traditional single signal acquisition or offline assessment.

Description

Technical Field

[0001] This invention relates to the field of stroke hand function rehabilitation technology, and in particular to an intelligent training health data assessment system for stroke hand function rehabilitation. Background Technology

[0002] After a stroke, damage to the brain's motor control areas often leads to a decline in the coordination of hand muscles. This manifests as weakness in finger flexion and extension, difficulty in performing fine motor skills such as grasping and pinching, and some patients may also experience muscle spasms or compensatory activation of non-target muscle groups. These problems directly affect the completion of daily self-care activities such as dressing, eating, and washing, thus reducing their quality of life.

[0003] Traditional stroke hand function rehabilitation health data assessments mostly use single signal acquisition or offline assessment methods. Due to the lack of systematic collaborative extraction, real-time adjustment and safety monitoring mechanisms, the assessment and training needs are disconnected and the fit is insufficient. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a smart training health data assessment system for stroke hand function rehabilitation. It aims to improve the traditional stroke hand function rehabilitation health data assessment, which mostly adopts single signal acquisition or offline assessment methods. Due to the lack of systematic collaborative extraction, real-time adjustment and safety monitoring mechanisms, the assessment and training needs are disconnected and the adaptability is insufficient.

[0005] This invention provides the following technical solution: a smart training and health data assessment system for stroke hand function rehabilitation includes: The acquisition and mapping module is used to acquire multi-channel electromyographic signals of target muscle groups in the hand and forearm of stroke patients, establish a mapping between the acquisition area and the target muscle group, and record action trigger information. The spatial filtering module is used to perform spatial filtering on the acquired electromyographic signals and generate a filtered electromyographic signal matrix. The spatiotemporal separation module is used to input the filtered electromyography signal matrix into the deep sparse separation model, extract the spatiotemporal coordination patterns of the hand containing time series and spatial weights, determine the complexity based on the dimension of hand joint activity, adaptively adjust the number of coordination patterns, and update the sparsity constraint parameters of the model in real time. The template alignment module is used to spatially align spatiotemporal collaborative patterns based on a collaborative template library, using deformable constraint strategies such as rotation, scaling, and translation, and to calculate collaborative deviation, compensatory activation index, and comprehensive collaborative quality score. The training adjustment module is used to locate target muscle groups that deviate from the normal template based on the comprehensive coordination quality score, compensatory activation index and target muscle group mapping relationship, adjust the difficulty and load of training movements in real time, and generate feedback prompts. The safety monitoring module is used to monitor the amplitude of electromyographic signals and the angle of hand joints during rehabilitation training. When the amplitude or angle exceeds the safe range, the training parameters are adjusted and an alarm is triggered.

[0006] By adopting the above technical solution, multi-channel electromyography (EMG) signals of the hand and forearm target muscle groups of stroke patients are collected and a mapping between the collection area and the target muscle group is established. Spatial filtering is performed on the collected EMG signals to extract the spatiotemporal coordination patterns of the hand and adaptively adjust the number of coordination patterns. The coordination patterns are spatially aligned and scored. Training parameters are adjusted in real time and safety indicators are monitored, thereby realizing a complete closed loop of health data assessment and training regulation. This improves the problem that traditional stroke hand function rehabilitation health data assessment mostly adopts single signal acquisition or offline assessment methods. Due to the lack of systematic collaborative extraction, real-time adjustment and safety monitoring mechanisms, assessment and training needs are disconnected and the adaptability is insufficient.

[0007] Furthermore, the acquisition of multichannel electromyographic signals from the target muscle groups of the hand and forearm of stroke patients includes the following steps: Multichannel surface electrode arrays were deployed on the target muscle groups of the patient's hand and forearm; The signal of each electrode was calibrated through resting and slight contraction tests; Electromyographic signals were acquired at a high sampling rate, and the start and end timestamps of the training movements were recorded simultaneously. The acquired raw electromyography signals were subjected to DC offset removal and basic filtering.

[0008] Furthermore, establishing the mapping between the acquisition area and the target muscle group includes the following steps: The hand and forearm are divided into standardized collection areas, with each area corresponding to a specific muscle group; Establish a mapping relationship between each electrode signal and the corresponding muscle group region; The mapping relationship was fine-tuned based on the patient's hand shape and muscle distribution; The mapping relationship is stored synchronously with the acquired electromyographic signals.

[0009] Furthermore, the spatial filtering of the acquired electromyographic signals includes the following steps: Determine the neighborhood range of each electrode; Perform Laplacian filtering on each center electrode and calculate the difference between the center electrode signal and the neighboring electrode signals; Gaussian weights are applied to the neighboring electrode signals for weighted summation, and the weighted values ​​are then superimposed onto the center electrode signal. The filtered signals from each electrode are combined into a matrix to generate a filtered electromyographic signal matrix. The filter matrix is ​​standardized.

[0010] Furthermore, inputting the filtered electromyographic signal matrix into the deep sparse separation model includes the following steps: The filtered electromyographic signal matrix is ​​segmented according to the time sequence to form an input window; The input window is input into the encoding layer of the deep sparse separation model to extract local spatiotemporal features and apply sparsity constraints. The encoded features are input into the model's decoding layer to generate a spatiotemporal collaborative pattern output matrix for hand muscle groups; The output matrix contains the activation time series and spatial weight distribution of each muscle group.

[0011] Furthermore, the process of determining complexity based on the gesture joint movement dimension includes the following steps: Dimensional statistics were performed on the finger and palm joints involved in current rehabilitation training gestures; Calculate the number of joint movements or range of motion to generate a gesture complexity score; Map gesture complexity scores to the number of collaborative patterns and adaptively adjust the output structure of the deep sparse separation model. The adjusted number of collaborative patterns will be applied to subsequent spatiotemporal collaborative pattern extraction.

[0012] Furthermore, the spatial alignment of the spatiotemporal cooperative mode includes the following steps: Select the standard collaborative template that is closest to the current gesture from the collaborative template library; The extracted spatiotemporal collaborative patterns are mapped to the template coordinate system; Apply a rotation matrix to the cooperative mode to adjust its orientation so that the mode direction is consistent with the template direction; Apply scaling operations to the collaborative mode to make the spatial weight magnitude range of the mode consistent with the template; Perform a translation operation on the collaborative pattern to align the center of the pattern with the center of the template; The rotation, scaling, and translation parameters are iteratively adjusted according to the minimum mean square error criterion until the cooperative mode and the template achieve the best match.

[0013] Furthermore, the calculation of the collaborative deviation degree, the compensatory activation index, and the comprehensive collaborative quality score includes the following steps: For each muscle group, the difference between its activation time series and the template time series is calculated to obtain the co-deviation degree. Compensatory activation indices are calculated for activation signals that deviate from the normal template but are compensated by other muscle groups. The collaborative deviation and the compensation activation index are weighted and integrated to generate a comprehensive collaborative quality score; Output the synergistic deviation, compensatory activation index, and comprehensive score for each gesture and muscle group, which can be used for subsequent training adjustments.

[0014] Furthermore, the real-time adjustment of training movement difficulty and load includes the following steps: Obtain the comprehensive coordination quality score, compensatory activation index, and target muscle group mapping relationship corresponding to the current rehabilitation training gestures; A muscle group deviation matrix is ​​generated based on the degree to which the target muscle group deviates from the normal template; Based on the preset training difficulty range and load limits, calculate the appropriate resistance, amplitude, and speed parameters for the current movement; The calculated training parameters are sent to the rehabilitation training equipment so that the equipment can adjust the resistance and load of the movement in real time. During training, electromyographic signals and movement execution are continuously monitored, and training parameters are dynamically updated to form a closed-loop regulation. When muscle overactivation or abnormal movement is detected, the load is automatically reduced or the movement is stopped, and a prompt message is generated.

[0015] Furthermore, the monitoring of electromyographic signal amplitude and hand joint angle includes the following steps: By attaching surface electromyography electrodes to target muscle groups in the hand and forearm, multi-channel electromyography signals are collected and converted from analog to digital to generate digital electromyography signal sequences. By fixing wearable angle sensors or inertial measurement units to key joints of the fingers and wrists, data on the rotation angle and bending amplitude of the hand joints are collected, and then filtered and calibrated. The instantaneous amplitude of the digital electromyography signal is calculated, the joint angle data is sampled in real time, and compared with the preset threshold. When the electromyographic amplitude or joint angle exceeds the preset threshold, an alarm signal is generated and the abnormal information is fed back to the training adjustment module.

[0016] The present invention has the following beneficial effects: 1. In this invention, multi-channel electromyographic (EMG) signals of the hand and forearm target muscle groups of stroke patients are collected and a mapping between the collection area and the target muscle groups is established. Spatial filtering is performed on the collected EMG signals to extract the spatiotemporal coordination patterns of the hand and adaptively adjust the number of coordination patterns. The coordination patterns are spatially aligned and scores are calculated. Training parameters are adjusted in real time and safety indicators are monitored, thereby realizing a complete closed loop of health data assessment and training regulation. This improves the problem that traditional stroke hand function rehabilitation health data assessment mostly adopts single signal acquisition or offline assessment methods. Due to the lack of systematic collaborative extraction, real-time adjustment and safety monitoring mechanisms, assessment and training needs are disconnected and the adaptability is insufficient.

[0017] 2. In this invention, by sequentially performing Laplacian filtering, Gaussian neighborhood weighted filtering, and standardization on the acquired electromyographic signals, a high-resolution, low-noise electromyographic signal matrix is ​​generated. This improves upon the problem that traditional electromyographic signal processing mostly uses a single basic filtering method, which does not optimize for the characteristics of easy crosstalk in hand muscle group signals, thus causing signal aliasing and affecting the accuracy of cooperative pattern extraction.

[0018] 3. In this invention, by selecting similar standard collaborative templates, the collaborative pattern is rotated, scaled, translated and adjusted and iteratively matched to achieve individualized and accurate alignment. This improves the problem that traditional collaborative pattern alignment mostly uses fixed templates for direct comparison, which does not take into account individual patient differences, resulting in large alignment errors and inability to accurately quantify the degree of deviation. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the architecture of a stroke hand function rehabilitation intelligent training health data assessment system proposed in this invention; Figure 2 This is a flowchart illustrating a health data assessment method for intelligent training in stroke hand function rehabilitation proposed in this embodiment. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: In a first embodiment of the present invention, the present invention provides a health data assessment system for intelligent training in stroke hand function rehabilitation, such as... Figure 1 As shown, it includes a data acquisition and mapping module, which is used to acquire multi-channel electromyographic signals of the target muscle groups in the hand and forearm of stroke patients, establish a mapping between the acquisition area and the target muscle groups, and record action trigger information. Furthermore, collecting multichannel electromyographic signals from target muscle groups in the hand and forearm of stroke patients includes the following steps: Multichannel surface electrode arrays were deployed on the target muscle groups of the patient's hand and forearm; The signal of each electrode was calibrated through resting and slight contraction tests; Electromyographic signals were acquired at a high sampling rate, and the start and end timestamps of the training movements were recorded simultaneously. The acquired raw electromyography signals were subjected to DC offset removal and basic filtering.

[0022] Specifically, multi-channel electromyography (EMG) signals from target muscle groups in the hand and forearm of stroke patients are acquired by deploying multi-channel surface electrode arrays in these areas. Each electrode corresponds to a muscle region to collect local EMG activity. The electrode signals are first calibrated using resting and slight contraction tests to obtain baseline potentials and determine the signal response characteristics of each channel. High sampling rate EMG signals are acquired while simultaneously recording the start and end timestamps of the training movements. The acquired data are then compiled into a matrix. Indicates the number of electrode channels. Indicates the number of sampling points, each element Indicates the first The first channel in the The voltage values ​​at each sampling time point are used to collect the raw electromyography (EMG) signals. The raw signals are first processed to remove DC offset to eliminate static baseline interference, and then subjected to basic filtering to suppress high-frequency noise and low-frequency drift. The processed EMG signal matrix is ​​then analyzed. This provides input data for subsequent signal analysis and feature extraction. The next step can be... As input, the matrix is ​​fed into a deep sparse separation model or other electromyographic signal analysis algorithms to extract spatiotemporal features and synergistic patterns. Each row of the matrix corresponds to the time series signal of a single electrode, and each column corresponds to the sampled value of different channels at the same time point. Through this processing, continuous electromyographic activity information can be obtained, which can be used to analyze the activation and synergistic patterns of target muscle groups in the hand and forearm during rehabilitation training movements, providing basic data for training adjustment and feedback.

[0023] Furthermore, establishing the mapping between the acquisition area and the target muscle group includes the following steps: The hand and forearm are divided into standardized collection areas, with each area corresponding to a specific muscle group; Establish a mapping relationship between each electrode signal and the corresponding muscle group region; The mapping relationship was fine-tuned based on the patient's hand shape and muscle distribution; The mapping relationship is stored synchronously with the acquired electromyographic signals.

[0024] Specifically, establishing a mapping between the acquisition area and the target muscle group is achieved by dividing the hand and forearm into standardized acquisition areas. Each acquisition area corresponds to a specific muscle group to form a preliminary mapping relationship. Each electrode signal is then associated with its corresponding muscle group region to construct a mapping matrix. Indicates the number of electrode channels. Represents the number of muscle groups, matrix elements A value of 1 indicates the first The signal collected by the first electrode corresponds to the first... If a muscle group is specified, then 0 is used; otherwise, the matrix is ​​adjusted based on the patient's hand shape and muscle distribution. Fine-tuning is performed to adjust the correspondences to accommodate individual differences, ultimately resulting in a mapping matrix. With the acquired electromyographic signal matrix Synchronized storage to form associated datasets This indicates the number of sampling points. In this associated dataset, each row corresponds to the time-series signal of a single electrode channel, and each column corresponds to the mapping identifier of a specific muscle group. This associated dataset can be used as input for deep sparse separation models or other electromyography (EMG) analysis algorithms to extract the spatiotemporal coordination patterns of the target muscle group and achieve targeted adjustment of training movements; mapping matrix. Each element Obtained through standardized region division and electrode positioning calibration, ensuring that the correspondence between electromyographic signals and muscle group locations can be used for subsequent synergistic pattern analysis and training feedback.

[0025] The spatial filtering module is used to perform spatial filtering on the acquired electromyographic signals and generate a filtered electromyographic signal matrix. Furthermore, spatial filtering of the acquired electromyographic signals includes the following steps: Determine the neighborhood range of each electrode; Perform Laplacian filtering on each center electrode and calculate the difference between the center electrode signal and the neighboring electrode signals; Gaussian weights are applied to the neighboring electrode signals for weighted summation, and the weighted values ​​are then superimposed onto the center electrode signal. The filtered signals from each electrode are combined into a matrix to generate a filtered electromyographic signal matrix. The filter matrix is ​​standardized.

[0026] Specifically, spatial filtering of the acquired electromyographic signals is achieved by determining the neighborhood range of each electrode; for the central electrode... First, define its neighborhood set. Then, a Laplacian filter was used to calculate the difference between the center electrode signal and the neighboring electrode signals. The filtering formula is as follows: This represents the time-series signal of the neighboring electrode. The number of neighborhood electrodes; Gaussian weights are applied to the neighborhood electrode signals. The spatial distance between the central electrode and the neighboring electrodes. The standard deviation of the Gaussian function is calculated by weighted summation. Update the center electrode signal; after filtering each electrode, combine all filtered signals into a matrix. Indicates the number of electrode channels. Represents the number of sampling points in the matrix. Standardization is performed to unify the amplitude scale, and the output is a filtered electromyography signal matrix. This matrix can be used as input to a subsequent deep sparse separation model to extract spatiotemporal collaborative patterns of the hand. The standardized parameters are obtained by averaging the values ​​of each channel. with standard deviation The calculations ensure numerical stability and consistency with the analysis in subsequent model processing.

[0027] The spatiotemporal separation module is used to input the filtered electromyography signal matrix into the deep sparse separation model, extract the spatiotemporal coordination patterns of the hand containing time series and spatial weights, determine the complexity based on the dimension of hand joint activity, adaptively adjust the number of coordination patterns, and update the sparsity constraint parameters of the model in real time. Furthermore, inputting the filtered electromyography signal matrix into the deep sparse separation model includes the following steps: The filtered electromyographic signal matrix is ​​segmented according to the time sequence to form an input window; The input window is fed into the encoding layer of the deep sparse separation model to extract local spatiotemporal features and apply sparsity constraints. The encoded features are input into the model's decoding layer to generate a spatiotemporal collaborative pattern output matrix for hand muscle groups; The output matrix contains the activation time series and spatial weight distribution of each muscle group.

[0028] Specifically, the filtered electromyography signal matrix The length is formed by segmenting the time series. input window The input to the deep sparse separation model is used as the encoding layer; the encoding layer extracts local spatiotemporal features through convolution or fully connected operations. And apply sparsity constraints by minimizing the loss function. Optimize model parameters; where To decode the output matrix, For sparse constraint coefficients, It is the Frobenius norm. for Norm; encoding features Input to the decoding layer to generate the spatiotemporal collaborative pattern output matrix of hand muscles. ,matrix Each row in the matrix represents the activation time sequence of the corresponding muscle group, and each column represents the spatial weight distribution of each muscle group at that time point. The output matrix can be used for subsequent cooperative pattern alignment and complexity determination; parameters of the encoding and decoding layers. , The input time series data was obtained through training samples and backpropagation optimization. From the filter matrix The segmentation can be achieved using a sliding window or a non-overlapping window approach. The output matrix S is used to characterize the spatiotemporal coordination pattern of hand muscle groups in rehabilitation training, providing basic data for coordination template alignment and training movement adjustment.

[0029] Furthermore, determining complexity based on the gesture joint movement dimension includes the following steps: Dimensional statistics were performed on the finger and palm joints involved in current rehabilitation training gestures; Calculate the number of joint movements or range of motion to generate a gesture complexity score; Map gesture complexity scores to the number of collaborative patterns and adaptively adjust the output structure of the deep sparse separation model. The adjusted number of collaborative patterns will be applied to subsequent spatiotemporal collaborative pattern extraction.

[0030] Specifically, the complexity is determined based on the dimensions of the hand gestures and joint movements. This is achieved by statistically analyzing the dimensions of the finger and hand joints involved in the current rehabilitation training gestures and constructing joint movement vectors. The number of joints involved in the hand. Indicates the first The range of motion or number of movements of each joint during training exercises; the range of motion is measured by the change in joint angles as measured by sensors. Calculated, i.e. For the first The maximum range of motion of each joint For the first The minimum range of motion of each joint; the number of joint movements is determined by counting joint angles exceeding a preset threshold. Obtain the number of events; obtain the joint motion vector. After normalization, using the formula Generate a gesture complexity score; among which Weights for each joint and satisfying Scoring based on gesture complexity Through mapping function Adaptive determination of the number of collaborative modes and will This method is applied to adjust the output structure of deep sparse separation models, specifically by adjusting the number of output channels in the model's decoding layer, thereby generating a corresponding number of spatiotemporal cooperative pattern matrices in subsequent spatiotemporal cooperative pattern extraction. ;in The input window length is specified; each row in this matrix represents the time series and spatial weight distribution of a cooperative pattern, which serves as the data input for the next step of template alignment and training adjustment modules, accurately reflecting the complexity of gestures and guiding the adjustment of training actions.

[0031] The template alignment module is used to spatially align spatiotemporal collaborative patterns based on a collaborative template library, using deformable constraint strategies such as rotation, scaling, and translation, and to calculate collaborative deviation, compensatory activation index, and comprehensive collaborative quality score. Furthermore, spatial alignment of the spatiotemporal cooperative model includes the following steps: Select the standard collaborative template that is closest to the current gesture from the collaborative template library; The extracted spatiotemporal collaborative patterns are mapped to the template coordinate system; Apply a rotation matrix to the cooperative mode to adjust its orientation so that the mode direction is consistent with the template direction; Apply scaling operations to the collaborative mode to make the spatial weight magnitude range of the mode consistent with the template; Perform a translation operation on the collaborative pattern to align the center of the pattern with the center of the template; The rotation, scaling, and translation parameters are iteratively adjusted according to the minimum mean square error criterion until the cooperative mode and the template achieve the best match.

[0032] Specifically, when performing spatial alignment of spatiotemporal cooperative patterns, the closest standard cooperative template is first selected from the cooperative template library based on the current training gesture. The cooperative template is represented in matrix form. This indicates the number of muscle groups contained in the template. The time series length is represented by a template pre-constructed from the demonstration training data; the extracted spatiotemporal co-pattern is represented as... Its elements are derived from the spatial weights and activation time series output by the deep sparse separation model, and are used in the alignment process. Map the coordinates to the template; then set the rotation matrix. Scaling factor and translation vector Forming a deformable mapping structure to transform cooperative modes into ;in The vector consists of rows with all elements equal to 1, ensuring the translation operation adapts to the time series dimension. To make the transformed pattern as consistent as possible with the template, the objective function is constructed using the minimum mean square error criterion. for Norm, and adjust the rotation matrix through iterative optimization. Each element and scaling factor and translation vector The components cause the error Continuously decrease; the iterative process is based on the change in error. ;in Using a preset convergence threshold and a stopping condition, the optimal transformation parameters are finally obtained, forming the aligned cooperative mode. The alignment result is used to subsequently calculate the synergistic deviation, compensatory activation index, and comprehensive synergistic quality score. The synergistic deviation can be calculated by measuring the difference between the aligned pattern and the template, the compensatory activation index can be calculated based on the weight changes of non-target muscle groups, and the comprehensive synergistic quality score serves as the input for the next step of rehabilitation training difficulty adjustment and feedback generation module, which is used to improve the matching degree of training movements and the personalized adaptability of training programs.

[0033] Furthermore, the calculation of the collaborative deviation degree, the compensatory activation index, and the comprehensive collaborative quality score includes the following steps: For each muscle group, the difference between its activation time series and the template time series is calculated to obtain the co-deviation degree. Compensatory activation indices are calculated for activation signals that deviate from the normal template but are compensated by other muscle groups. The collaborative deviation and the compensation activation index are weighted and integrated to generate a comprehensive collaborative quality score; Output the synergistic deviation, compensatory activation index, and comprehensive score for each gesture and muscle group, which can be used for subsequent training adjustments.

[0034] Specifically, the calculation of the collaborative deviation degree, the compensatory activation index, and the comprehensive collaborative quality score uses the template-aligned collaborative pattern as input; where the collaborative pattern is represented as a matrix. Standard collaborative templates are represented as matrices. The rows of the matrix correspond to muscle groups, and the columns correspond to time series. The elements are derived from the aforementioned deep sparse separation model and spatial alignment steps. First, the difference between the activation time series of each muscle group and the template is calculated, and a deviation metric is constructed using the squared error form. The degree of cooperative deviation can be expressed as This result indicates the degree of deviation of the synergistic behavior of muscle groups from the template over the complete time series. Subsequently, based on the definition of compensation, muscle groups with low activation in the template but abnormally high activation in the synergistic pattern were identified, and a compensatory activation index was constructed with reference to the degree of deviation of the target muscle group, which can be expressed as: This indicator reflects the additional contribution of muscle groups during non-target activation phases; then, the synergistic deviation and compensatory activation indicators are weighted and fused, and the comprehensive score can be expressed as... and For preset weights and satisfying The degree of coordination deviation of each muscle group is determined by the configuration of the rehabilitation plan; Compensation activation indicators and comprehensive collaborative quality score The output is used in the subsequent training adjustment module. The training adjustment module adjusts the intensity of training movements, muscle load distribution, and feedback prompts based on the score, so that the training process is more in line with the target coordination mode and provides compensatory adjustments when the deviation is large, thereby improving the pertinence and effectiveness of rehabilitation training.

[0035] The training adjustment module is used to locate target muscle groups that deviate from the normal template based on the comprehensive coordination quality score, compensatory activation index and target muscle group mapping relationship, adjust the difficulty and load of training movements in real time, and generate feedback prompts. Furthermore, adjusting the difficulty and load of training exercises in real time includes the following steps: Obtain the comprehensive coordination quality score, compensatory activation index, and target muscle group mapping relationship corresponding to the current rehabilitation training gestures; A muscle group deviation matrix is ​​generated based on the degree to which the target muscle group deviates from the normal template; Based on the preset training difficulty range and load limits, calculate the appropriate resistance, amplitude, and speed parameters for the current movement; The calculated training parameters are sent to the rehabilitation training equipment so that the equipment can adjust the resistance and load of the movement in real time. During training, electromyographic signals and movement execution are continuously monitored, and training parameters are dynamically updated to form a closed-loop regulation. When muscle overactivation or abnormal movement is detected, the load is automatically reduced or the movement is stopped, and a prompt message is generated.

[0036] Specifically, the training and adjustment module uses a comprehensive collaborative quality scoring matrix. Compensation activation indicators and the mapping relationship of target muscle groups As input, the aforementioned scores and indicators are derived from the collaborative pattern analysis process, and the mapping relationship is provided by the acquisition and mapping module, which is used to identify the muscle groups that are expected to be activated in the gesture. The total number of muscle groups; firstly based on and Constructing muscle group deviation matrix This matrix describes the degree of deviation in the synergistic behavior of each muscle group, and can be represented as follows: and The fusion coefficients set for the training strategy and satisfying Then, combined with the preset training difficulty range and load limit range The appropriate resistance, amplitude, and speed parameters for the current action are calculated using the deviation matrix. For example, the resistance parameter can be expressed as... The minimum resistance threshold, The load adjustment factor determines whether the resistance is increased or decreased based on the degree of deviation from the target muscle group; amplitude parameter. With speed parameters Using the same method, through linear mapping rules Calculation; where As a reference amplitude and speed, , The adjustment coefficients ultimately form the motion control parameter vector. The generated training parameters are sent to the rehabilitation training equipment, enabling the equipment to adjust resistance and load in real time during the movement execution; the system continuously monitors electromyographic signals and movement execution, and detects a new frame of synergistic pattern signal. At that time, the module recalculates the deviation matrix and updates the parameters to achieve closed-loop regulation; if muscle group activation values ​​are detected during the monitoring process... Clearly exceeded the template activation limit Then by condition Triggering protection logic, automatically reducing the load to The device can either stop the movement and generate a prompt message; the final output training parameters are used to control the rehabilitation training equipment to perform the next movement, so that the training difficulty matches the patient's condition, improves the standardization of muscle activation and synergy patterns during training, and ensures that the movement is performed within a safe range.

[0037] The safety monitoring module is used to monitor the amplitude of electromyographic signals and the angle of hand joints during rehabilitation training. When the amplitude or angle exceeds the safe range, the training parameters are adjusted and an alarm is triggered. Furthermore, monitoring the amplitude of electromyographic signals and the angle of the hand joints includes the following steps: By attaching surface electromyography electrodes to target muscle groups in the hand and forearm, multi-channel electromyography signals are collected and converted from analog to digital to generate digital electromyography signal sequences. By fixing wearable angle sensors or inertial measurement units to key joints of the fingers and wrists, data on the rotation angle and bending amplitude of the hand joints are collected, and then filtered and calibrated. The instantaneous amplitude of the digital electromyography signal is calculated, the joint angle data is sampled in real time, and compared with the preset threshold. When the electromyographic amplitude or joint angle exceeds the preset threshold, an alarm signal is generated and the abnormal information is fed back to the training adjustment module.

[0038] Specifically, the safety monitoring module uses multi-channel digital electromyography signal sequences. With joint angle sequence For input; where This represents the number of electromyographic signal channels. To monitor the number of joints, the former (electromyography) is acquired by surface electromyography electrodes attached to the muscles of the hand and forearm and obtained through analog-to-digital conversion, while the latter (electromyography) is acquired by angle sensors or inertial measurement units fixed to key joints of the fingers and wrist and obtained through filtering and calibration. First, the instantaneous amplitude of the electromyography signal for each channel is calculated and expressed as an absolute value average. ; Given the length of the sliding window, the angle sample value of each joint is directly expressed as... Subsequently, the instantaneous electromyographic amplitude and joint angle were compared with the corresponding preset safety thresholds. and Comparison, by conditions or The module determines whether there is a risk of exceeding limits during the training process; when any condition is met, the module generates an alarm signal and outputs an index of abnormal muscle groups. Index of Abnormal Joints This forms a set of abnormal information. The abnormal information is sent to the training adjustment module in real time, which is used to limit or reduce the parameters of movement resistance, amplitude and speed in the next training cycle. This allows the training equipment to make adjustments when the patient exerts excessive force or the joints exceed the safe range, thereby avoiding muscle fatigue, joint hyperextension or movement injury in subsequent training and improving the safety and stability of the rehabilitation training process.

[0039] Example 2: A method for assessing health data in intelligent training for stroke hand function rehabilitation includes the following steps: Multichannel electromyographic signals of target muscle groups in the hand and forearm of stroke patients were collected, and a mapping between the collection area and the target muscle group was established. At the same time, action triggering information was recorded. Spatial filtering is performed on the acquired electromyography (EMG) signals to generate a filtered EMG signal matrix. The filtered electromyography signal matrix is ​​input into the deep sparse separation model to extract the spatiotemporal coordination pattern of the hand containing time series and spatial weights. The complexity is determined according to the dimension of hand joint movement, the number of coordination patterns is adaptively adjusted, and the sparsity constraint parameters of the model are updated in real time. Based on the collaborative template library, a deformable constraint strategy of rotation, scaling, and translation is adopted to spatially align the spatiotemporal collaborative patterns, and calculate the collaborative deviation degree, compensatory activation index, and comprehensive collaborative quality score. Based on the comprehensive synergistic quality score, compensatory activation index and target muscle group mapping relationship, the target muscle groups that deviate from the normal template are located, the difficulty and load of training movements are adjusted in real time, and feedback prompts are generated. During rehabilitation training, the amplitude of electromyographic signals and the angle of hand joints are monitored. When the amplitude or angle exceeds the safe range, the training parameters are adjusted and an alarm is triggered.

[0040] A 52-year-old stroke patient, three months post-stroke, was admitted to the rehabilitation department of a top-tier hospital. The patient presented with weakness in the right hand's grasp and incoordination of finger flexion and extension. Traditional rehabilitation training relied on subjective adjustments by therapists, failing to quantify muscle group coordination deficiencies and easily leading to muscle fatigue or joint discomfort due to excessive exertion, making it difficult to establish a safe and personalized training loop. To address these issues, the intelligent training health data assessment method for stroke hand function rehabilitation provided by this invention was adopted. The process is as follows: Figure 2 As shown. The specific implementation process of this method is as follows: A multi-channel surface electrode array was deployed on the back of the patient's right hand and forearm, covering key muscle groups. The electrodes were calibrated using resting-state testing and gentle grasping movements. Electromyographic (EMG) signals were acquired at a high sampling rate, and training movement timestamps were recorded. A mapping relationship between the electrodes and muscle groups was established by combining hand shape data. This step achieves comprehensive coverage and precise localization of key muscle group signals, avoiding the limitations of traditional single-electrode signal analysis. It provides stable and traceable raw data for subsequent collaborative analysis, solving the problem of not being able to accurately correlate EMG signals with specific muscle groups.

[0041] First, Laplacian filtering is performed on the raw EMG signal. Then, Gaussian weights are applied based on the physical distance between the electrodes for neighborhood weighting. Finally, the filtered signals are combined to form a matrix and standardized. This step reduces crosstalk and noise in the EMG signal, generates a high-resolution signal matrix with consistent amplitude, avoids noise interference in subsequent analysis, solves the problem of inaccurate evaluation caused by poor signal quality, and provides high-quality input for deep sparse separation models.

[0042] The filter matrix is ​​segmented over time to form an input window, which is then input into a lightweight convolutional autoencoder. Spatiotemporal features are extracted and sparse constraints are applied. The complexity score is calculated by statistically analyzing the gesture joint activity dimension. The number of collaborative patterns is adaptively adjusted, and the decoding layer outputs a spatiotemporal collaborative pattern matrix. This step adapts to different gesture complexities, achieves accurate extraction of muscle group collaborative patterns, avoids the limitations of traditional fixed patterns, and solves the problem of difficulty in quantifying the spatiotemporal collaboration of hand muscles, providing quantifiable features for subsequent alignment evaluation.

[0043] By selecting similar standard coordination templates, the coordination patterns are rotated, scaled, and translated for adjustment, iteratively optimized until the best match is achieved. Then, the coordination deviation degree and compensatory activation index of each muscle group are calculated, and a weighted fusion is used to generate a comprehensive score. This step eliminates the influence of individual differences in hand shape, accurately quantifies coordination deviation and compensation, avoids the evaluation errors of traditional fixed templates, solves the problem of not being able to individually determine muscle group coordination defects, and provides a clear basis for training adjustment based on deviations.

[0044] Based on the comprehensive score, compensatory indicators, and muscle group mapping relationship, a deviation matrix is ​​generated. Appropriate parameters are calculated by combining the training range and load limitations and sent to the training device. The parameters are continuously monitored and dynamically updated, generating feedback to guide the patient. This step enables individualized dynamic adjustment of the training plan, specifically improving muscle group coordination deficiencies, avoiding the shortcomings of traditional one-size-fits-all training, solving the problems of poor training adaptability and inability to accurately correct deviations, and improving the targeting of training.

[0045] The system collects digital electromyography (EMG) signals and joint angle data, calculates the instantaneous amplitude of the EMG signals and samples the joint angle, compares it with preset thresholds, and generates an alarm signal and adjusts training parameters if the thresholds are exceeded. This step mitigates safety risks in real time, avoiding muscle fatigue injury or joint hyperextension, solving the problem of traditional training lacking real-time safety monitoring, ensuring training safety, and improving patient compliance.

[0046] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart training and health data assessment system for stroke hand function rehabilitation, characterized in that, include: The acquisition and mapping module is used to acquire multi-channel electromyographic signals of target muscle groups in the hand and forearm of stroke patients, establish a mapping between the acquisition area and the target muscle group, and record action trigger information. The spatial filtering module is used to perform spatial filtering on the acquired electromyographic signals and generate a filtered electromyographic signal matrix. The spatiotemporal separation module is used to input the filtered electromyography signal matrix into the deep sparse separation model, extract the spatiotemporal coordination patterns of the hand containing time series and spatial weights, determine the complexity based on the dimension of hand joint activity, adaptively adjust the number of coordination patterns, and update the sparsity constraint parameters of the model in real time. The template alignment module is used to spatially align spatiotemporal collaborative patterns based on a collaborative template library, using deformable constraint strategies such as rotation, scaling, and translation, and to calculate collaborative deviation, compensatory activation index, and comprehensive collaborative quality score. The training adjustment module is used to locate target muscle groups that deviate from the normal template based on the comprehensive coordination quality score, compensatory activation index and target muscle group mapping relationship, adjust the difficulty and load of training movements in real time, and generate feedback prompts. The safety monitoring module is used to monitor the amplitude of electromyographic signals and the angle of hand joints during rehabilitation training. When the amplitude or angle exceeds the safe range, the training parameters are adjusted and an alarm is triggered.

2. The intelligent training and health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The process of collecting multichannel electromyographic signals from the target muscle groups in the hand and forearm of stroke patients includes the following steps: Multichannel surface electrode arrays were deployed on the target muscle groups of the patient's hand and forearm; The signal of each electrode was calibrated through resting and slight contraction tests; Electromyographic signals were acquired at a high sampling rate, and the start and end timestamps of the training movements were recorded simultaneously. The acquired raw electromyography signals were subjected to DC offset removal and basic filtering.

3. The intelligent training and health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The process of establishing the mapping between the acquisition area and the target muscle group includes the following steps: The hand and forearm are divided into standardized collection areas, with each area corresponding to a specific muscle group; Establish a mapping relationship between each electrode signal and the corresponding muscle group region; The mapping relationship was fine-tuned based on the patient's hand shape and muscle distribution; The mapping relationship is stored synchronously with the acquired electromyographic signals.

4. The intelligent training health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The spatial filtering of the acquired electromyographic signals includes the following steps: Determine the neighborhood range of each electrode; Perform Laplacian filtering on each center electrode and calculate the difference between the center electrode signal and the neighboring electrode signals; Gaussian weights are applied to the neighboring electrode signals for weighted summation, and the weighted values ​​are then superimposed onto the center electrode signal. The filtered signals from each electrode are combined into a matrix to generate a filtered electromyographic signal matrix. The filter matrix is ​​standardized.

5. The intelligent training and health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The process of inputting the filtered electromyographic signal matrix into the deep sparse separation model includes the following steps: The filtered electromyographic signal matrix is ​​segmented according to the time sequence to form an input window; The input window is input into the encoding layer of the deep sparse separation model to extract local spatiotemporal features and apply sparsity constraints. The encoded features are input into the model's decoding layer to generate a spatiotemporal collaborative pattern output matrix for hand muscle groups; The output matrix contains the activation time series and spatial weight distribution of each muscle group.

6. The intelligent training health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The process of determining complexity based on the dimension of hand gesture joint movement includes the following steps: Dimensional statistics were performed on the finger and palm joints involved in current rehabilitation training gestures; Calculate the number of joint movements or range of motion to generate a gesture complexity score; Map gesture complexity scores to the number of collaborative patterns and adaptively adjust the output structure of the deep sparse separation model. The adjusted number of collaborative patterns will be applied to subsequent spatiotemporal collaborative pattern extraction.

7. The intelligent training health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The spatial alignment of the spatiotemporal cooperative mode includes the following steps: Select the standard collaborative template that is closest to the current gesture from the collaborative template library; The extracted spatiotemporal collaborative patterns are mapped to the template coordinate system; Apply a rotation matrix to the cooperative mode to adjust its orientation so that the mode direction is consistent with the template direction; Apply scaling operations to the collaborative mode to make the spatial weight magnitude range of the mode consistent with that of the template; Perform a translation operation on the collaborative pattern to align the center of the pattern with the center of the template; The rotation, scaling, and translation parameters are iteratively adjusted according to the minimum mean square error criterion until the cooperative mode and the template achieve the best match.

8. The intelligent training and health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The calculation of the collaborative deviation, compensatory activation index, and comprehensive collaborative quality score includes the following steps: For each muscle group, the difference between its activation time series and the template time series is calculated to obtain the co-deviation degree. Compensatory activation indices are calculated for activation signals that deviate from the normal template but are compensated by other muscle groups. The collaborative deviation and the compensation activation index are weighted and integrated to generate a comprehensive collaborative quality score; Output the synergistic deviation, compensatory activation index, and comprehensive score for each gesture and muscle group, which can be used for subsequent training adjustments.

9. The intelligent training health data assessment system for stroke hand function rehabilitation according to claim 1, characterized in that, The real-time adjustment of training exercise difficulty and load includes the following steps: Obtain the comprehensive coordination quality score, compensatory activation index, and target muscle group mapping relationship corresponding to the current rehabilitation training gestures; A muscle group deviation matrix is ​​generated based on the degree to which the target muscle group deviates from the normal template; Based on the preset training difficulty range and load limits, calculate the appropriate resistance, amplitude, and speed parameters for the current movement; The calculated training parameters are sent to the rehabilitation training equipment so that the equipment can adjust the resistance and load of the movement in real time. During training, electromyographic signals and movement execution are continuously monitored, and training parameters are dynamically updated to form a closed-loop regulation. When muscle overactivation or abnormal movement is detected, the load is automatically reduced or the movement is stopped, and a prompt message is generated.

10. A stroke hand function rehabilitation intelligent training health data assessment system according to claim 1, characterized in that, The monitoring of electromyographic signal amplitude and hand joint angle includes the following steps: By attaching surface electromyography electrodes to target muscle groups in the hand and forearm, multi-channel electromyography signals are collected and converted from analog to digital to generate digital electromyography signal sequences. By fixing wearable angle sensors or inertial measurement units to key joints of the fingers and wrists, data on the rotation angle and bending amplitude of the hand joints are collected, and then filtered and calibrated. The instantaneous amplitude of the digital electromyography signal is calculated, the joint angle data is sampled in real time, and compared with the preset threshold. When the electromyographic amplitude or joint angle exceeds the preset threshold, an alarm signal is generated and the abnormal information is fed back to the training adjustment module.

Images