Rheumatoid arthritis patient hand function rehabilitation effect evaluation system combined with bcw theory

Abstract

The present application relates to the technical field of health monitoring, in particular to a rheumatoid arthritis patient hand function rehabilitation effect evaluation system combined with BCW theory, which comprises a behavior analysis module, a capability extraction module, an opportunity evaluation module, an effect analysis module and a stage optimization module.In the present application, by collecting hand trajectory and task record, the execution rate and duration are extracted and normalized fitting, the dynamic behavior image is constructed, the principal component extraction is carried out combined with flexion angle and electromyographic signal, the high weight capability feature is generated, the response time difference of monitoring device trigger is calculated, the response delay change rate is compared with task time, the relationship between external stimulus and capability activation is revealed, the effect surface is constructed by fusing behavior, capability and response feature, the stability and gradient direction are analyzed, the rehabilitation effect stereoscopic quantization and progress trend presentation are realized, the differentiated rehabilitation path is adapted, and the precision and regulation ability of evaluation are improved.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and in particular to an evaluation system for the rehabilitation effect of hand function in rheumatoid arthritis patients that incorporates the BCW theory. Background Technology

[0002] The field of health monitoring technology encompasses technologies related to the acquisition, analysis, and management of human physiological information. Its core components include acquiring physiological or motor state information of specific parts of the human body through biosignal acquisition devices, performing feature identification and classification of the acquired information using sensing and recognition methods, and then monitoring and assessing human health status through medical data analysis methods. Covering multiple areas such as physiological signal acquisition, motor function recognition, disease monitoring and assessment, and information management of health status, it has formed a continuous and quantitative health status monitoring technology system based on physiological data, capable of supporting monitoring and medical analysis scenarios for various populations under different health states.

[0003] Among them, the hand function rehabilitation effect evaluation system for rheumatoid arthritis patients based on BCW theory refers to the quantitative monitoring and evaluation of hand function status in rheumatoid arthritis patients during rehabilitation training. Its technical aspects include: structuring rehabilitation training behaviors based on the behavior change wheel theory framework; acquiring joint movement trajectories, muscle exertion information, and movement completion rates during training using hand movement recognition and biomechanical feature extraction technologies; acquiring data on hand function status-related indicators through rehabilitation process information collection methods; and quantitatively evaluating changes in hand function based on a disease rehabilitation assessment rule system.

[0004] While existing technologies cover areas such as physiological signal acquisition, motion recognition, and disease assessment, they suffer from isolated indicator systems and single feedback paths at the implementation level. They generally rely on static feature extraction and rule-driven assessment models, resulting in a lack of continuous dynamic tracking of changes in patients' actual training behaviors. For example, simply collecting electromyography or joint motion trajectories often only reflects a single dimension of functional status, lacking cross-dimensional behavior-ability-response coupling analysis, making it difficult to reveal the intrinsic mechanisms of functional recovery. Furthermore, most systems employ staged sampling and single-point judgment, failing to form real-time training behavior evolution modeling and intervention adaptation feedback, easily leading to response delays, intervention blind spots, and resource waste during the rehabilitation process. In addition, health status assessments often rely on expert experience to construct indicator threshold systems. Faced with rehabilitation scenarios with high individual variability and non-linear changes, misjudgments or insufficient generalization of assessment results are prone to occur, limiting their applicability in refined management and individualized assessment. For example, in the hand function rehabilitation of rheumatoid arthritis patients, different patients respond significantly differently to external device stimuli. Without a dynamic opportunity domain monitoring mechanism, it is impossible to accurately determine the sensitivity and effectiveness of rehabilitation strategies, thus affecting the optimization of rehabilitation pathways and treatment outcomes. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a system for evaluating the rehabilitation effect of hand function in rheumatoid arthritis patients based on the BCW theory. The technical solution is as follows:

[0006] On the one hand, a system for evaluating the rehabilitation effect of hand function in rheumatoid arthritis patients, combining the BCW theory, is provided. This system includes: The behavior analysis module collects the patient's hand trajectory and training task records during rehabilitation training, extracts the hand movement execution rate and task duration, normalizes them, performs correlation fitting, generates a behavior feature set, and transmits it to the ability extraction module. The ability extraction module, based on the behavioral feature set, collects the interphalangeal flexion and extension angles and surface electromyography signals of the patient's hand, analyzes the angle deviation and muscle group signal synchronization sequence, inputs them into principal component analysis for principal component extraction, calculates the principal component load, generates the ability feature set, and transmits it to the opportunity assessment module. The opportunity assessment module monitors the time difference between the signal triggering of the external rehabilitation assistive device and the patient's response, extracts the external response delay and intervention frequency, calculates the rate of change of the external response delay in combination with the ability feature set, compares it with the task duration, generates a dynamic feature set of the opportunity domain, and transmits it to the effect analysis module. The effect analysis module performs weight normalization on the behavioral feature set, ability feature set and opportunity domain dynamic feature set, calculates the cross-weighted shift value and constructs the effect surface, analyzes the stability, volatility and gradient direction of the surface, and generates the hand function rehabilitation effect evaluation result. The phase optimization module, based on the hand function rehabilitation effect evaluation results, locates the areas of abnormal fluctuations in the effect surface and performs smoothing and error correction, analyzes the trend stability of the corrected surface, and generates optimized rehabilitation phase evaluation results.

[0007] As a further aspect of the present invention, the behavioral feature set includes hand movement execution rate features and task duration features; the ability feature set includes principal components of finger flexion and extension angles, principal components of muscle group signal synchronization, and principal component load distribution features; the opportunity domain dynamic feature set includes external response delay change rate, intervention frequency, and task duration comparison; the hand function rehabilitation effect evaluation results include effect surface stability, surface fluctuation, and surface gradient direction; and the optimized rehabilitation stage evaluation results include abnormal fluctuation area correction and abnormal surface fluctuation.

[0008] As a further aspect of the present invention, the behavior analysis module includes: The data acquisition submodule collects the trajectory coordinate sequence of the patient's hand during rehabilitation training and the task execution record and performs synchronous correction, calculates the displacement change between adjacent trajectory points and performs smoothing, organizes the displacement change data in each training task stage, and generates a hand trajectory dataset. The rate calculation submodule calculates the instantaneous rate between adjacent trajectory points based on the hand trajectory dataset, then calculates the action execution rate for multiple time periods within the task, extracts the start and end times of the task to calculate the task duration, normalizes the action execution rate, generates a normalized action rate sequence, and records the task duration. The feature fitting submodule calls the normalized rate sequence of the action, takes the task duration as the input variable and the rate normalization value as the output variable, calculates the correlation coefficient and constructs a trend distribution curve, and combines the results with the task number to cluster and integrate them to generate a behavioral feature set.

[0009] As a further aspect of the present invention, the capability extraction module includes: The signal acquisition submodule, based on the behavioral feature set, acquires the flexion and extension angles of the patient's fingers, records multi-joint angle data according to the sampling time points within the training cycle, calculates the angle difference between adjacent frames and filters out outliers, integrates and sums the continuous angle differences in the time series, and generates a finger angle deviation sequence. The electromyography synchronization submodule, based on the interdigital angle deviation sequence, synchronously acquires and registers electromyography signals on the surface of the hand, calculates the average potential of the muscle group in the corresponding interval and analyzes the electromyography amplitude sequence, calls the angle deviation value and the electromyography amplitude value to perform correlation calculation, and generates an angle electromyography synchronization sequence. The principal component calculation submodule calculates the covariance between multiple signal channels based on the angle electromyography synchronization sequence, performs feature decomposition, and inputs it into principal component analysis to extract principal components. It analyzes the weight ratio of multiple components in the original signal, selects key signal contribution factors based on the load vector, and generates a capability feature set.

[0010] As a further aspect of the present invention, the principal component analysis consists of principal components, eigenvalues, and loading vectors.

[0011] As a further aspect of the present invention, the opportunity assessment module includes: The signal monitoring submodule monitors the trigger signal generated by the external rehabilitation assistive device when it is triggered and the response signal recorded by the patient's hand sensor. It calculates the time interval between each set of trigger and response signals, filters out recorded values ​​with abnormal time intervals, calculates the mean value after filtering, and generates an external response delay sequence. The frequency extraction submodule, based on the external response delay sequence, counts the number of times the external rehabilitation assistive device is triggered and the number of times the patient responds within a set period, calculates the ratio between the two and normalizes it as the time-triggered response frequency, analyzes the frequency change trend, and generates intervention frequency distribution results. The feature generation submodule calls the intervention frequency distribution results and combines them with the capability feature set. Based on the time series of external response delay, it calculates the delay change rate within a continuous period, compares the corresponding ratio of the delay change rate with the task duration, extracts time-related change parameters, and generates a dynamic feature set of the opportunity domain.

[0012] As a further aspect of the present invention, the effect analysis module includes: The weight normalization submodule obtains the behavioral feature set, capability feature set and opportunity domain dynamic feature set for sampling and statistics, calculates the weight ratio coefficients of multiple feature dimensions under a unified scale, performs linear normalization on the weight ratio coefficients, and generates a feature weight normalization value set. The cross-weight shift submodule extracts the corresponding weight values ​​of behavioral features and ability features based on the normalized set of feature weights, calculates the weight deviation between the two and compares it with the rate of change in the dynamic features of the opportunity domain, analyzes and summarizes the cross-coupling between features, analyzes the average weight shift index, and generates a cross-weight shift value set. The surface assessment submodule calls the cross-weighted shift value set, maps multiple cross-weighted shift values ​​to a three-dimensional coordinate plane, associates them with the time series to construct an effect surface, calculates the surface volatility and extracts the surface gradient direction, and integrates the numerical values ​​with the overall surface stability index to generate a hand function rehabilitation effect assessment result.

[0013] As a further aspect of the present invention, the stage optimization module includes: The fluctuation detection submodule, based on the hand function rehabilitation effect evaluation results, extracts the response value sequence at multiple times in the effect surface, calculates the response change rate between adjacent times and counts the change rate difference value, filters out the region where the change rate difference value exceeds the fluctuation threshold, and generates a set of abnormal fluctuation intervals of the surface. The surface correction submodule calls the set of abnormal fluctuation intervals of the surface, calculates the mean of the time point response for the time point response sequence within multiple intervals, performs adjacent interpolation, recalculates the continuous change sequence of the interval time point response, adjusts the deviation according to the time point smoothness coefficient, and generates a set of surface smoothing correction values. The trend analysis submodule extracts the response change sequence under the time series based on the surface smoothing correction value set, calculates the rate of change of adjacent time periods and calculates the variance of the rate of change, calculates the stability coefficient based on the variance change trend, and generates the optimized rehabilitation stage evaluation results.

[0014] As a further aspect of the present invention, the fluctuation threshold is set by calculating the response change rate at multiple time points within the time series and statistically analyzing the distribution characteristics of the difference values, with the standard deviation of the difference values ​​as the benchmark.

[0015] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By acquiring hand trajectories and training task records during rehabilitation training, and performing normalized correlation fitting between hand movement execution rate and task duration, a precise profile with dynamic behavioral characteristics can be constructed, providing an accurate behavioral basis for subsequent interpretation of ability status. Based on this, the synchronous acquisition of interdigital flexion-extension angles and surface electromyography (EMG) signals is introduced. Through coupled analysis of angle deviation sequences and muscle group signals, and principal component loading calculation, high-weight ability features can be extracted from multidimensional EMG and motion parameters, improving the objectivity and sensitivity of rehabilitation status quantification. Furthermore, by monitoring the time difference between the triggering of external rehabilitation assistive devices and the patient's response, external response delay and intervention frequency are extracted. Combined with the ability feature sequence, the rate of change of external response delay is calculated and cross-compared with task duration, revealing the dynamic relationship between external stimuli and intrinsic ability activation. Finally, behavioral, ability, and opportunity domain features are uniformly weighted and cross-weighted shift values ​​are calculated to construct an effect surface and analyze its stability and gradient change direction. This not only achieves a three-dimensional quantification of rehabilitation effects but also presents rehabilitation progress trends and efficacy-sensitive changes. The overall logic embodies a closed-loop feedback system that collaboratively models three elements: behavioral dynamics, ability status, and external intervention opportunities. This system can enhance the interpretability and controllability of assessment results, is suitable for rheumatoid hand functional rehabilitation scenarios with significant individual differences and complex recovery paths, and significantly improves the accuracy and real-time response capability of rehabilitation assessment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the system of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention; Figure 3 This is a flowchart of the behavior analysis module in this invention; Figure 4 This is a flowchart of the capability extraction module in this invention; Figure 5 This is a flowchart of the opportunity assessment module in this invention; Figure 6 This is a flowchart of the effect analysis module in this invention; Figure 7 This is a flowchart of the stage optimization module in this invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] This invention provides an evaluation system for the rehabilitation effect of hand function in rheumatoid arthritis patients based on the BCW theory, such as... Figure 1-2 The diagram shown illustrates a system for evaluating the rehabilitation effects of hand function in rheumatoid arthritis patients, incorporating the BCW theory. This system includes: The behavior analysis module collects the patient's hand trajectory and training task records during rehabilitation training, extracts the hand movement execution rate and task duration, normalizes them, performs correlation fitting, generates a behavior feature set, and transmits it to the ability extraction module. The ability extraction module, based on the behavioral feature set, collects the flexion and extension angles of the patient's fingers and surface electromyography signals, analyzes the angle deviation and muscle group signal synchronization sequence, inputs them into principal component analysis for principal component extraction, calculates the principal component load, generates the ability feature set, and transmits it to the opportunity assessment module. The opportunity assessment module monitors the time difference between the signal triggering of external rehabilitation assistive devices and the patient's response, extracts the external response delay and intervention frequency, calculates the rate of change of external response delay by combining the ability feature set, compares it with the task duration, generates a dynamic feature set of the opportunity domain, and transmits it to the effect analysis module. The effect analysis module normalizes the weights of the behavioral feature set, ability feature set, and opportunity domain dynamic feature set, calculates the cross-weighted shift value, constructs the effect surface, analyzes the stability, volatility, and gradient direction of the surface, and generates the hand function rehabilitation effect assessment results. The phase optimization module, based on the hand function rehabilitation effect assessment results, locates the areas of abnormal fluctuations in the effect surface and performs smoothing and error correction, analyzes the trend stability of the corrected surface, and generates optimized rehabilitation phase evaluation results.

[0024] The behavioral feature set includes hand movement execution rate characteristics and task duration characteristics; the ability feature set includes principal components of finger flexion and extension angles, principal components of muscle group signal synchronization, and principal component load distribution characteristics; the opportunity domain dynamic feature set includes external response delay change rate, intervention frequency, and task duration comparison; the hand function rehabilitation effect assessment results include effect surface stability, surface variability, and surface gradient direction; and the optimized rehabilitation stage evaluation results include abnormal fluctuation area correction and abnormal surface variability.

[0025] Specifically, such as Figure 2 , 3 As shown, the behavior analysis module includes: The data acquisition submodule collects the trajectory coordinate sequence of the patient's hand during rehabilitation training and the task execution record and performs synchronous correction, calculates the displacement change between adjacent trajectory points and performs smoothing, organizes the displacement change data in each training task stage, and generates a hand trajectory dataset. First, a 3D motion capture sensor deployed on the rehabilitation training equipment continuously records the spatial position of the patient's hand while performing rehabilitation tasks at a fixed sampling frequency of 100 Hz. Each raw record contains a timestamp accurate to milliseconds and the X, Y, and Z coordinates of the hand at that moment. When the patient performs a designated "cup translation" rehabilitation task, the task is assigned a unique number T001, and the task control system simultaneously records the start timestamp as 1.50 seconds and the end timestamp as 5.80 seconds. To accurately match the collected continuous trajectory data with the specific task execution process, the system performs synchronous correction processing. The processing iterates through all collected raw trajectory coordinate points, comparing the timestamp of each point with the start and end time range of task T001 (1.50 seconds to 5.80 seconds). Coordinate points whose timestamps fall within this range are selected and arranged in chronological order to form a trajectory coordinate sequence unique to task T001. Next, the module calculates the spatial displacement change between two adjacent trajectory points based on this sequence. This calculation is performed point-by-point, specifically calculating the straight-line distance between two points in three-dimensional space. This involves taking the square root of the sum of the squares of the differences in the X, Y, and Z coordinates of the two points. For example, if the coordinates of the first two points P1 and P2 in the sequence are (15.1, 30.2, 8.5) cm and (15.3, 30.7, 8.6) cm respectively, the calculated displacement change between them is 0.548 cm. This calculation is applied to all adjacent point pairs in the sequence, generating an original displacement change sequence. Considering the slight random fluctuations in the original sensor data, a data smoothing operation is then performed on this displacement change sequence. The operation uses a moving average method with a window size of 3. That is, for the third point in the sequence and all subsequent points, the smoothed new value is obtained by adding the original value of the point itself and the original values ​​of its two preceding points, and then dividing by 3. Finally, this processed smoothed displacement change data sequence, along with its corresponding task number T001 and task start and end time information, is integrated to form a structured record and stored in the hand trajectory dataset.

[0026] The rate calculation submodule calculates the instantaneous rate between adjacent trajectory points based on the hand trajectory dataset, then calculates the action execution rate for multiple time periods within the task, extracts the start and end times of the task to calculate the task duration, normalizes the action execution rate, generates a normalized action rate sequence, and records the task duration. Upon receiving a task record from the hand trajectory dataset, such as the record for task T001, rate-related calculations begin. The record contains the task's start and end timestamps and a smoothed displacement change sequence. First, the instantaneous rate between adjacent trajectory points is calculated. Since the data acquisition time interval is a fixed 0.01 seconds (determined by a 100 Hz sampling frequency), the instantaneous rate of each point is obtained by dividing its corresponding smoothed displacement change value by 0.01 seconds. For example, a data point with a smoothed displacement value of 0.585 cm has an instantaneous rate of 58.5 cm / s. This process iterates through all smoothed displacement data points within the T001 task cycle, generating a complete instantaneous rate sequence. Subsequently, the overall action execution rate of the task is calculated based on this instantaneous rate sequence. This calculation is performed by taking the arithmetic mean of all instantaneous rate values ​​within the T001 task cycle. If 430 instantaneous rate values ​​are generated during the duration of task T001, and the sum of these values ​​is 9890 cm / s, then the calculated action execution rate is 23.0 cm / s. Meanwhile, the start time (1.50 seconds) and end time (5.80 seconds) of the task were extracted from the task record. Subtracting these two times yielded a task duration of 4.30 seconds. Next, the action execution rate and task duration were normalized. The maximum and minimum values ​​used for normalization were established through a large-scale experimental test with a control group of 50 healthy adults. In this experiment, the effective range for action execution rate was determined to be 5.0 cm / s to 85.0 cm / s, and the effective range for task duration was 1.0 second to 15.0 seconds. The normalization process for the T001 data was as follows: subtract the minimum value of the corresponding range from the measured value, and then divide by the difference between the maximum and minimum values ​​of that range. Specifically, the normalized value for rate was (23.0 - 5.0) / (85.0 - 5.0) = 0.225; the normalized value for duration was (4.30 - 1.0) / (15.0 - 1.0) = 0.236. The same calculation and normalization process is performed on each task completed by the patient. The (normalized duration, normalized rate) data pairs generated by all tasks are combined into an action normalized rate sequence according to the order in which the tasks are executed.

[0027] The feature fitting submodule calls the action normalized rate sequence, takes the task duration as the input variable and the rate normalization value as the output variable, calculates the correlation coefficient and constructs the trend distribution curve, and combines the results with the task number to cluster and integrate the results to generate a behavioral feature set. The generated normalized rate sequence of actions is invoked and subjected to in-depth analysis. This sequence consists of a series of data pairs, each containing the normalized duration of a task and its corresponding normalized rate value. In this embodiment, the sequence generated from five training tasks (T001 to T005) completed consecutively by the patient is used as the processing object, where the normalized task duration is defined as the input variable and the normalized rate value is defined as the output variable. First, the Pearson correlation coefficient between these two variables is calculated. The calculation process involves the following steps: calculating the arithmetic mean of the five sets of normalized duration data and the five sets of normalized rate data respectively; then, for each task, calculating the deviation of the duration and rate value from the corresponding mean; then, multiplying the duration deviation and rate deviation for each task, and summing these five products; finally, dividing this sum by the square root of the product of the sums of squared deviations of the two sets of data to obtain the final correlation coefficient value, for example, 0.995. After calculating the correlation coefficient, the trend distribution curve between these two variables is constructed. This process is achieved by establishing a univariate linear regression equation. Specifically, the slope and intercept of the linear equation that best fits these five data points were calculated using the least squares method. The slope was calculated to be 1.016, and the intercept to be 0.007. This yielded a linear trend curve describing the normalization rate as a function of the normalization duration. The final step was to cluster and integrate the analysis results. This clustering was based on the calculated correlation coefficient values. Four cluster categories were preset, and the thresholds for these categories were determined through statistical analysis of thousands of historical rehabilitation data. These four categories are defined as follows: correlation coefficients greater than 0.8 are defined as "high positive correlation," those between 0.5 and 0.8 are defined as "moderate positive correlation," those between -0.5 and 0.5 are defined as "weak correlation or no correlation," and those less than -0.5 are defined as "negative correlation." Since the correlation coefficient calculated in this study was 0.995, which is greater than 0.8, the analysis results of the task group consisting of tasks T001 to T005 were classified into the "high positive correlation" category. Finally, the module generates a set of behavioral features, which includes the task group number, the calculated correlation coefficient of 0.995, the trend curve equation, and the complete information of the clustering result "high positive correlation".

[0028] Specifically, such as Figure 2 , 4 As shown, the capability extraction module includes: The signal acquisition submodule, based on the behavioral feature set, collects the flexion and extension angles of the patient's fingers, records multi-joint angle data according to the sampling time points within the training cycle, calculates the angle difference between adjacent frames and filters out outliers, integrates and sums the continuous angle differences in the time series, and generates a finger angle deviation sequence. After receiving the generated behavioral feature set, signal acquisition of the patient's fine hand movements is initiated for the corresponding task group (T001 to T005). Using a data glove worn on the patient's hand, the flexion and extension angles of the proximal interphalangeal joint (PIP) and metacarpophalangeal joint (MCP) of the index finger are simultaneously acquired at a frequency of 50Hz. Based on each sampling time point within the training cycle (e.g., 1.50 seconds to 5.80 seconds in task T001), the angle data of these two joints are recorded, forming a multi-joint angle time series. For example, at time point 2.00, the acquired MCP angle is 35.2 degrees and the PIP angle is 48.5 degrees; at the next time point 2.02, the acquired MCP angle is 36.1 degrees and the PIP angle is 49.0 degrees. Subsequently, the angle difference between adjacent data frames is calculated. For the MCP joint, the angle difference between 2.02 seconds and 2.00 seconds is... The calculation is performed across the entire time series, generating an angle difference sequence for each joint. Next, outlier filtering is performed on the generated angle difference sequence. The criterion for outlier judgment is a preset angle change rate threshold. This threshold is based on experimental data from 50 healthy subjects performing rapid finger flexion and extension movements. The experiment recorded the maximum angle change between adjacent frames within the physiological limits at a sampling rate of 50Hz. Statistical results showed that 99.5% of the angle difference data were within 4 degrees every 20 milliseconds; therefore, this value was set as the outlier judgment threshold. When processing the angle difference sequence, if the absolute value of the angle difference at a point, such as -5.2 degrees, is found to be greater than 4 degrees, that point is judged as an outlier. After outlier removal, the value of the previous valid data point (e.g., 0.9 degrees) is used to fill the gap. After outlier processing, the continuous angle differences in the time series are integrated and summed. This process starts from the task start point, accumulating the processed angle differences of each frame one by one to generate a continuously changing cumulative angle value. Taking the MCP joint as an example, if the first four effective angle differences are [0.8, 0.9, 0.7, 1.1] degrees, then the integral summation sequence is [0.8, 1.7, 2.4, 3.5] degrees. Each value in this sequence represents the total change in the joint angle from the start of the task to the current moment. This operation is performed on both the MCP and PIP joints, ultimately generating two parallel interfinite finger angle deviation sequences.

[0029] The electromyography synchronization submodule, based on the finger angle deviation sequence, synchronously acquires and registers electromyography signals on the surface of the hand, calculates the average potential of the muscle group in the corresponding interval and analyzes the electromyography amplitude sequence, calls the angle deviation value and the electromyography amplitude value to perform correlation calculation, and generates the angle electromyography synchronization sequence. Based on the obtained finger angle deviation sequence (including MCP and PIP sequences), simultaneous acquisition and registration of electromyographic (EMG) signals from the hand surface were initiated. EMG signals were acquired at a sampling frequency of 1000 Hz using electrode pads placed on the surfaces of the patient's forearm flexors (e.g., flexor digitorum superficialis) and extensors (e.g., extensor digitorum commonis). To align the EMG signals with the angle deviation sequence in time, data registration was performed. Since the sampling rate of the EMG signals (1000 Hz) is 20 times that of the angle data (50 Hz), the registration process used the timestamps of the angle data as a reference. For each angle data point's timestamp, the closest EMG signal timestamp was selected, and 20 EMG data points (i.e., 20 milliseconds) were traced back from this timestamp, forming a data window. Next, the average potential of the muscle groups within this data window was calculated. The root mean square (RMS) value of the EMG signal was used for the average potential calculation. The specific calculation is as follows: First, square each of the 20 electromyographic signal samples (unit: millivolts) within the window; then calculate the average of these 20 squared values; finally, take the square root of the average. For example, if the sum of squares of the electromyographic signals within a 20-millisecond window is 0.081 (mV)², then its root mean square value is... Millivolts. This calculation is repeated for both flexor and extensor electromyographic (EMG) signal channels, as well as at all time points within the task cycle, generating two EMG amplitude sequences. At this point, four time-synchronized sequences are obtained: MCP angle deviation sequence, PIP angle deviation sequence, flexor EMG amplitude sequence, and extensor EMG amplitude sequence. Subsequently, correlation calculations are performed between the angle deviation values ​​and EMG amplitude values. This calculation is performed separately in the flexion and extension dimensions: the angle deviation values ​​of MCP and PIP are added to represent the overall flexion / extension trend; the EMG amplitude values ​​of flexors and extensors are processed separately. Taking flexion as an example, the angle deviation values ​​of MCP and PIP are paired with the flexor EMG amplitude values ​​to form a data pair sequence, and then the Pearson correlation coefficient between these two variables is calculated. The calculation process is consistent with the correlation coefficient calculation method in the aforementioned behavioral feature set generation. For task T001, the correlation coefficient between the total angle deviation and flexor EMG amplitude is calculated to be 0.891, and the correlation coefficient with extensor EMG amplitude is -0.753. These correlation values, along with the four original synchronization sequences, were integrated to generate the angular electromyography synchronization sequence.

[0030] Table 1: Input Table for Synchronization Signal Data and Correlation Calculation 2.00 83.7 0.061 0.015 2.02 85.2 0.064 0.014 2.04 86.8 0.069 0.013 2.06 88.5 0.075 0.012 2.08 90.1 0.082 0.012 As shown in Table 1, this table lists some of the data after synchronization and preprocessing. This data will be used to calculate the correlation between angle and electromyographic signal, which is the basis for generating angle electromyographic synchronization sequences.

[0031] The principal component calculation submodule calculates the covariance between multiple signal channels based on the angle electromyography synchronization sequence, performs feature decomposition, and inputs it into principal component analysis to extract principal components. It analyzes the weight ratio of multiple components in the original signal, selects key signal contribution factors based on the load vector, and generates a capability feature set. Based on the generated angular electromyography (EMG) synchronization sequence, the parallel multi-channel signal data is extracted for analysis. In this embodiment, three key signals are selected: total angular deviation (representing kinematic output), flexor EMG amplitude (representing agonist activation), and extensor EMG amplitude (representing antagonist activation). First, the covariance matrix among these three signal channels is calculated. Covariance is used to measure the overall error between two variables, and it is calculated by taking the deviations of the two signal sequences at corresponding time points, multiplying them, and then taking the mean. Taking the total angular deviation sequence and the flexor EMG amplitude sequence as examples, their means are set to 55.0 degrees and 0.045 millivolts, respectively. At a certain time point, their instantaneous values ​​are 83.7 degrees and 0.061 millivolts, respectively, then the product of the deviations at that time point is... The covariance of the two signals is obtained by averaging the products of the deviations at all time points. A 3x3 covariance matrix is ​​constructed by calculating the covariance of all signal pairs (3 pairs in total) and the variance of each signal itself (i.e., its covariance with itself). Next, eigenvalue decomposition is performed on this covariance matrix. Eigenvalue decomposition decomposes the covariance matrix into a set of eigenvalues ​​and corresponding eigenvectors. Each eigenvector represents a new coordinate axis direction (i.e., a principal component), and its corresponding eigenvalue represents the variance of the original signal data projected onto this new coordinate axis direction. The results of eigenvalue decomposition are input into principal component analysis to extract principal components. The principal components are sorted according to the magnitude of their corresponding eigenvalues, with the largest eigenvalue being principal component one (PC1), followed by principal component two (PC2), and so on. Subsequently, the weight ratio of each component in the original signal is analyzed. This weight ratio is determined by calculating the percentage of each eigenvalue relative to the sum of all eigenvalues. For example, if the calculated three eigenvalues ​​are 2.85, 0.12, and 0.03, the sum is 3.00. Therefore, the weight ratio of principal component 1 is... Principal component two is Principal component three is Key signal contribution factors were selected based on the load vector (i.e., feature vector). Each element in the load vector represents the contribution of the original signal (such as total angle deviation) to the principal component. A load threshold of 0.5 was set; when the absolute value of an element in the load vector was greater than 0.5, its corresponding original signal was considered a key contribution factor of the principal component. The load vector of principal component one was set to [0.72, 0.68, -0.15], with elements corresponding to total angle deviation, flexor electromyographic amplitude, and extensor electromyographic amplitude, respectively. Since the absolute values ​​of the first two elements (0.72 and 0.68) were both greater than 0.5, total angle deviation and flexor electromyographic amplitude were identified as key signal contribution factors constituting principal component one. The results showed that the most significant variation (95%) in the patient's rehabilitation movement pattern was driven by the synergistic changes in flexion-extension angle and flexor activation. Finally, the weight ratios of all principal components and the key signal contribution factors of each principal component were integrated to generate an ability feature set.

[0032] Specifically, such as Figure 2 , 5 As shown, the opportunity assessment module includes: The signal monitoring submodule monitors the trigger signal generated by the external rehabilitation assistive device when it is triggered and the response signal recorded by the patient's hand sensor. It calculates the time interval between each set of trigger and response signals, filters out recorded values ​​with abnormal time intervals, calculates the mean value after filtering, and generates an external response delay sequence. During rehabilitation training, this module continuously monitors trigger signals emitted by external rehabilitation aids (such as a device that provides vibration cues at the patient's wrist) and reaction signals recorded by sensors on a data glove worn by the patient's hand. The trigger signal is a digital pulse with a precise timestamp, for example, emitted at 15 minutes 30.500 seconds. The reaction signal is defined as the moment when the patient's proximal interphalangeal joint (PIP) flexion angle first exceeds 10 degrees after receiving the trigger. This angle threshold is determined based on baseline data of healthy individuals performing rapid grasping movements under no-weight-bearing conditions; 10 degrees is considered a clear and unambiguous initiation marker. In this case, if the data glove records a PIP joint angle of 10.1 degrees at 15 minutes 30.885 seconds after the trigger, this moment is recorded as the timestamp of the reaction signal. The time interval between the two is calculated accordingly. The time interval is 150 milliseconds, or 385 milliseconds. This calculation is repeated for each pair of trigger and response signals throughout the training task (e.g., a 5-minute session), forming a raw list of recorded time intervals. Outliers are then filtered out from this list. The filtering is based on a predefined effective time window, ranging from 150 to 500 milliseconds. This range is based on standard neuroscience research on simple tactile response times, which shows that for healthy adults, responses below 150 milliseconds are typically anticipatory or accidental, while responses above 500 milliseconds indicate inattention or ineffective response. This benchmark was validated in a controlled experiment involving 100 healthy adults, where data showed that 98% of effective response times fell within this range. Therefore, when processing the raw list of time intervals, any recorded values ​​less than 150 milliseconds or greater than 500 milliseconds, such as 120 milliseconds or 610 milliseconds, are identified and removed. After filtering out all outliers, the arithmetic mean of the remaining effective recorded values ​​is calculated. Within a 30-second evaluation period, 10 valid time intervals were recorded, with values ​​of [385, 401, 392, 410, 378, 399, 405, 388, 393, 394] milliseconds, totaling 3945 milliseconds. The average time interval for this period is then calculated as follows: Milliseconds. This average value is used as the representative latency data for this evaluation period. The above calculation process is repeated in consecutive 30-second periods to generate a time series consisting of the average latency values ​​of multiple periods, i.e., the external response latency series.

[0033] The frequency extraction submodule, based on the external response delay sequence, counts the number of times the external rehabilitation assistive device is triggered and the number of times the patient responds within a set period, calculates the ratio between the two and normalizes it as the time-triggered response frequency, analyzes the frequency change trend, and generates intervention frequency distribution results. Analysis is performed based on the generated external response delay sequence and raw data. First, trigger and response events during the rehabilitation process are statistically analyzed in predefined cycles (e.g., 60 seconds). Within a 60-second cycle, the total number of triggers issued by the external rehabilitation assistive device is counted. For example, in the first minute of training, the device issued 25 vibration trigger signals. Simultaneously, the number of effective responses generated by the patient is counted. The criteria for a valid response are the same as before: the response time interval must fall within a valid window of 150 to 500 milliseconds. Of these 25 triggers, 22 produced responses within this window, while the other 3 (1 no response, 2 responses exceeding 500 milliseconds) were considered invalid responses. Next, the ratio between the number of triggers and the number of effective responses is calculated. In this example, this ratio is [value missing]. This ratio is directly used as the response frequency within that time period because it is a normalized value between 0 and 1, where 1 represents an effective response to all triggers. The advantage of this approach is that, by calculating a simple ratio, the consistency of a patient's response to external interventions can be directly quantified over a specific time period. Subsequently, the module analyzes the response frequency over multiple consecutive periods to observe its trends. This analysis is performed by linearly fitting the response frequency values ​​for consecutive time periods (e.g., five consecutive one-minute periods).

[0034] Table 2: Response Frequency Statistics and Trend Analysis First minute 25 22 0.88 Second minute 25 23 0.92 Third minute 25 23 0.92 Fourth minute 25 24 0.96 Fifth minute 25 24 0.96 As shown in Table 2, this table records response frequency data over a continuous 5-minute period. Based on this data, a trend line can be calculated. By calculating the time series slope of these five frequency values, a positive slope (e.g., 0.018) indicates that the patient's response frequency is increasing over time; a negative slope indicates a decreasing trend; and a slope close to zero indicates stability. Finally, the response frequency values ​​for each period and the calculated overall trend (e.g., denoted as "increasing trend, slope 0.018") are combined to generate the intervention frequency distribution results.

[0035] The feature generation submodule calls the intervention frequency distribution results and combines them with the capability feature set. It calculates the delay change rate within a continuous period based on the time series of external response delay, compares the corresponding ratio of the delay change rate with the task duration, extracts time-related change parameters, and generates a dynamic feature set of the opportunity domain. The generated intervention frequency distribution results were retrieved and combined with an earlier generated capability feature set describing the patient's intrinsic motor coordination patterns. The capability feature set indicated that the patient's main motor variability (95%) was driven by two key signal contributing factors: "total angular deviation" and "flexor electromyographic amplitude." Then, based on the time series of external response delays (e.g., a sequence consisting of average delay values ​​[394.5, 380.1, 375.5, 368.9, 371.2] milliseconds across five consecutive 30-second cycles), the rate of change in delay within consecutive time periods was calculated. The rate of change in delay was calculated as: the average delay of the later time period minus the average delay of the previous time period, divided by the duration of the time period (30 seconds). Taking the second time period as an example, the rate of change in delay was... Milliseconds per second. This negative value indicates that the patient's reaction speed is increasing within the time period. This calculation is applied to the entire delay sequence, generating a delay change rate sequence. Next, the corresponding proportion of the delay change rate to the task duration is compared. Here, "task duration" refers to the length of the time period that generates the delay change rate, i.e., 30 seconds. The corresponding proportion is calculated as: the total delay change within the time period (e.g., -14.4 milliseconds for the second time period) divided by the total duration of that time period (30,000 milliseconds). Therefore, the proportion for the second time period is... This dimensionless parameter represents the magnitude of reaction time change per unit time and is extracted as a key time-related variation parameter. This calculation is performed on all consecutive time periods, forming a sequence of time-related variation parameters. Finally, the key signal contributing factors (total angle deviation, flexor electromyography amplitude), intervention frequency distribution results (response frequency sequence and trend), external response delay sequence, and the newly calculated sequence of time-related variation parameters are integrated. These multi-dimensional time series data are combined into a structured dataset that comprehensively describes the dynamic evolution of patients' intrinsic physiological responses and extrinsic behavioral performance in the face of external interventions, generating a dynamic feature set of the opportunity domain.

[0036] Specifically, such as Figure 2 , 6 As shown, the effect analysis module includes: The weight normalization submodule obtains the behavioral feature set, ability feature set and opportunity domain dynamic feature set for sampling and statistics, calculates the weight ratio coefficient of multiple feature dimensions under a unified scale, performs linear normalization on the weight ratio coefficient, and generates a set of feature weight normalization values. The generated behavioral feature set, capability feature set, and opportunity domain dynamic feature set are obtained, and the core quantitative indicators in these three sets are sampled and statistically analyzed. From the behavioral feature set, the Pearson correlation coefficient representing the consistency between task planning and execution is extracted, with a value of 0.995. From the capability feature set, the weight ratio of Principal Component 1 (PC1), representing the core neuromuscular synergy pattern, is extracted, with a value of 95%, or 0.950. From the opportunity domain dynamic feature set, the average response frequency representing the stability of responses to external interventions is extracted, with a value of 0.928 (obtained by calculating the arithmetic mean of the response frequencies [0.88, 0.92, 0.92, 0.96, 0.96] over the aforementioned five periods). These three sampled indicator values ​​[0.995, 0.950, 0.928] constitute the original feature weight vector. Next, the weight ratio coefficients of multiple feature dimensions at a uniform scale are calculated. This process is achieved by dividing each original weight value by the sum of all original weight values. First, the sum is calculated: Then, the weight ratio coefficient for each feature is calculated: the weight ratio coefficient for the behavioral feature is... The weighting coefficient for ability characteristics is: The weighting coefficients for the dynamic features of the opportunity domain are: Linear normalization is then performed on these calculated weight ratio coefficients. Since the sum of the results from the previous step is already 1, it already satisfies the normalization requirement and constitutes the relative importance of each feature in the comprehensive evaluation system. The advantage of this approach is that by jointly normalizing feature indicators from different sources and with different physical meanings, a unified quantitative comparison framework is constructed. Finally, the module integrates this set of normalized coefficient values ​​[0.3463, 0.3307, 0.3230] to generate a set of normalized feature weight values.

[0037] The cross-weight shift submodule extracts the corresponding weight values ​​of behavioral features and ability features based on the normalized set of feature weights, calculates the weight deviation between the two and compares it with the rate of change in the dynamic features of the opportunity domain, analyzes and summarizes the cross-coupling between features, analyzes the average weight shift index, and generates a cross-weight shift value set. After receiving the normalized feature weight set [0.3463, 0.3307, 0.3230], an in-depth analysis of the intrinsic relationships between the features begins. First, the behavioral feature weight value (0.3463) representing high-level motion planning and the capability feature weight value (0.3307) representing low-level neuromuscular execution are extracted from the value set. Then, the weight deviation between these two is calculated, specifically the difference between the behavioral feature weight value and the capability feature weight value. This positive bias indicates that, within this assessment period, the patient's motor planning consistency (behavioral characteristic) carries a slightly higher weight in the overall assessment than their neuromuscular synergy efficiency (ability characteristic). Next, this weighting bias is compared to the rate of change of delay in the opportunity domain dynamic feature set. From the opportunity domain dynamic feature set, the previously calculated sequence of rates of change of delay for four consecutive time periods (in milliseconds / second) is extracted: [-0.48, -0.153, -0.22, 0.077]. The average of this sequence is calculated to obtain the average rate of change of delay. Milliseconds per second. This negative value represents the overall trend of improvement in patient reaction speed over the observation period. The comparison process is accomplished by calculating the product of the weighted bias and the absolute value of the average rate of change of delay, to quantify the strength of the cross-coupling between the two: This product is defined as the cross-coupling amount. This calculation is applied to each assessment period, forming a time series of cross-coupling amounts, which are then summarized. Finally, the average weighted shift index is analyzed. The average weighted shift index is defined as the arithmetic mean of the cross-coupling amounts calculated over multiple consecutive assessment periods (e.g., three consecutive recovery days). Setting the cross-coupling amounts for the three consecutive days to 0.0030264, 0.0032100, and 0.0028500, the average weighted shift index is... The cross-coupling amount calculated for each evaluation period and the final average weight shift index are integrated to generate a cross-weight shift value set.

[0038] The surface assessment submodule calls the cross-weighted shift value set, maps multiple cross-weighted shift values ​​to a three-dimensional coordinate plane and associates them with the time series to construct an effect surface, calculates the surface volatility and extracts the surface gradient direction, and integrates the numerical values ​​with the overall surface stability index to generate the hand function rehabilitation effect assessment result. The cross-weighted shift value set is invoked, which contains a series of cross-coupling values ​​arranged in chronological order. In this embodiment, the cross-coupling values ​​[0.0030264, 0.0032100, 0.0028500] calculated for three consecutive rehabilitation days are used as input. These cross-weighted shift values ​​are mapped to a two-dimensional coordinate system with time as the X-axis and cross-weighted shift values ​​as the Z-axis to construct an effect curve. This curve visually demonstrates the dynamic evolution of the patient's internal characteristic coupling state as the rehabilitation process progresses. Next, the volatility of this curve is calculated. The volatility is calculated using the standard deviation of the data sequence. First, the mean is calculated to be 0.0030288, and then the standard deviation is calculated to be 0.001469. This standard deviation value is the surface (curve in this example) volatility; a smaller value indicates a stable rehabilitation state. Subsequently, the surface gradient direction is extracted. The gradient direction is determined by calculating the slope between the first and last points on the effect curve. This negative slope indicates a slight decreasing trend in cross-coupling over time. Finally, the overall surface stability index is used for numerical integration. The overall surface stability index is defined as a comprehensive score, calculated as follows: The normalized gradient is the result of mapping the calculated gradient value to an interval. Its baseline range was determined through statistical analysis of historical data from 200 recovered patients, and the normal fluctuation range of the gradient was set to -0.001 to 0.001. In this example, the normalized value of the gradient -0.0000882 is... Substitute the values ​​into the equation and integrate them: The integrated results of these values ​​were used for the final evaluation. Based on a pre-defined evaluation interval (also based on the statistical distribution of historical data from 200 patients: greater than 4000 for "high stable recovery," 2000 to 4000 for "moderate stable recovery," and less than 2000 for "low stable recovery"), the final score of 4084.63 was classified as "high stable recovery." This result indicates that the patient's hand function rehabilitation process exhibited high stability and a positive recovery trend. The score of 4084.63 and its corresponding evaluation category "high stable recovery" were combined to generate the hand function rehabilitation effect evaluation result.

[0039] Specifically, such as Figure 2 ,7 As shown, the stage optimization module includes: The fluctuation detection submodule extracts the response value sequence at multiple times from the effect surface based on the hand function rehabilitation effect assessment results, calculates the rate of change of response at adjacent times and counts the difference in rate of change, filters out regions where the difference in rate of change exceeds the fluctuation threshold, and generates a set of abnormal fluctuation intervals on the surface. After obtaining the hand function rehabilitation effect assessment results, a sequence of response values ​​at multiple time points is extracted from the effect surface. In this embodiment, the response value sequence consists of cross-weighted shift values ​​for 12 consecutive rehabilitation days, specifically: [0.00301, 0.00322, 0.00295, 0.00334, 0.00351, 0.00253, 0.00264, 0.00362, 0.00371, 0.00380, 0.00315, 0.00324]. First, the rate of change of response between adjacent time points is calculated, i.e., the daily change. For example, the rate of change on day 2 relative to day 1 is... This calculation is performed point-by-point along the time series, generating a response rate of change sequence containing 11 values: [0.00021, -0.00027, 0.00039, 0.00017, -0.00098, 0.00011, 0.00098, 0.00009, 0.00009, -0.00065, 0.00009]. Next, this rate of change sequence is statistically analyzed, calculating the difference between adjacent rates of change. For example, the difference in the rate of change on day 3 is the rate of change on day 3 minus the rate of change on day 2, i.e. This process generates a sequence of rate-of-change difference values ​​containing 10 values: [-0.00048, 0.00066, -0.00022, -0.00115, 0.00109, 0.00087, -0.00089, 0.00000, -0.00074, 0.00074]. Subsequently, the rate-of-change difference value sequence is filtered according to a preset fluctuation threshold to identify areas of significant fluctuation. This fluctuation threshold is set based on historical data analysis of 200 patients who have completed rehabilitation with good results during the later stages of rehabilitation. The daily cross-weighted shift values ​​of these 200 patients are processed to calculate their rate-of-change difference values, forming a large-scale reference dataset. Statistical distribution analysis of this dataset determines its 95th percentile, which is 0.00090. Therefore, the fluctuation threshold is set to 0.00090. Any point where the absolute value of the difference in rate of change exceeds this threshold is identified as an abnormal fluctuation point. During the screening of the difference value sequence in this embodiment, it was found that the absolute values ​​of the 5th value (-0.00115) and the 6th value (0.00109) both exceeded 0.00090. These two outliers correspond to the changes from day 5 to day 6 and from day 6 to day 7 in the original data, respectively. Therefore, the continuous segment containing these time points, namely day 5, day 6, and day 7, is marked as an abnormal fluctuation interval. All identified intervals of this type are summarized to generate a set of abnormal fluctuation intervals on the surface.

[0040] The surface correction submodule calls the abnormal fluctuation interval set of the surface, calculates the mean of the time point response for the time point response sequence within multiple intervals, performs adjacent interpolation, recalculates the continuous change sequence of the interval time point response, adjusts the deviation according to the time point smoothness coefficient, and generates a set of surface smoothing correction values. After receiving the set of abnormal fluctuation intervals on the surface, the time point response sequences [0.00351, 0.00253, 0.00264] within the identified interval (days 5 to 7) are corrected. First, the arithmetic mean of the response values ​​at all time points within this abnormal interval is calculated: Next, adjacent interpolation is performed on the points within the interval. Specifically, for the point with the most drastic fluctuation within the interval (i.e., the response value of 0.00253 on day 6), linear interpolation is performed using two adjacent points in the original sequence located on the boundary of the abnormal interval (0.00351 on day 5 and 0.00264 on day 7) to calculate a replacement value: This interpolation provides a smoothed estimate based on local trends. After obtaining the interpolation, the bias is adjusted according to a smoothness coefficient at a given time point. The coefficient is set in relation to the severity of the anomalous fluctuations. Severity is defined as the ratio of the absolute value of the largest rate of change difference (0.00115) within the anomalous segment to the fluctuation threshold (0.00090), i.e. The formula for calculating the smoothness coefficient is as follows: In this example, it is This coefficient determines the extent to which the correction value adopts the interpolation result and the interval mean. The adjustment process involves taking a weighted average of the interpolation (0.003075) and the interval mean (0.00289), with weights of (1 - smoothness coefficient) and smoothness coefficient, respectively. The corrected response value on day 6 is... After correcting the outliers, the corrected values ​​replace the corresponding values ​​in the original sequence, and the continuous change sequence within this interval is recalculated. After correction, the sequence from day 5 to day 7 becomes [0.00351, 0.003035, 0.00264], with continuous change values ​​of [-0.000475, -0.000395]. Finally, the complete time series containing these corrected values ​​is integrated to generate a set of surface smoothing correction values, with the following values: [0.00301, 0.00322, 0.00295, 0.00334, 0.00351, 0.003035, 0.00264, 0.00362, 0.00371, 0.00380, 0.00315, 0.00324].

[0041] The trend analysis submodule extracts the response change sequence under the time series based on the surface smoothing correction value set, calculates the rate of change of adjacent time periods and calculates the variance of the rate of change, calculates the stability coefficient based on the trend of variance change, and generates the optimized rehabilitation stage evaluation results. Based on the generated set of surface smoothing correction values ​​[0.00301, 0.00322, 0.00295, 0.00334, 0.00351, 0.003035, 0.00264, 0.00362, 0.00371, 0.00380, 0.00315, 0.00324], the final trend analysis is initiated. First, the response change sequence under this time series is extracted, that is, the daily rate of change is recalculated. The calculation result is a sequence containing 11 values: [0.00021, -0.00027, 0.00039, 0.00017, -0.000475, -0.000395, 0.00098, 0.00009, 0.00009, -0.00065, 0.00009]. Next, the variance of this rate of change sequence is calculated. The calculation process is as follows: first, the arithmetic mean of the sequence is obtained. Then, the square of the difference between each rate of change value and the mean is calculated, and all these squared differences are summed to obtain a total sum of squares of 0.000001642. Finally, the total sum of squares is divided by the number of data points, 11, to obtain the variance of the rate of change. Based on the calculated variance, the stability coefficient is further calculated. The formula for the stability coefficient is a calibrated inverse function: The calibration coefficient of 10,000,000 in this formula was determined based on the analysis of historical variance data from 200 recovered patients. This coefficient transforms typical variance values ​​(on the order of 1e-7) to around 1.0, making the stability coefficient fall within an intuitive range of 0 to 1000. Substituting the variance calculated in this embodiment into the formula: The advantage of this approach is that, through variance calculation and scaling transformation, the dynamic stability of the entire rehabilitation process is quantified into a single, easily comparable indicator. Finally, the calculated stability coefficient of 401.6 is compared with a preset evaluation interval. This evaluation system is also based on statistical analysis of the historical stability coefficients of 200 patients: greater than 750 indicates "stable recovery trend," 400 to 750 indicates "basically stable recovery trend," and less than 400 indicates "fluctuating recovery trend." In this embodiment, the score of 401.6 falls within the "basically stable recovery trend" interval. This result indicates that, after data smoothing correction, the overall rehabilitation process of the patients shows a basically stable trend. Integrating the stability coefficient value of 401.6 with the evaluation level of "basically stable recovery trend" generates an optimized rehabilitation stage evaluation result.

[0042] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A system for evaluating the rehabilitation effect of hand function in rheumatoid arthritis patients based on the BCW theory, characterized in that, The system includes: The behavior analysis module collects the patient's hand trajectory and training task records during rehabilitation training, extracts the hand movement execution rate and task duration, normalizes them, performs correlation fitting, generates a behavior feature set, and transmits it to the ability extraction module. The ability extraction module, based on the behavioral feature set, collects the interphalangeal flexion and extension angles and surface electromyography signals of the patient's hand, analyzes the angle deviation and muscle group signal synchronization sequence, inputs them into principal component analysis for principal component extraction, calculates the principal component load, generates the ability feature set, and transmits it to the opportunity assessment module. The opportunity assessment module monitors the time difference between the signal triggering of the external rehabilitation assistive device and the patient's response, extracts the external response delay and intervention frequency, calculates the rate of change of the external response delay in combination with the ability feature set, compares it with the task duration, generates a dynamic feature set of the opportunity domain, and transmits it to the effect analysis module. The effect analysis module performs weight normalization on the behavioral feature set, ability feature set and opportunity domain dynamic feature set, calculates the cross-weighted shift value and constructs the effect surface, analyzes the stability, volatility and gradient direction of the surface, and generates the hand function rehabilitation effect evaluation result. The effect analysis module includes: The weight normalization submodule obtains the behavioral feature set, capability feature set and opportunity domain dynamic feature set for sampling and statistics, calculates the weight ratio coefficients of multiple feature dimensions under a unified scale, performs linear normalization on the weight ratio coefficients, and generates a feature weight normalization value set. The cross-weight shift submodule extracts the corresponding weight values ​​of behavioral features and ability features based on the normalized set of feature weights, calculates the weight deviation between the two and compares it with the rate of change in the dynamic features of the opportunity domain, analyzes and summarizes the cross-coupling between features, analyzes the average weight shift index, and generates a cross-weight shift value set. The surface assessment submodule calls the cross-weighted shift value set, maps multiple cross-weighted shift values ​​to a three-dimensional coordinate plane, associates them with the time series to construct an effect surface, calculates the surface volatility and extracts the surface gradient direction, and integrates the numerical values ​​with the overall surface stability index to generate a hand function rehabilitation effect assessment result.

2. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 1, characterized in that: The behavioral feature set includes hand movement execution rate features and task duration features; the ability feature set includes principal components of finger flexion and extension angles, principal components of muscle group signal synchronization, and principal component load distribution features; the opportunity domain dynamic feature set includes external response delay change rate, intervention frequency, and task duration comparison; and the hand function rehabilitation effect evaluation results include effect surface stability, surface variability, and surface gradient direction.

3. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 1, characterized in that, The behavior analysis module includes: The data acquisition submodule collects the trajectory coordinate sequence of the patient's hand during rehabilitation training and the task execution record and performs synchronous correction, calculates the displacement change between adjacent trajectory points and performs smoothing, organizes the displacement change data in each training task stage, and generates a hand trajectory dataset. The rate calculation submodule calculates the instantaneous rate between adjacent trajectory points based on the hand trajectory dataset, then calculates the action execution rate for multiple time periods within the task, extracts the start and end times of the task to calculate the task duration, normalizes the action execution rate, generates a normalized action rate sequence, and records the task duration. The feature fitting submodule calls the normalized rate sequence of the action, takes the task duration as the input variable and the rate normalization value as the output variable, calculates the correlation coefficient and constructs a trend distribution curve, and combines the results with the task number to cluster and integrate them to generate a behavioral feature set.

4. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 1, characterized in that, The capability extraction module includes: The signal acquisition submodule, based on the behavioral feature set, acquires the flexion and extension angles of the patient's fingers, records multi-joint angle data according to the sampling time points within the training cycle, calculates the angle difference between adjacent frames and filters out outliers, integrates and sums the continuous angle differences in the time series, and generates a finger angle deviation sequence. The electromyography synchronization submodule, based on the interdigital angle deviation sequence, synchronously acquires and registers electromyography signals on the surface of the hand, calculates the average potential of the muscle group in the corresponding interval and analyzes the electromyography amplitude sequence, calls the angle deviation value and the electromyography amplitude value to perform correlation calculation, and generates an angle electromyography synchronization sequence. The principal component calculation submodule calculates the covariance between multiple signal channels based on the angle electromyography synchronization sequence, performs feature decomposition, and inputs it into principal component analysis to extract principal components. It analyzes the weight ratio of multiple components in the original signal, selects key signal contribution factors based on the load vector, and generates a capability feature set.

5. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 4, characterized in that, The principal component analysis consists of principal components, eigenvalues, and loading vectors.

6. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 1, characterized in that, The opportunity assessment module includes: The signal monitoring submodule monitors the trigger signal generated by the external rehabilitation assistive device when it is triggered and the response signal recorded by the patient's hand sensor. It calculates the time interval between each set of trigger and response signals, filters out recorded values ​​with abnormal time intervals, calculates the mean value after filtering, and generates an external response delay sequence. The frequency extraction submodule, based on the external response delay sequence, counts the number of times the external rehabilitation assistive device is triggered and the number of times the patient responds within a set period, calculates the ratio between the two and normalizes it as the time-triggered response frequency, analyzes the frequency change trend, and generates intervention frequency distribution results. The feature generation submodule calls the intervention frequency distribution results and combines them with the capability feature set. Based on the time series of external response delay, it calculates the delay change rate within a continuous period, compares the corresponding ratio of the delay change rate with the task duration, extracts time-related change parameters, and generates a dynamic feature set of the opportunity domain.

7. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 1, characterized in that, The system also includes: The phase optimization module, based on the hand function rehabilitation effect evaluation results, locates the areas of abnormal fluctuations in the effect surface and performs smoothing and error correction, analyzes the trend stability of the corrected surface, and generates optimized rehabilitation phase evaluation results. The optimized rehabilitation stage evaluation results include correction of abnormal fluctuation areas and abnormal fluctuations of curved surfaces.

8. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 7, characterized in that, The stage optimization module includes: The fluctuation detection submodule, based on the hand function rehabilitation effect evaluation results, extracts the response value sequence at multiple times in the effect surface, calculates the response change rate between adjacent times and counts the change rate difference value, filters out the region where the change rate difference value exceeds the fluctuation threshold, and generates a set of abnormal fluctuation intervals of the surface. The surface correction submodule calls the set of abnormal fluctuation intervals of the surface, calculates the mean of the time point response for the time point response sequence within multiple intervals, performs adjacent interpolation, recalculates the continuous change sequence of the interval time point response, adjusts the deviation according to the time point smoothness coefficient, and generates a set of surface smoothing correction values. The trend analysis submodule extracts the response change sequence under the time series based on the surface smoothing correction value set, calculates the rate of change of adjacent time periods and calculates the variance of the rate of change, calculates the stability coefficient based on the variance change trend, and generates the optimized rehabilitation stage evaluation results.

9. The evaluation system for hand function rehabilitation in rheumatoid arthritis patients based on BCW theory according to claim 8, characterized in that: The fluctuation threshold is set by calculating the rate of change of response at multiple time points within the time series and statistically analyzing the distribution characteristics of the difference values, with the standard deviation of the difference values ​​as the benchmark.

Images