Stroke patient stage hand movement rehabilitation evaluation method and system

Abstract

The application discloses a stroke patient stage hand movement rehabilitation evaluation method and system, relates to the field of stroke patient hand rehabilitation evaluation, and comprises the following steps: acquiring kinematics signals, mechanical signals and sEMG signals synchronously collected by a data glove system in a standardized hand function test process; preprocessing the kinematics signals, the mechanical signals and the sEMG signals; based on the preprocessing result, extracting features of each test action segment to obtain heterogeneous feature vectors of the test action segments; based on the heterogeneous feature vectors of all the test action segments, performing multi-dimensional feature fusion and time alignment to obtain a multi-dimensional hand function feature matrix; based on the multi-dimensional hand function feature matrix, performing Brunnstrom staging division to obtain a Brunnstrom staging evaluation reference result of the stroke patient as rehabilitation evaluation reference information; and the application can improve the evaluation accuracy of the hand rehabilitation condition of the stroke patient.

Description

Technical Field

[0001] This application relates to the field of hand rehabilitation evaluation technology for stroke patients, and in particular to a method and system for phased hand motor rehabilitation evaluation of stroke patients. Background Technology

[0002] Stroke is one of the leading causes of long-term disability in adults, with over 50% of stroke patients experiencing long-term upper limb and hand motor dysfunction, severely impacting their quality of life. In rehabilitation engineering practice, the recovery process of hand function is typically divided into stages based on Brunnstrom's classic six-stage theory. Each stage exhibits different neuromuscular control data characteristics in dimensions such as muscle strength, muscle tone, movement patterns, and coordination. Therefore, automated stage classification and quantitative characterization of multimodal sensor data is a core prerequisite for achieving closed-loop control in rehabilitation assessment systems. However, existing assessment methods largely rely on manual scale scoring, resulting in limited data consistency and difficulty in objectively capturing the continuous signal characteristics during the transition from commensurate movement to partially isolated movement. This leads to intervention parameter adjustments often lagging behind the patient's actual recovery level.

[0003] In recent years, wearable sensing technology has provided a data collection basis for the objectification of hand function assessment. However, in practical engineering implementation, the assessment accuracy of hand function assessment methods based on wearable sensing technology is low. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for evaluating the phased hand motor rehabilitation of stroke patients, which can improve the accuracy of evaluating the phased hand motor rehabilitation of stroke patients.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for evaluating the phased hand motor rehabilitation of stroke patients, applied to a phased hand motor rehabilitation evaluation system for stroke patients. The system includes a data glove system and a processing module. The data glove system is worn on the stroke patient's hand during standardized hand function testing to collect kinematic signals, mechanical signals, and sEMG signals from the stroke patient's hand during the standardized hand function testing process. The method for evaluating the phased hand motor rehabilitation of stroke patients includes: During the standardized hand function test, the data glove system synchronously collects kinematic signals, mechanical signals, and sEMG signals. Preprocessing is performed on the acquired kinematic signals, mechanical signals, and sEMG signals; Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each test action segment; wherein, the features include kinematic features, mechanical features and / or electromyographic features, and the heterogeneous feature vectors include kinematic heterogeneous feature vectors, mechanical heterogeneous feature vectors and / or electromyographic heterogeneous feature vectors; Based on the heterogeneous feature vectors of all test action segments, multi-dimensional feature fusion and time alignment are performed to obtain a multi-dimensional hand function feature matrix; Based on the multidimensional hand function feature matrix, Brunnstrom staging is performed to obtain Brunnstrom staging assessment reference results for stroke patients, which can be used as reference information for rehabilitation assessment.

[0006] Optionally, the data glove system includes a first inertial measurement unit, five second inertial measurement units, five flexible bending sensors, thirteen flexible pressure sensors, and a surface electromyography (EMG) acquisition component. The first inertial measurement unit is located on the back of the hand, the five second inertial measurement units are correspondingly arranged on the fingertips, the five flexible bending sensors are correspondingly arranged along the back of the fingers, and the thirteen flexible pressure sensors are correspondingly embedded in the fingertips, palmar surfaces of the metacarpophalangeal joints, ulnar surfaces of the thumb, the thenar eminence, and the hypothenar eminence. The surface EMG acquisition component includes a first bipolar surface electrode group, a second bipolar surface electrode group, and a reference... The system includes a reference electrode and two electromyography (EMG) acquisition channels for the flexor and extensor sides, respectively, to receive differential signals. A first bipolar surface electrode group is placed on the skin surface corresponding to the flexor or extensor muscles of the forearm in stroke patients, and a second bipolar surface electrode group is placed on the skin surface corresponding to the extensor or extensor muscles of the forearm in stroke patients, respectively, for detecting surface EMG signals at the corresponding sites. The two EMG acquisition channels synchronously acquire the raw surface EMG signals EMG_raw,F(t) on the flexor side and EMG_raw,E(t) on the extensor side based on the same sampling clock or unified timestamp, and share the reference electrode. The kinematic signals include triaxial angular velocity, triaxial acceleration, triaxial magnetic field and corresponding original attitude quaternions collected by the first inertial measurement unit and each second inertial measurement unit, as well as the original bending electrical signals collected by each flexible bending sensor. The raw mechanical signals include the raw pressure signals collected by each flexible pressure sensor; The raw sEMG signals include the raw surface electromyography (EMG) signals EMG_raw,F(t) on the flexor side and EMG_raw,E(t) on the extensor side, which are synchronously acquired by the flexor side EMG acquisition channel and the extensor side EMG acquisition channel during the same test action.

[0007] Optionally, the preprocessing of the acquired kinematic signals, mechanical signals, and sEMG signals specifically includes: The original attitude quaternions collected by the first inertial measurement unit and each second inertial measurement unit are initialized with attitude and aligned with the local coordinate system. For each finger, the relative attitude quaternion of the corresponding second inertial measurement unit relative to the first inertial measurement unit is calculated based on the aligned quaternion of the first inertial measurement unit and the corresponding second inertial measurement unit. Each relative posture quaternion is converted into the overall Euler angle of the corresponding finger relative to the back of the hand, and the angle component corresponding to the finger flexion and extension movement is extracted as the IMU solution angle of the corresponding finger. An angle fusion is performed on the IMU-derived angle of each finger and the bending reference angle of the corresponding flexible bending sensor based on an adaptive complementary filter to obtain the overall fused angle of each finger; wherein, the bending reference angle is obtained by mapping the original bending electrical signal of each flexible bending sensor on each finger to the bending reference angle through factory calibration or multi-point attitude calibration. Based on the finger joint angle allocation parameters pre-calibrated and fixed for the subject or device model, the overall fusion angle of each finger is allocated to the corresponding finger joint to obtain the estimated fusion angle of each finger joint; wherein, the finger joints of the 2nd to 5th fingers include the metacarpophalangeal joint, proximal interphalangeal joint, and distal interphalangeal joint, and the finger joints of the thumb include the carpometacarpophalangeal joint, metacarpophalangeal joint, and interphalangeal joint; the estimated fusion angle of each finger joint is obtained based on the overall fusion angle allocation of the corresponding finger. Zero-point drift compensation and temperature drift calibration are performed on the raw pressure signals collected by each flexible pressure sensor. Based on the sampling rate, the effective bandwidth of the surface electromyography acquisition component, and the local power frequency, the two raw sEMG signals were subjected to bandpass filtering and power frequency notch filtering using the same parameters. The attitude of the three-axis angular velocities output by the first inertial measurement unit and each of the second inertial measurement units is aligned, and the three-axis angular velocities of each finger relative to the back of the hand are calculated based on the three-axis angular velocities after transformation to a unified coordinate system. Based on the triaxial angular velocities of each finger relative to the back of the hand, the angular velocity modulus of each finger relative to the back of the hand is calculated. The multimodal raw data stream is time-stamped according to a unified sampling clock, and the time-synchronized multimodal raw data stream is output, along with the corresponding overall fusion angle, the estimated fusion angle of each finger joint, and the triaxial angular velocity and angular velocity magnitude of each finger relative to the back of the hand. The multimodal raw data stream includes preprocessed kinematic signals, mechanical signals, and sEMG signals.

[0008] Optionally, the method for evaluating the phased hand motor rehabilitation of stroke patients, before extracting features from each test movement segment based on the preprocessing results, further includes: An automatic action cycle segmentation algorithm based on fused kinematic thresholds is used to segment the action of the multimodal raw data stream: For each test action, the primary driving finger is determined; where, for test actions that require only one finger to move, the primary driving finger is the finger that needs to move when performing the test action; for test actions that require multiple fingers to move simultaneously, the primary driving finger is determined according to the provisions of the action standard protocol. The relative angular velocity modulus corresponding to the main driving finger of each test action is detected when the relative angular velocity modulus is greater than the preset starting threshold ω_start and the duration reaches the first target time. The moment when the relative angular velocity modulus first exceeds ω_start in each target time period is determined as the starting point of the corresponding action. For test actions that do not include a hold phase in the action protocol, detect the time period during which the relative angular velocity modulus corresponding to the main driving finger of the test action falls below the preset termination threshold ω_end and the duration reaches the second target time, and determine the moment when the relative angular velocity modulus first falls below ω_end in the time period as the action end point of the corresponding action. For test actions that include a hold phase in the action protocol, the moment when the relative angular velocity modulus corresponding to the main driving finger of the test action first falls below ω_end and satisfies the second target time is determined as the end point of the motion phase or the time when the target posture is reached t_arr. The hold phase is then retained, and the earliest of the test action end prompt, release start, or the end of the maximum hold window specified in the action protocol is determined as the end point of the test action. Based on the start and end points of each test action, the time-synchronized multimodal raw data stream is divided into independent data segments corresponding to each test action to obtain each test action segment.

[0009] Optionally, the test action segment includes active action segments and passive action segments. The active action segment includes active fist clenching segment, active finger extension segment, active grasping segment, and active pinching segment. The passive action segment includes passive finger extension segment. Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each test action segment, specifically including: For each active motion segment, the joint range of motion of each finger joint, the total active range of motion of the five fingers, and the smoothness of the motion trajectory are extracted. Separation motion indices are extracted for selective motion segments pre-specified by the motion standard protocol. For active fist clenching segments and active five-finger extension segments, the correlation coefficient ρ of the common motion of the five fingers is calculated to obtain the kinematic heterogeneous feature vector of each active motion segment. Among them, the finger joints of the 2nd to 5th fingers include the metacarpophalangeal joint (MCP), the proximal interphalangeal joint (PIP), and the distal interphalangeal joint (DIP), and the finger joints of the thumb include the carpometacarpophalangeal joint (CMC), the metacarpophalangeal joint (MCP), and the interphalangeal joint (IP). For each active grasp segment and each active pinch segment, the maximum muscle strength index is extracted to obtain the mechanical heterogeneous feature vector of each active grasp segment and each active pinch segment; For each passive extension segment of a finger, muscle tone quantification index is extracted to obtain the mechanical heterogeneous feature vector of each passive extension segment of a finger; For each test action segment, the degree of muscle activation, the antagonist muscle synergistic contraction rate, and the degree of muscle fatigue are extracted. For active action segments, the time difference Δt=|t_EMG_on-t_motion_on| between the onset of electromyographic activation t_EMG_on and the onset of movement t_motion_on is further extracted to obtain the corresponding electromyographic heterogeneous feature vector.

[0010] Optionally, for each active motion segment, the joint range of motion of each finger joint, the total active range of motion of the five fingers, and the smoothness of the motion trajectory are extracted, and a separation motion index is extracted for selective motion segments pre-specified by the motion standard protocol, specifically including: For each active movement segment, the range of motion of each finger joint is calculated using the following formula: ROM_(i,j)=max(θ_(i,j)(t))-min(θ_(i,j)(t)); Where i represents the finger number, i∈{1,2,3,4,5}, representing the thumb, index finger, middle finger, ring finger, and little finger respectively; j represents the joint type of finger i, ROM_(i,j) represents the joint mobility of the finger joint j of finger i; θ_(i,j)(t) represents the estimated fusion angle of the middle finger joint j of finger i, and max(θ_(i,j)(t)) and min(θ_(i,j)(t)) are both obtained within t∈[t_start,t_end], where t_start and t_end represent the start and end points of the current active action segment; For fingers 2-5, the active range of motion TAM_finger,i of finger i is calculated using the following formula: TAM_finger,i=ROM_MCP,i+ROM_PIP,i+ROM_DIP,i; Wherein, ROM_MCP,i represents the range of motion of the metacarpophalangeal joint of finger i, ROM_PIP,i represents the range of motion of the proximal interphalangeal joint of finger i, and ROM_DIP,i represents the range of motion of the distal interphalangeal joint of finger i. The active range of motion (TAM_thumb) of the thumb is calculated using the following formula: TAM_thumb=ROM_CMC+ROM_MCP+ROM_IP; Wherein, ROM_CMC represents the range of motion of the carpometacarpal joint of the thumb, ROM_MCP represents the range of motion of the metacarpophalangeal joint of the thumb, and ROM_IP represents the range of motion of the interphalangeal joint of the thumb. For each active motion segment, the total active range of motion (TAM_total) of the five fingers is calculated using the following formula: ; For each active motion segment, the smoothness of the motion trajectory is characterized by a normalized Jerk value, calculated using the following formula: Jerk_norm=(T 5 / D 2 )·∫[d 3 θ_out,i(t) / dt 3 ] 2 dt; Where Jerk_norm represents the normalized Jerk value, T represents the duration of each active action segment, T=t_end-t_start, t_start and t_end represent the start and end points of the current active action segment, D represents the total change in the overall fusion angle of the main driving finger i within the current active action segment, D=max(θ_out,i(t))-min(θ_out,i(t)), θ_out,i(t) represents the overall fusion angle of the unique main driving finger i pre-specified for the current test action by the action standard protocol, and max(θ_out,i(t)) and min(θ_out,i(t)) are both obtained within t∈[t_start,t_end]. For selective motion segments pre-specified by the motion standard protocol, extract the separation motion index according to the following steps: Identify the non-target fingers to obtain a set N of non-target fingers; wherein, the non-target fingers refer to the fingers that should remain still when performing the test action of the selective action segment; Extract the angular velocity modulus |ω_rel,i(t)| of the main driving finger i relative to the back of the hand and the angular velocity modulus |ω_rel,k(t)| of each non-target finger k relative to the back of the hand, k∈N; If the number of valid sampling points of the main driving finger i and the non-target finger k meets the requirements, and the standard deviations of |ω_rel,i(t)| and |ω_rel,k(t)| within the selective motion segment are both greater than a preset fluctuation threshold, which indicates that the main driving finger i has valid motion, the Pearson correlation coefficient r_inter,k of |ω_rel,i(t)| and |ω_rel,k(t)| is calculated to obtain the correlation coefficient of each non-target finger k: r_inter,k=Cov[|ω_rel,i(t)|,|ω_rel,k(t)|] / (σ_rel,i·σ_rel,k); Wherein, σ_rel,i is the standard deviation of |ω_rel,i(t)| within the selective motion segment, σ_rel,k is the standard deviation of |ω_rel,k(t)| within the selective motion segment, the calculation interval is the start and end time of the selective motion segment, σ_rel,i>0, σ_rel,k>0; r_inter,k represents the Pearson correlation coefficient between |ω_rel,i(t)| and |ω_rel,k(t)| within the selective motion segment, k∈N; Cov[·] is the covariance operator; If σ_rel,i is greater than the preset fluctuation threshold, while σ_rel,k is not greater than the preset fluctuation threshold, and |ω_rel,k(t)| does not exceed the preset rest threshold within the selective motion segment, the corresponding r_inter,k is determined as 0; Based on the correlation coefficients of the non-target fingers of the selective motion segment, the fractional independence index FII is determined according to the following formula: FII = 1 - max(|r_inter,k|),k∈N; If the main driving finger i has no valid motion, mark FII as missing and prompt for retest until the main driving finger i has valid motion; For the active fist-clenching segment and the active five-finger stretching segment, the five-finger co-movement correlation coefficient ρ is calculated according to the following steps: Extract the overall fusion angles of the five fingers, and calculate the Pearson correlation coefficient r_{i,k} of any two fingers among the five fingers according to the following formula: r_{i,k}=corr(θ_out,i(t),θ_out,k(t)); Wherein, r_{i,k} represents the Pearson correlation coefficient of the i-th finger and the k-th finger, 1≤i<k≤5, θ_out,i(t) represents the overall fusion angle of the i-th finger, θ_out,k(t) represents the overall fusion angle of the k-th finger, and corr() represents a statistical function for calculating the Pearson correlation coefficient of θ_out,i(t) and θ_out,k(t); Calculate the correlation coefficient ρ of the coordinated movement of the five fingers using the following formula: ρ=(1 / 10)Σ|r{i,k}|.

[0011] Optionally, the maximum muscle strength index includes the maximum grip strength quantitative value F_peak,grip and the maximum pinch strength quantitative value F_peak,pinch; The extraction of maximum muscle strength indices for each active grasping segment and each active pinching segment specifically includes: For each active grasping segment and active pinching segment, the total muscle force F_total(t) is calculated according to the following formula: F_total(t) = ΣF_i(t); F_i(t) = k_i·P_i(t); Where F_i(t) represents the contact force of the calibrated i-th flexible pressure sensor, k_i represents the calibration coefficient, and P_i(t) represents the original pressure signal collected by the i-th flexible pressure sensor; the total muscle force F_total(t) is obtained by summing the contact forces F_i(t) of the relevant flexible pressure sensors specified in the current action rule. The peak value of the total muscle force F_total(t) for each active grasping segment, F_peak,grip, is extracted as the maximum grip force quantification value of the active grasping segment, and the peak value of the total muscle force F_total(t) for each active pinching segment, F_peak,pinch, is extracted as the maximum pinch force quantification value of the active pinching segment. For each passive extension segment of the finger, the extraction of muscle tone quantification indicators specifically includes: For any passive extension segment of a finger, within the preset contact position, loading direction, and angular velocity range, the average value of the contact force F_i(t) of the calibrated flexible pressure sensor that is consistent with the passive extension loading direction of the finger is calculated or the sum is calculated according to the preset weight to obtain the composite resistance F_resist(t) of the finger. For any passive extension segment of a finger, define the non-negative passive extension angular displacement. (t) = |θ_MCP(t) - θ_MCP(t_start)|, where θ_MCP(t) is the estimated fusion angle of the MCP joint of any finger at time t, and θ_MCP(t_start) is the estimated fusion angle of the MCP joint of any finger at the start of the movement. With F_resist(t) as the x-axis and F_resist(t) as the y-axis, construct the drag-extension angular displacement curve F_resist(t). ); Based on the aforementioned resistance-extension angular displacement curve F_resist( ),according to Extract the area under the non-negative curve (AUC), where... _end represents the non-negative extension angular displacement at the end of the passive extension segment of the finger; and the resistance-extension angular displacement curve F_resist( In the linear ascending segment of ), F_resist is relative to The non-negative fitting slope is used as the stiffness coefficient K_stiff; For each test action segment, the muscle activation level, antagonistic muscle synergistic contraction rate, and muscle fatigue are extracted, specifically including: For each test action segment, the preprocessed sEMG signals from the flexor side EMG acquisition channel F and the extensor side EMG acquisition channel E are subjected to full-wave rectification and low-pass envelope processing to obtain EMG_env,F(t) and EMG_env,E(t). For each path c∈{F,E}, the root mean square value of the integral electromyography (RMS_c) is calculated based on the electromyographic envelope EMG_env,c(t) according to the following formula: RMS_c = ; Where T = t_end - t_start represents the duration of each test action segment, t ∈ [t_start, t_end]; In the MVIC calibration test, an analysis window of the same duration as each test action segment is used to calculate RMS_MVIC,c, and the calculation is performed when the corresponding MVIC reference value is greater than the minimum effective value and the calibration signal quality is qualified. %RMS_MVIC,c=(RMS_c / RMS_MVIC,c)×100%.

[0012] Where %RMS_MVIC,c represents the standardized muscle activation level; Within the same effective action segment, the electromyographic envelopes of the flexor-side electromyographic acquisition channel F and the extensor-side electromyographic acquisition channel E are normalized according to their respective effective reference values ​​to obtain E_F(t) and E_E(t). The co-contraction rate CR of the flexor-side and extensor-side muscle groups is then calculated according to the following formula: CR=[∫min(E_F(t),E_E(t))dt / ∫max(E_F(t),E_E(t))dt]×100%.

[0013] Optionally, the Brunnstrom staging based on the multi-dimensional hand function feature matrix to obtain the Brunnstrom staging assessment reference results for the stroke patients specifically includes: Based on the multi-dimensional hand function feature matrix, it is determined whether the conditions of various features of each Brunnstrom stage are met; wherein, the various features include three categories: kinematic features, mechanical features, and electromyographic features. For each category of features in each Brunnstrom period: If the required features in the class features used for the Brunnstrom staging determination are complete and the required features satisfy the corresponding feature conditions of the Brunnstrom staging, then the indicator function of the class feature takes the value of 1; If the required features for the class feature used in the Brunnstrom staging determination are complete, but at least one of the required features does not meet the corresponding feature conditions of the Brunnstrom staging, then the indicator function of the class feature takes the value of 0. If any required feature is missing from the category feature used for Brunnstrom staging determination, the output data is incomplete and a retest prompt is given until the required feature is found to be complete in the category feature during retesting. Based on the above situation, the value of the indicator function of the category feature is determined. For each Brunnstrom stage, the score for that stage is calculated based on the indicator function value for each feature class, according to the following formula: S_stage =Σ_d(w_d×I(feature_d)); Where S_stage represents the score of the Brunnstrom stage, w_d represents the weight of the d-th feature of the Brunnstrom stage, feature_d represents the d-th feature of the Brunnstrom stage, and I(feature_d) represents the value of the indicator function of the d-th feature of the Brunnstrom stage. Based on the scores of each Brunnstrom stage, the Brunnstrom stage assessment reference results for the stroke patients are output, specifically including: If only one Brunnstrom period achieves the highest score, the Brunnstrom period with the highest score will be output as the candidate period.

[0014] Optionally, the step of outputting the Brunnstrom staging assessment reference results for the stroke patient based on the scores of each Brunnstrom stage further includes: If all features involved in the score calculation for any Brunnstrom phase are complete, and the scores for all Brunnstrom phases are 0, a prompt will be output indicating that the current quantitative rule base phase conditions have not been met. If the highest and second-highest scores of all Brunnstrom periods correspond to adjacent Brunnstrom periods, and the difference between the highest and second-highest scores satisfies the transition judgment conditions pre-set in the quantitative rule base, and at least one pre-specified key feature appears in a cross-period fulfillment state between the two adjacent Brunnstrom periods, then the transition period prompt for the two adjacent Brunnstrom periods is output, and the conflict feature is recorded; if the calibration of the transition judgment conditions is not completed, the scores and conflict features of the two adjacent Brunnstrom periods are output for manual review. If the above conditions are not met and two or more Brunnstrom stages achieve the highest score, or if the two Brunnstrom stages corresponding to the highest score are not adjacent, output a message indicating a conflict between the two Brunnstrom stages with the highest scores and prompt for retesting or manual review.

[0015] Secondly, this application provides a phased hand motor rehabilitation evaluation system for stroke patients, including: The data glove system is worn on the hands of stroke patients during standardized hand function testing to collect kinematic, mechanical, and sEMG signals from the stroke patient's hand during the standardized hand function testing process. The processing module is used to execute the phased hand motor rehabilitation evaluation method for stroke patients described in any of the above-mentioned methods.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method and system for evaluating the staged hand motor rehabilitation of stroke patients. It utilizes a data glove system to synchronously collect kinematic, mechanical, and sEMG signals from the stroke patient's hand, comprehensively capturing the kinematic trajectory, biomechanical resistance, and neuromuscular coordinated control information of hand movements. This avoids blind spots in single-dimensional approaches and more accurately and completely reveals the multidimensional pathological nature of post-stroke hand dysfunction. Preprocessing of the acquired kinematic, mechanical, and sEMG signals suppresses noise and feature distortion. Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors. The original feature values ​​and standardized values ​​are organized according to action type and feature fields to form a multidimensional hand function feature matrix, reducing the impact of heterogeneous signal response frequency differences and data temporal inconsistencies on feature calculation. Using a quantitative rule base based on Brunnstrom's recovery staging theory, combined with weighted scores of kinematic, mechanical, and electromyographic features, the system outputs Brunnstrom staging, staging score, and deviation of each dimension's features. In summary, this application can comprehensively utilize hand function information from different modalities, solving the problems that traditional devices often rely on a single sensor and have large differences in response frequency and noise characteristics of heterogeneous signals, which can easily lead to timing misalignment and feature distortion. This is beneficial to improving the accuracy, interpretability, and verifiability of phased hand movement rehabilitation assessment for stroke patients. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of the structure of a phased hand movement rehabilitation evaluation system for stroke patients provided in an embodiment of this application; Figure 2 for Figure 1 A schematic diagram of the back of the hand structure of the data glove system; Figure 3 for Figure 1 A schematic diagram of the hand structure of the data glove system; Figure 4 A flowchart illustrating a method for evaluating the phased hand motor rehabilitation of stroke patients, provided as an embodiment of this application; Figure 5 This is a schematic diagram of the resistance-nonnegative passive extension angular displacement curve, the area under the curve, the stiffness coefficient, and the hysteresis loop area according to an embodiment of this application. Figure 6 for Figure 1A schematic diagram of the processing module; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Stroke is one of the leading causes of long-term disability in adults, with over 50% of stroke patients experiencing long-term upper limb and hand motor dysfunction, severely impacting their quality of life. In rehabilitation engineering practice, the recovery process of hand function is typically divided into stages based on Brunnstrom's classic six-stage theory. Each stage exhibits different neuromuscular control data characteristics in dimensions such as muscle strength, muscle tone, movement patterns, and coordination. Therefore, automated stage classification and quantitative characterization of multimodal sensor data is a core prerequisite for achieving closed-loop control in rehabilitation assessment systems. However, existing assessment methods largely rely on manual scale scoring, resulting in limited data consistency and difficulty in objectively capturing the continuous signal characteristics during the transition from commensurate movement to partially isolated movement. This leads to intervention parameter adjustments often lagging behind the patient's actual recovery level.

[0022] In recent years, wearable sensing technology has provided a data acquisition foundation for the objectification of hand function assessment. However, in practical engineering implementation and multimodal signal processing, hand function assessment methods based on wearable sensing technology have the following defects ① to ④. Among them: ① Traditional devices often rely on a single sensor (measuring only angle or only electromyography), and the heterogeneous signals have large differences in response frequency and noise characteristics, which can easily lead to timing misalignment and feature distortion.

[0023] ② Traditional fixed threshold segmentation algorithms are difficult to adapt to the slow, small-amplitude, and irregular movement characteristics of stroke patients, resulting in inaccurate segmentation of the start and end points of the movement cycle, misalignment of subsequent feature extraction windows, and low signal-to-noise ratio; ③ Insufficient accuracy of inertial signal drift and multi-source data fusion under micro-motion: Traditional IMUs exhibit significant integral drift during low-speed, small-angle joint movements, while relying solely on flexible bending sensors will result in the loss of high-frequency dynamic details.

[0024] ④ The existing scheme lacks an adaptive weighted filtering and fusion mechanism, resulting in large cumulative errors in joint angles and angular velocity trajectories, making it difficult to support reliable calculation of refined kinematic features.

[0025] The aforementioned shortcomings result in low accuracy of existing hand function assessment methods based on wearable sensing technology.

[0026] To address, or at least partially address, the deficiencies in the aforementioned related technologies, in one exemplary embodiment, such as Figure 1 As shown, a phased hand motor rehabilitation evaluation system for stroke patients is provided, including a data glove system 1 and a processing module 2. The data glove system 1 is worn on the stroke patient's hand during standardized hand function testing to collect kinematic signals, mechanical signals, and sEMG signals from the stroke patient's hand during the standardized hand function testing process. The processing module 2 is used to execute the phased hand motor rehabilitation evaluation method for stroke patients described below.

[0027] During the standardized hand function test, stroke patients performed the following sequence of actions in sequence: ①Active movements: actively clenching fists in the resting position, actively extending all five fingers in the resting position, pinching the thumb against the other four fingers in sequence, pinching the thumb and index finger sideways, cylindrical grasping, and spherical grasping; ② Passive movement: This includes the passive extension phase and the passive rebound phase. In the passive extension phase, the patient's hand is in a relaxed state. The rehabilitation training equipment or therapist passively extends the patient's relaxed fingers to the maximum extension position at a constant rate. In the passive rebound phase, the rehabilitation training equipment or therapist slowly and passively moves the fingers in the opposite direction at the same or similar angular velocity as in the passive extension phase, so that the fingers move from the maximum extension position to the flexion position.

[0028] The resting hand position refers to a relaxed posture in which the hand is naturally still and free from stress. At this time, the tension of the intrinsic and extrinsic muscles, joint capsules, and ligaments of the hand is in a relatively balanced state, which is equivalent to the natural posture of the hand when a person is asleep or under anesthesia.

[0029] In another exemplary embodiment of this application, such as Figures 2 to 3As shown, the data glove system 1 includes a first inertial measurement unit 3, five second inertial measurement units 4, five flexible bending sensors 5, thirteen flexible pressure sensors 6, and a surface electromyography (EMG) acquisition component 7. The first inertial measurement unit 3 is located on the back of the hand, the five second inertial measurement units 4 are correspondingly arranged on the fingertips, the five flexible bending sensors 5 are correspondingly arranged along the back of the fingers, and the thirteen flexible pressure sensors 6 are correspondingly embedded in the fingertips, palmar surface of the metacarpophalangeal joint, ulnar surface of the thumb, the thenar eminence, and hypothenar eminence. The surface EMG acquisition component 7 includes a first bipolar surface electrode group, a second bipolar surface electrode group, and... The system includes a reference electrode and two electromyography (EMG) acquisition channels for the flexor and extensor sides, respectively, to receive differential signals. A first bipolar surface electrode group is placed on the skin surface corresponding to the flexor or extensor muscles of the forearm in stroke patients, and a second bipolar surface electrode group is placed on the skin surface corresponding to the extensor or extensor muscles of the forearm in stroke patients, respectively detecting surface EMG signals at the corresponding locations. The two EMG acquisition channels synchronously acquire the raw surface EMG signals EMG_raw,F(t) on the flexor side and EMG_raw,E(t) on the extensor side based on the same sampling clock or unified timestamp, and share the reference electrode.

[0030] Among them, the first inertial measurement unit 3 and the second inertial measurement unit are used to collect the triaxial angular velocity ω_raw(t), triaxial acceleration a_raw(t), triaxial magnetic field m_raw(t) and the corresponding original attitude quaternion q_raw(t) at the current position. The flexible bending sensor 5 is used to collect the original bending electrical signal Vbend,i(t) at the current position. The flexible pressure sensor 6 is used to collect the original pressure signal P_i(t) at the current position. The flexor side electromyography acquisition channel and the extensor side electromyography acquisition channel of the surface electromyography acquisition component 7 simultaneously acquire two original voltage sequences EMG_raw,F(t) and EMG_raw,E(t).

[0031] In one exemplary embodiment, a method for evaluating the phased hand motor rehabilitation of stroke patients is provided, which is applied to the aforementioned system for evaluating the phased hand motor rehabilitation of stroke patients, such as... Figure 4 As shown, the process includes steps 101 to 105. Wherein: Step 101: Acquire the kinematic signals, mechanical signals, and sEMG signals synchronously collected by the data glove system during the standardized hand function test.

[0032] The kinematic signals include the triaxial angular velocity ω_raw(t), triaxial acceleration a_raw(t), and triaxial magnetic field m_raw(t) collected by the first inertial measurement unit 3 and each of the second inertial measurement units 4, and the corresponding original attitude quaternion q_raw(t) calculated from the triaxial angular velocity ω_raw(t), triaxial acceleration a_raw(t), and triaxial magnetic field m_raw(t), as well as the original bending electrical signal Vbend,i(t) collected by each flexible bending sensor 5. The original attitude quaternion q_raw(t) is a unit quaternion calculated by the sensor fusion algorithm built into each IMU (Inertial Measurement Unit) using the triaxial angular velocity ω_raw(t), triaxial acceleration a_raw(t), and triaxial magnetic field m_raw(t), representing the real-time attitude of the IMU rigid body coordinate system relative to the Earth inertial coordinate system.

[0033] The raw mechanical signals include the raw pressure signals P_i(t) acquired by each flexible pressure sensor 6. The raw sEMG signals include the raw voltage sequences EMG_raw,F(t) and EMG_raw,E(t) acquired by the surface electromyography acquisition component 7.

[0034] In this embodiment, two sets of bipolar surface electrodes are arranged along the muscle fiber direction of the target flexor and extensor sides, respectively. Before acquisition, the electrode position, skin treatment, and contact status are recorded; if any electrode falls off, signal saturation occurs, power frequency interference exceeds the limit, or other signal quality is unqualified, the corresponding reason is recorded, and the CR field that depends on the dual signals is marked as missing and not replaced by a single signal.

[0035] Step 102: Preprocess the acquired kinematic signals, mechanical signals and sEMG signals.

[0036] Step 103: Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each test action segment; wherein, the features include kinematic features, mechanical features and / or electromyographic features, and the heterogeneous feature vectors include kinematic heterogeneous feature vectors, mechanical heterogeneous feature vectors and / or electromyographic heterogeneous feature vectors.

[0037] Step 104: Based on the heterogeneous feature vectors of all test action segments, perform multi-dimensional feature fusion and time alignment to obtain a multi-dimensional hand function feature matrix.

[0038] Step 105: Based on the multi-dimensional hand function feature matrix, Brunnstrom staging is performed to obtain reference results for Brunnstrom staging assessment of stroke patients, which can be used as reference information for rehabilitation assessment.

[0039] In this embodiment of the application, the Brunnstrom staging assessment results of stroke patients are used as reference information for rehabilitation assessment to assist doctors in assessing the rehabilitation status of stroke patients.

[0040] By implementing steps 101 to 104 above, kinematic signals, mechanical signals, and sEMG signals of the stroke patient's hand are synchronously acquired through the data glove system. This comprehensively captures the kinematic trajectory, biomechanical resistance, and neuromuscular coordinated control information of hand movements, avoiding blind spots in a single dimension and revealing the multidimensional pathological nature of post-stroke hand dysfunction more accurately and completely. Preprocessing the acquired kinematic, mechanical, and sEMG signals suppresses noise and feature distortion. Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each segment. Multidimensional feature fusion and time alignment are then performed based on these heterogeneous feature vectors to suppress temporal misalignment and feature distortion caused by large differences in heterogeneous signal response frequencies. In summary, this application solves the problem that traditional devices often rely on a single sensor, and the large differences in heterogeneous signal response frequencies and noise characteristics easily lead to temporal misalignment and feature distortion, thus improving the accuracy of phased hand movement rehabilitation assessment for stroke patients.

[0041] In another exemplary embodiment of this application, the above-described method for evaluating the phased hand motor rehabilitation of stroke patients, before step 102, further includes: The quality of the acquired kinematic signals, mechanical signals, and sEMG signals is checked to see if they meet the preset quality requirements. If the kinematic signal, mechanical signal and sEMG signal all meet the preset quality requirements, proceed to step 102; If any of the kinematic signal, mechanical signal, or sEMG signal fails to meet the preset quality requirements, a retest action will be prompted to reacquire the signal until all three signals meet the preset quality requirements.

[0042] In another exemplary embodiment of this application, step 102 specifically includes steps 201 to 2010, wherein: Step 201: Initialize the attitude and align the original attitude quaternions q_raw(t) collected by the first inertial measurement unit 3 and each of the second inertial measurement units 4 with the local coordinate system to obtain the attitude-aligned quaternion q_0(t) of the first inertial measurement unit 3 and the attitude-aligned quaternion q_i(t) of each of the second inertial measurement units 4.

[0043] Step 202: For each finger, calculate the relative attitude quaternion of the corresponding second inertial measurement unit relative to the first inertial measurement unit based on the aligned quaternion of the first inertial measurement unit and the corresponding second inertial measurement unit.

[0044] In this embodiment of the application, step 202 specifically includes: Step 2021: Using the first inertial measurement unit 3 as a reference, calculate its inverse quaternion q_0(t)^(-1); Step 2022: For each quaternion q_i(t) of the second inertial measurement unit 4, calculate its relative attitude quaternion q_rel,i(t) relative to the first inertial measurement unit 3: q_rel,i(t) = q_0(t)^(-1) q_i(t); Wherein, q_rel,i(t) reflects the overall posture change of the i-th finger relative to the back of the hand.

[0045] Step 203: Convert the relative posture quaternion q_rel,i(t) into the overall Euler angle of the corresponding finger relative to the back of the hand, and extract the angle component corresponding to the finger flexion and extension movement as the IMU solution angle θ_IMU,i(t) of the i-th finger.

[0046] In this embodiment of the application, the relative attitude quaternion q_rel,i(t) is converted into Euler angles (ZYX order) using the conventional quaternion to Euler angle conversion formula, where the pitch angle corresponds to the flexion and extension motion of the finger.

[0047] Step 204: Based on the adaptive complementary filter, the angle calculated by the IMU on each finger is fused with the bending reference angle of the flexible bending sensor to obtain the overall fused angle of each finger.

[0048] In this embodiment of the application, the original bending electrical signal V_bend,i(t) of the flexible bending sensor 5 on each finger is mapped to a bending reference angle θ_f,i(t) through factory calibration or multi-point posture calibration. The mapping relationship can be a linear model or a calibration function obtained by polynomial fitting. The linear model is θ_f,i(t)=k_i·V_bend,i(t)+b_i, where k_i and b_i are calibration coefficients.

[0049] Meanwhile, the Euler angle θ_IMU,i(t) obtained in step 203 above is used as the inertial measurement angle θ_i(t).

[0050] Angle fusion of θ_i(t) and θ_f,i(t) is performed based on an adaptive complementary filter to obtain a fused angle θ_out,i(t) with high denoising and steady-state accuracy. The specific formula is as follows: θ_out,i(t) =α_i(t)·θ_IMU,i(t) + [1-α_i(t)]·θ_f,i(t); Wherein, α_i(t) is the adaptively adjusted weighting coefficient; when the signal of the second inertial measurement unit is greatly affected by short-time acceleration interference, α(t) is decreased, and when the flexible bending sensor has obvious drift, α(t) is increased.

[0051] Specifically, the adaptive adjustment process of the weight coefficient α(t) includes the following steps 2041 to 2042, wherein: Step 2041: Based on the adjustment rules for acceleration interference detection and the adjustment rules for drift detection based on flexible bending sensor, adjust α(t), and the adjustment rate of α(t) shall not exceed the preset maximum adjustment step size.

[0052] In this embodiment, the adjustment rule based on acceleration interference detection is as follows: Acceleration deviation Δa(t) = |a_mag(t) - g| / g. Here, a_mag(t) is the acceleration modulus of each second inertial measurement unit (IMU), and g is the gravitational acceleration. If Δa(t) of any second IMU is greater than the pre-calibrated acceleration deviation threshold a_th for the current IMU range, sampling rate, and wearing method, it is determined that the second IMU is subject to linear acceleration interference. The α_i(t) corresponding to the finger where the second IMU is located is reduced, and α_i(t) is gradually restored when Δa(t) is not greater than a_th.

[0053] The adjustment rule based on the drift detection of the flexible bending sensor is as follows: During the static period (i.e., when the angular velocity modulus |ω_i(t)| of the second inertial measurement unit corresponding to the i-th finger is lower than the preset modulus threshold and continues for more than a set time), if the angle difference between the second inertial measurement unit and the corresponding flexible bending sensor exceeds the set angle difference threshold, it is determined that the flexible bending sensor has drifted, and the α_i(t) used to calculate the fusion angle θ_out,i(t) of the finger where the flexible bending sensor is located is increased. Here, |ω_i(t) is used for the static period determination.

[0054] Step 2042, detect the overall limit: whether the value of α_i(t) is within the range of [α_min,i, α_max,i]. If α_i(t) is less than α_min,i, adjust the value of α_i(t) to α_min,i. If α_i(t) is greater than α_max,i, adjust the value of α_i(t) to α_max,i.

[0055] The maximum adjustment amounts of a_th, static modulus threshold, angle difference threshold, α_min,i, α_max,i, and α_i(t) per unit time are all determined by the device calibration and stored in the configuration with version number.

[0056] Step 205: Based on the finger joint angle allocation parameters pre-calibrated and fixed for the subject or device model, the overall fusion angle of each finger is allocated to the corresponding finger joint to obtain the estimated fusion angle of each finger joint. Specifically, the finger joints of fingers 2-5 include the metacarpophalangeal joint, proximal interphalangeal joint, and distal interphalangeal joint; the finger joints of the thumb include the carpometacarpophalangeal joint, metacarpophalangeal joint, and interphalangeal joint. The estimated fusion angle of each finger joint is obtained by allocating it according to the allocation ratio of the overall fusion angle of the corresponding finger.

[0057] In this embodiment, the angle distribution ratios of the metacarpophalangeal joints, proximal interphalangeal joints, and distal interphalangeal joints of the 2nd to 5th fingers, as well as the angle distribution ratios of the carpometacarpophalangeal joints, metacarpophalangeal joints, and interphalangeal joints of the thumb, are all determined and recorded based on pre-collected calibration motion data, and the sum of the angle distribution ratios corresponding to each joint of the same finger is 100%. The estimated fusion angle of each joint is calculated based on the overall fusion angle of the corresponding finger and the corresponding angle distribution ratio, and is not expressed as an independent direct measurement value of each joint.

[0058] Step 206: Perform zero-point drift compensation and temperature drift calibration on the raw pressure signals collected by each flexible pressure sensor.

[0059] Step 207: Based on the sampling rate, the effective bandwidth of the surface electromyography acquisition component, and the local power frequency, EMG_raw,F(t) and EMG_raw,E(t) are subjected to bandpass filtering and power frequency notch processing using the same parameters.

[0060] In this embodiment of the application, when the sampling rate is sufficient to cover the corresponding bandwidth and the local power frequency is 50Hz, a 20-450Hz bandpass filter and a 50Hz power frequency notch filter can be used to suppress low-frequency baseline drift, high-frequency noise and power frequency interference; the actual parameters are recorded in the processing configuration file and saved with the results.

[0061] Step 208: The three-axis angular velocities ω_raw(t) output by the first inertial measurement unit 3 and each of the second inertial measurement units 4 are aligned in attitude, and the three-axis angular velocities of each finger relative to the back of the hand are calculated based on the three-axis angular velocities after transformation to a unified coordinate system.

[0062] In this embodiment, the three-axis angular velocity of the first inertial measurement unit 3 in a unified coordinate system... Represented as: ; No. This refers to the triaxial angular velocity of the corresponding second inertial measurement unit 4 in a unified coordinate system. Represented as: ; in, express x-axis component, express The y-axis component, express The z-axis component.

[0063] For the This refers to, based on the same sampling time and Calculate the triaxial angular velocity relative to the back of the hand using the following formula. : ; in: ; in, These correspond to the thumb, index finger, middle finger, ring finger, and little finger, respectively. express x-axis component, express The y-axis component, express The z-axis component.

[0064] Step 209: Based on the triaxial angular velocities of each finger relative to the back of the hand obtained in step 208, calculate the angular velocity modulus of each finger relative to the back of the hand.

[0065] In this embodiment of the application, the first The angular velocity modulus relative to the back of the hand is calculated using the following formula: ; in, They represent the first The three axial components of the triaxial angular velocity relative to the back of the hand in a unified coordinate system; Indicates the first It refers to the angular velocity modulus relative to the back of the hand, which is used for motion segmentation, motion initiation detection, and motion velocity parameter calculation for the corresponding test action.

[0066] Step 2010: Timestamp align the multimodal raw data stream according to a unified sampling clock, and output the time-synchronized multimodal raw data stream and the corresponding overall fusion angle, estimated fusion angle of each finger joint, and triaxial angular velocity and angular velocity magnitude of each finger relative to the back of the hand; wherein, the multimodal raw data stream includes kinematic signals, mechanical signals, and sEMG signals from the preprocessed standardized hand function test process.

[0067] In another exemplary embodiment of this application, the above-described method for evaluating the phased hand motor rehabilitation of stroke patients, before step 103, further includes: Step 301 involves using an automatic motion cycle segmentation algorithm based on fused kinematic thresholds to segment the multimodal raw data stream into action segments, obtaining each test action fragment. Step 301 specifically includes steps 3011 to 3014, wherein: Step 3011: For each test action, determine the primary driving finger; wherein, for test actions that require only one finger to move, the primary driving finger is the finger that needs to move when performing the test action; for test actions that require multiple fingers to move simultaneously, the primary driving finger is determined according to the provisions of the action standard protocol.

[0068] Step 3012: Detect each target time period where the relative angular velocity modulus corresponding to the main driving finger of each test action is greater than the preset starting threshold ω_start and the duration reaches the first target time, and determine the moment when the relative angular velocity modulus first exceeds ω_start in each target time period as the action start point of the corresponding action.

[0069] Step 3013: For test actions that do not include a hold phase in the action protocol, detect the time period during which the relative angular velocity modulus corresponding to the main driving finger of the test action falls below a preset termination threshold ω_end and the duration reaches the second target time. The moment when the relative angular velocity modulus first falls below ω_end in the time period is determined as the action end point of the corresponding action. For test actions that include a hold phase in the action protocol, the moment when the relative angular velocity modulus corresponding to the main driving finger of the test action first falls below ω_end and meets the second target time is determined as the motion phase end point or the target posture arrival time t_arr. The subsequent hold phase is retained, and the earliest time among the test action end prompt, release start, or the end of the maximum hold window specified in the action protocol is determined as the action end point of the test action.

[0070] Among them, the action rules for grasping or pinching can also be configured with corresponding force signal holding conditions.

[0071] Step 3014: Based on the start and end points of each test action, the time-synchronized multimodal raw data stream is divided into independent data segments corresponding to each test action to obtain each test action segment.

[0072] In this embodiment, the starting threshold ω_start and the ending threshold ω_end are preset based on the mean and standard deviation of the relative angular velocity modulus during the resting period, or adaptively calibrated based on the patient's baseline test, and satisfy ω_end < ω_start to form a hysteresis interval for action start and end detection; the ω_start, ω_end, first target time, second target time and maximum hold window are determined by the resting noise and calibrated action data under the same device, wearing method and action protocol, satisfy ω_end < ω_start, and are fixed in the versioned rule base along with the applicable conditions.

[0073] The test actions include active and passive actions, and the test action segments include active action segments and passive action segments. Active action segments include active fist clenching segment, active five-finger extension segment, thumb and other four fingers pinching segment, thumb and index finger side pinching segment, cylindrical grasp segment, and ball-shaped grasp segment. Passive action segments include passive finger extension segment.

[0074] In another exemplary embodiment of this application, step 103 specifically includes: Step 401: For each active motion segment, extract kinematic features such as the joint range of motion of each finger joint, the total active range of motion of the five fingers, and the smoothness of the motion trajectory; for the selective motion segments pre-specified by the motion standard protocol, further extract the separation motion index (FII); for the active fist clenching segment and the active five-finger extension segment, calculate the correlation coefficient ρ of the common motion of the five fingers to obtain the corresponding kinematic heterogeneous feature vector; wherein, the finger joints of the 2nd to 5th fingers include the metacarpophalangeal joint (MCP), the proximal interphalangeal joint (PIP), and the distal interphalangeal joint (DIP), and the finger joints of the thumb include the carpometacarpophalangeal joint (CMC), the metacarpophalangeal joint (MCP), and the interphalangeal joint (IP).

[0075] For each active movement segment, the range of motion of each finger joint is calculated according to the following formula: ROM_(i,j)=max(θ_(i,j)(t))-min(θ_(i,j)(t)); Where i represents the finger number, i∈{1,2,3,4,5}, representing the thumb, index finger, middle finger, ring finger, and little finger respectively; j represents the joint type of finger i, ROM_(i,j) represents the joint mobility of the finger joint j of finger i; θ_(i,j)(t) represents the estimated fusion angle of the middle finger joint j of finger i, and max(θ_(i,j)(t)) and min(θ_(i,j)(t)) are both obtained within t∈[t_start,t_end], where t_start and t_end represent the start and end points of the current active action segment.

[0076] For fingers 2-5, the active range of motion TAM_finger,i of finger i is calculated using the following formula: TAM_finger,i=ROM_MCP,i+ROM_PIP,i+ROM_DIP,i; Wherein, ROM_MCP,i represents the range of motion of the metacarpophalangeal joint of finger i, ROM_PIP,i represents the range of motion of the proximal interphalangeal joint of finger i, and ROM_DIP,i represents the range of motion of the distal interphalangeal joint of finger i.

[0077] For the thumb, the active range of motion (TAM_thumb) is calculated using the following formula: TAM_thumb=ROM_CMC+ROM_MCP+ROM_IP; Wherein, ROM_CMC represents the range of motion of the carpometacarpal joint of the thumb, ROM_MCP represents the range of motion of the metacarpophalangeal joint of the thumb, and ROM_IP represents the range of motion of the interphalangeal joint of the thumb.

[0078] For each active motion segment, the total active range of motion (TAM_total) of the five fingers is calculated using the following formula: ; Among them, TAM_total is the comprehensive active range of motion index of the five fingers defined in this application based on the above-mentioned joint composition.

[0079] For each active motion segment, the smoothness of the motion trajectory is characterized by the normalized Jerk value, calculated using the following formula: Jerk_norm=(T 5 / D 2 )·∫[d 3 θ_out,i(t) / dt 3 ] 2 dt; Where Jerk_norm represents the normalized Jerk value, T represents the duration of each active action segment, T=t_end-t_start, t_start and t_end represent the start and end points of the current active action segment, D represents the total change in the overall fusion angle of the main driving finger i within the current active action segment, D=max(θ_out,i(t))-min(θ_out,i(t)), θ_out,i(t) represents the overall fusion angle of the unique main driving finger i pre-specified for the current test action by the action standard protocol, and max(θ_out,i(t)) and min(θ_out,i(t)) are both obtained within t∈[t_start,t_end]. Jerk_norm is calculated only when D is greater than the minimum effective amplitude D_min calibrated for the current action in the rule base and the number of effective samples and signal quality meet the requirements; otherwise, it is marked as missing and a retest is prompted.

[0080] In this embodiment, the motion standard protocol pre-assigns a unique primary driving finger i for each motion type, and uses the same primary driving finger and calculation interval in both patient data and health reference data. Jerk_norm is calculated only when D is greater than the minimum effective amplitude D_min calibrated for the current motion in the rule base and the number of effective samples and signal quality meet the requirements; otherwise, it is marked as missing and a retest is prompted. The smaller the Jerk_norm value, the smoother the motion trajectory under this definition.

[0081] Optionally, the maximum value ω_max of the relative angular velocity modulus |ω_rel,i(t)| corresponding to the target finger or the primary driving finger pre-specified in the standard protocol of the current test action within the effective active motion segment is extracted as an auxiliary motion velocity index. ω_max is not included in the calculation of the Brunnstrom stage score S_stage, but is retained as an auxiliary evaluation feature in the multi-dimensional hand function feature matrix X for motion quality verification, deviation from the healthy baseline calculation, and longitudinal comparison within the same patient.

[0082] For the selective motion segments pre-specified by the motion standard protocol, extract the separation motion index according to steps 4011 to 4014, wherein: Step 4011: Determine the non-target fingers to obtain a set N of non-target fingers; wherein, the non-target fingers refer to the fingers that should remain still when performing the test action of the selective action segment.

[0083] When the active motion segment involves the thumb pinching against the other four fingers in sequence, the finger directly in contact with the thumb is designated as the primary driving finger, the thumb as the participating finger, and the other three fingers that should remain stationary are designated as non-target fingers. When the active motion segment involves a thumb-index finger side-pinching segment, the index finger is designated as the primary driving finger, the thumb as the participating finger, and the middle, ring, and little fingers as non-target fingers. Active fist clenching, active five-finger extension, cylindrical grasping, and ball-shaped grasping segments are not directly included in the separation motion index.

[0084] Step 4012: Extract the angular velocity magnitude |ω_rel,i(t)| of the main driving finger i relative to the back of the hand and the angular velocity magnitude |ω_rel,k(t)| of each non-target finger k relative to the back of the hand, k∈N.

[0085] Step 4013: If the number of effective sampling points for the main driving finger i and the non-target finger k meets the requirements, and the standard deviations of |ω_rel,i(t)| and |ω_rel,k(t)| within the selected action segment are both greater than the preset fluctuation threshold, indicating that the main driving finger i has effective movement, calculate the Pearson correlation coefficient r_inter,k of |ω_rel,i(t)| and |ω_rel,k(t)| to obtain the correlation coefficient for each non-target finger k: r_inter,k=Cov[|ω_rel,i(t)|,|ω_rel,k(t)|] / (σ_rel,i·σ_rel,k); Where σ_rel,i is the standard deviation of |ω_rel,i(t)| within the selective action segment, σ_rel,k is the standard deviation of |ω_rel,k(t)| within the selective action segment, the calculation interval is the start and end time of the selective action segment, σ_rel,i>0, σ_rel,k>0; r_inter,k represents the Pearson correlation coefficient between the sequence |ω_rel,i(t)| and the sequence |ω_rel,k(t)| within the above selective action segment, k∈N; Cov[·] is the covariance operator.

[0086] If σ_rel,i is greater than the preset fluctuation threshold, while σ_rel,k is not greater than the preset fluctuation threshold, and |ω_rel,k(t)| does not exceed the preset resting threshold within the selective action segment, then the corresponding r_inter,k is set to 0.

[0087] The effective sampling point number of the main driving finger i and the non-target finger k refers to the number of sampling points of the kinematic signal of the main driving finger i and the non-target finger k that meet the quality requirements.

[0088] Step 4014: Based on the correlation coefficient of the non-target finger in the selective motion segment, the fractionalized movement index FII is determined according to the following formula: FII = 1 - max(|r_inter,k|),k∈N; Herein, FII characterizes the degree of motion decoupling between the main driving finger and other non-target fingers, max(|r_inter,k|) represents the maximum value of the absolute value of the correlation coefficient of the non-target finger in the selective motion segment. The closer the FII value is to 1, the higher the degree of motion decoupling between the target motion finger and other non-target fingers, and the stronger the independent motion capability.

[0089] If the main driving finger i has no valid motion, mark FII as missing and prompt for retest until the main driving finger i has valid motion.

[0090] For the active fist-clenching segment and the active five-finger stretching segment, the five-finger co-movement correlation coefficient ρ is calculated according to the following steps: The overall fusion angle θ_out,i(t) of the five fingers is extracted, and the Pearson correlation coefficient r_{i,k} between any two of the five fingers is calculated according to the following formula: r_{i,k}=corr(θ_out,i(t),θ_out,k(t)); Herein, r_{i,k} represents the Pearson correlation coefficient of the i-th finger and the k-th finger, 1≤i<k≤5, θ_out,i(t) represents the overall fusion angle of the i-th finger, θ_out,k(t) represents the overall fusion angle of the k-th finger, and corr() represents the statistical function for calculating the Pearson correlation coefficient of θ_out,i(t) and θ_out,k(t); And the five-finger co-movement correlation coefficient is calculated according to the following formula: ρ=(1 / 10)Σ|r{i,k}|; Herein, 10 is the number of pairwise combinations of the five fingers.

[0091] Step 402: For each active grasping segment and each active pinching segment, the maximum muscle force index is extracted, to obtain the mechanical heterogeneous feature vector of each active grasping segment and each active pinching segment.

[0092] Herein, the active grasping segment includes a cylindrical grasping segment and a spherical grasping segment, and the active pinching segment includes a segment of the thumb sequentially pinching the remaining four fingers and a segment of lateral pinching by the thumb and index finger. The maximum muscle force index includes the maximum grip force quantitative value F_peak,grip and the maximum pinch force quantitative value F_peak,pinch.

[0093] Specifically, step 402 specifically includes the following step 4021 to step 4022, where: Step 4021, for each active grasping segment and active pinching segment, calculate the total muscle force F_total(t) according to the following formula: F_total(t) = ΣF_i(t); F_i(t) = k_i·P_i(t); Where F_i(t) represents the contact force of the calibrated i-th flexible pressure sensor, k_i represents the calibration coefficient, and P_i(t) represents the original pressure signal collected by the i-th flexible pressure sensor; the total muscle force F_total(t) is obtained by summing the contact forces F_i(t) of the relevant flexible pressure sensors specified in advance by the current action rules.

[0094] Step 4022: Extract the peak value F_peak,grip of the total muscle force F_total(t) of each active grasping segment as the maximum grip force value of that active grasping segment, and extract the peak value F_peak,pinch of the total muscle force F_total(t) of each active pinching segment as the maximum pinch force value of that active pinching segment.

[0095] Among them, the two types of peaks (F_peak,grip and F_peak,pinch) are obtained by using the same action type, the same set of sensors and the same calculation method to obtain the health reference distribution.

[0096] Step 403: For each passive extension segment of a finger, extract the muscle tension quantification index to obtain the mechanical heterogeneous feature vector of each passive extension segment of a finger.

[0097] In this embodiment of the application, step 403 specifically includes: Step 4031: For any passive extension segment of a finger, within the preset contact position, loading direction and angular velocity range, the average value of the contact force F_i(t) of the calibrated flexible pressure sensor that is consistent with the passive extension loading direction of the finger is calculated or the sum is weighted according to the preset weight to obtain the synthetic resistance F_resist(t) of the finger.

[0098] In this embodiment, the passive extension direction of the finger is determined by the trend of the fusion angle change of the finger. The fusion angle changes from large to small to extended. The flexible pressure sensor on the same line is the flexible pressure sensor of the finger, including the flexible pressure sensor on the fingertip and the flexible pressure sensor on the palmar side of the metacarpophalangeal joint.

[0099] Step 4032: For any passive extension segment of a finger, define the non-negative passive extension angular displacement. (t) = |θ_MCP(t) - θ_MCP(t_start)|, where θ_MCP(t) is the estimated fusion angle of the MCP joint of any finger at time t, and θ_MCP(t_start) is the estimated fusion angle of the MCP joint of any finger at the start of the movement; With F_resist(t) as the x-axis and F_resist(t) as the y-axis, construct the drag-extension angular displacement curve F_resist(t). ).

[0100] In this embodiment, to ensure that different passive extension segments use a unified kinematic transverse axis, the estimated fusion angle of the MCP joint obtained in step 205 is used to construct a non-negative angular displacement. .

[0101] Step 4033, based on the constructed drag-extension angular displacement curve F_resist( ),according to Extract the area under the non-negative curve (AUC), where... _end represents the non-negative extension angular displacement at the end of the passive extension segment of the finger; and the resistance-extension angular displacement curve F_resist( In the linear ascending segment of ), F_resist is relative to The non-negative fitting slope is used as the stiffness coefficient K_stiff.

[0102] Among them, the area under the non-negative curve (AUC) and the stiffness coefficient (K_stiff) are used for subsequent comparisons of Brunnstrom staging conditions.

[0103] Optionally, the peak resistance F_peak_resist, the spasm capture angle θ_catch, and the hysteresis loop area are further extracted and retained as auxiliary evaluation features in the multi-dimensional hand function feature matrix X, without participating in the S_stage calculation.

[0104] Wherein, the resistance peak F_peak_resist is the curve F_resist( The maximum value of F_resist(t) over the entire angular range. Stiffness coefficient K_stiff = ΔF / Δ , for the curve F_resist( In the linear ascending segment, F_resist is relative to The non-negative fitted slope, Δ For the curve F_resist( The non-negative extension angular displacement difference of the linear ascending segment, ΔF is the curve F_resist( The difference in resistance F_resist(t) during the linear ascending segment; the linear segment selection rules, goodness of fit, and effective angular displacement range are saved with the rule base version. During passive finger extension, the reference resistance value and its dispersion in the resting state are calculated in real time; only when the spasticity capture judgment threshold, duration condition, and acceptance criteria have been determined and frozen by independent calibration data under the same loading protocol, is the value corresponding to the first satisfaction of the judgment condition of F_resist(t) calculated. The spasm capture angle θ_catch is determined; θ_catch is not output if the calibration is not completed. The hysteresis loop area refers to F_resist( during the passive extension and passive rebound phases of the finger). The area enclosed by the curves. The above F_peak_resist, θ_catch, and hysteresis loop areas are generated only when the corresponding signal quality is acceptable, and are used for characteristic deviation calculation and longitudinal comparison. AUC and K_stiff are only used for relative comparison under the same loading position, direction, velocity range, and estimated angle caliber, and are not expressed as direct measurements of MCP joint torque, intrinsic tissue stiffness, or clinical muscle tone.

[0105] In this embodiment of the application, F_resist( during the passive stretching phase) The curve (rising curve) and the F_resist (passive rebound phase) When the curves (descending curves) are plotted in the same coordinate system, the area of ​​the closed region enclosed by the two curves is the area of ​​the hysteresis loop. Figure 5 As shown.

[0106] Optionally, when enabling a muscle tone grading mapping model pre-calibrated based on the modified Ashworth scale MAS, the model adopts a pre-determined model type and input fields, and the model parameters and version are fixed after independent verification that the preset requirements are met.

[0107] The calibration data of the model were labeled by clinicians according to the MAS classification (0, 1, 1+, 2, 3, 4), and the same data collection and feature extraction process as the passive stretching test of this application was used.

[0108] Calculate AUC and K_stiff for each calibrated sample, and use the MAS level given by the clinician as the target output; A predetermined statistical classification or regression model is used to fit the mapping relationship from (AUC, K_stiff) to MAS level; after the model is independently validated, the model type, parameters and version are fixed.

[0109] In practical use, the AUC and K_stiff extracted from the passive extension segment of the fingers of stroke patients are input into the muscle tone grading mapping model with fixed parameters and version, and the output is the muscle tone quantitative score T_MAS. T_MAS can be a discrete grade or a continuous predicted value, and is only used as an auxiliary evaluation result in addition to Brunnstrom staging, and is marked in the output as a model predicted value; T_MAS is not generated when the model is not enabled.

[0110] Step 404: For each test action segment, extract the degree of muscle activation and the synergistic contraction rate of the antagonist muscle; for active action segments, further extract the time difference Δt=|t_EMG_on-t_motion_on between the onset time of electromyographic activation t_EMG_on and the onset time of movement t_motion_on to obtain the corresponding electromyographic heterogeneous feature vector.

[0111] For each test action segment, the degree of muscle activation, the rate of antagonistic muscle synergistic contraction, and the degree of muscle fatigue are extracted, specifically including: Step 4041: For each test action segment, the preprocessed sEMG signals from the flexor side EMG acquisition channel F and the extensor side EMG acquisition channel E are subjected to full-wave rectification and low-pass envelope processing to obtain EMG_env,F(t) and EMG_env,E(t).

[0112] Step 4042, for each path c∈{F, E}, based on the electromyographic envelope EMG_env,c(t), calculate the integrated root mean square value of electromyography RMS_c according to the following formula: RMS_c = ; Where T = t_end - t_start represents the duration of each test action segment, t ∈ [t_start, t_end].

[0113] Step 4043: In the MVIC calibration test, an analysis window of the same duration as the action segment to be evaluated is used to calculate RMS_MVIC,c. The calculation is performed when the corresponding MVIC reference value (determined based on baseline data from healthy individuals) is greater than the minimum effective value and the calibration signal quality is acceptable. %RMS_MVIC,c=(RMS_c / RMS_MVIC,c)×100%; %RMS_MVIC,c is the standardized muscle activation level used in the staging rules. If RMS_MVIC,c is invalid or the calibration signal quality is unqualified, %RMS_MVIC,c is not calculated, the reason for the missing value is recorded and a retest is prompted until RMS_MVIC,c is valid or the calibration signal quality is qualified.

[0114] Raw electromyographic amplitudes (e.g., μV) are highly susceptible to significant interference from non-experimental factors such as electrode position, skin impedance, subcutaneous fat thickness, muscle volume, and joint angle. Without standardization, individual differences or acquisition errors can easily be misinterpreted as differences in actual muscle activation. By correlating RMS with an individual's maximum contractile capacity, baseline bias is eliminated, making data from different subjects, different muscles, and different experimental time points directly comparable.

[0115] Step 4044: Within the same effective action segment, normalize the two electromyographic envelopes according to their respective effective reference values ​​to obtain E_F(t) and E_E(t), and calculate the co-contraction rate CR of the flexor and extensor muscle groups according to the following formula: CR=[∫min(E_F(t),E_E(t))dt / ∫max(E_F(t),E_E(t))dt]×100%; The quantitative rule base determines whether the hardware channels are permanently named as active muscle channels or antagonistic muscle channels based on the active or antagonistic roles of the flexor side electromyography acquisition channel F and the extensor side electromyography acquisition channel E in the current action.

[0116] When the denominator is not greater than the preset minimum effective value, any path is invalid, or there are insufficient effective sampling points, CR is recorded as missing.

[0117] In another exemplary embodiment of this application, in step 404 above, for each test action segment, while extracting the degree of muscle activation and the antagonistic muscle synergistic contraction rate, at least one of the median frequency sequence MDF_c(t), the linear fitting slope Slope_MDF,c of MDF_c(t), and the fatigue state flag is also extracted.

[0118] Accordingly, step 404 above further includes steps 4045 to 4046, wherein: Step 4045: For any electromyography (EMG) acquisition channel c, c∈{F, E}, in the flexor side EMG acquisition channel F and the extensor side EMG acquisition channel E, perform short-time Fourier transform on the preprocessed sEMG signal in the EMG acquisition channel c that satisfies the shortest duration contraction condition, extract the median frequency sequence MDF_c(t), and calculate the linear fitting slope Slope_MDF,c of MDF_c(t).

[0119] The preset shortest duration contraction condition refers to the following: the time during which the electromyographic envelope of the electromyographic acquisition channel c is continuously higher than the preset electromyographic activation threshold reaches the preset shortest analysis time, and the number of effective sampling points within the time period that is continuously higher than the preset electromyographic activation threshold meets the preset requirements; the electromyographic activation threshold, the preset shortest analysis time, and the number of effective sampling points are determined based on calibration data under the same muscle group, action, sampling rate, and analysis window conditions.

[0120] Step 4046, determine the fatigue state flag: If the sEMG signal quality of the EMG acquisition channel c is qualified and the EMG envelope meets the shortest continuous contraction condition, check whether the Slope_MDF,c of the EMG acquisition channel c is not greater than the negative fatigue threshold F_th,c corresponding to the EMG acquisition channel c; if so, set the fatigue state flag of the EMG acquisition channel c to 1, indicating that muscle fatigue has been detected; otherwise, set the fatigue state flag of the EMG acquisition channel c to 0, indicating that muscle fatigue has not been detected; where F_th,c is determined based on baseline data of healthy people obtained under the same muscle group, action, sampling conditions and analysis window, and the parameter version is recorded after calibration; If the sEMG signal quality of the electromyography acquisition channel c is unqualified or the electromyography envelope does not meet the preset shortest duration contraction condition, the fatigue state flag bit of the electromyography acquisition channel c will not be generated, and the fatigue state flag bit will be marked as uncalcifiable. When baseline data for healthy individuals is insufficient or calibration is incomplete, only MDF_c(t) and Slope_MDF,c are output, and the fatigue status flag of the electromyography acquisition channel c is not generated.

[0121] In this embodiment, MDF_c(t), Slope_MDF,c, and fatigue state flag are used as auxiliary fatigue evaluation fields for fatigue evaluation.

[0122] In another exemplary embodiment of this application, step 104 specifically includes steps 501 to 503, wherein: Step 501: Align the heterogeneous feature vectors of each test action segment according to the test batch, action type, and action order.

[0123] The action type refers to the type of test instruction for each test action segment, including active fist clenching, active finger extension, thumb pinching against the other four fingers in sequence, thumb-index finger side pinching, cylindrical grasping, ball-shaped grasping, and passive finger extension. Each segment is labeled with a corresponding action type. Multiple tests on the same patient are arranged according to the test batch identifier, action type, and the order of the action within the batch. Test action segments with correct action labels, complete start and end points, and acceptable signal quality for corresponding feature calculations are defined as valid test action segments.

[0124] Step 502: Within any action type, divide the field into continuous field and discrete field; retain the original value of the discrete field without calculating the Z-score; only for the continuous field f that is configured to be normalizable in the quantization rule base, has qualified data quality and a healthy baseline standard deviation σ_health,f>0, calculate the Z-score using the health mean μ_health,f and standard deviation σ_health,f of this action type, while retaining the original feature value, baseline version and reason for not being able to calculate.

[0125] In this embodiment of the application, the Z-score normalization formula is z = (x_raw -μ_health) / σ_health.

[0126] Where x_raw represents the raw feature value of the test action segment, and z represents the corresponding Z-score standardized value.

[0127] The health baseline is defined as: a statistical distribution reference of the feature values ​​of each dimension extracted from a pre-collected group of healthy subjects matched with the patients to be evaluated in terms of demographic characteristics such as age, sex, and dominant hand, using the same equipment, standard testing procedures and processing procedures; the minimum sample requirement is determined by a pre-defined statistical analysis protocol.

[0128] Structure and query dimensions of the baseline database for healthy subjects: The healthy subject baseline database stores health reference statistics corresponding to each characteristic, based on the necessary matching factors, action type, device model, action protocol, and processing flow version determined by a pre-defined statistical analysis protocol. These health reference statistics include central location, dispersion, and / or reference interval, according to the data type and distribution characteristics of the corresponding characteristic. The necessary matching factors may include one or more of age, gender, and user hand type; stratification factors, stratification boundaries, minimum sample requirements, stratification merging conditions, and statistic types are determined by the pre-defined statistical analysis protocol and recorded along with the healthy subject baseline database version.

[0129] In one embodiment, the baseline database of healthy subjects can be stratified according to age, gender, dominant hand, and movement type, wherein different movement types store corresponding health reference statistics. When the effective healthy samples of a certain stratum are insufficient to meet the minimum sample requirement specified in the statistical analysis scheme, the stratum is merged with adjacent or statistically similar strata according to the pre-defined stratification merging rules, or no health reference statistics are generated for that stratum.

[0130] For continuous features with acceptable data quality, a health reference standard deviation greater than zero, and suitable for description using mean and standard deviation, store the population mean μ_health and population standard deviation σ_health for Z-score standardization in step 502; for features that are not suitable for representation using mean and standard deviation, store the median, interquartile range, quantile reference interval, or other health reference statistics specified in the statistical analysis scheme, without requiring Z-score calculation.

[0131] When making longitudinal comparisons of the same patient at different test time points, the same equipment model, action protocol, processing procedure, and baseline database version for healthy subjects should be used; after the database version is updated, the results of the old version should not be directly mixed and compared with the results of the new version.

[0132] Step 503: Determine the representative values ​​of each dimension of the features for each action type according to the preset aggregation method, and concatenate them according to the preset action type order and feature field order to obtain the multi-dimensional hand function feature matrix X.

[0133] The representative value refers to a single numerical value calculated by a preset aggregation method from multiple valid test action fragments of the same test batch, the same action type, and the same feature dimension; unless otherwise specified, the preset aggregation method is the median. The feature fingerprint of an action type is formed by arranging the representative values ​​of each feature dimension in the preset feature field order; concatenation refers to arranging the feature fingerprints of each action type in a field-level order according to the preset action type order, not simply adding values ​​of different dimensions. When the quantization rule base has a specific aggregation method for the AUC mean, K_stiff maximum value, FII robust representative value, Jerk_norm arithmetic mean, or action success rate, it is directly calculated according to the feature values ​​of the corresponding valid test action fragments using that aggregation method. The action success rate is the ratio of the number of successful fragments to the total number of valid attempt fragments in the same test batch and the same composite grasping action type; ρ, stable holding time, and Δt are all written as independent fields in the multi-dimensional hand function feature matrix X. Each aggregation field simultaneously records the number of valid repetitions, the number of missing values, and the aggregation operator version.

[0134] Among them, the multi-dimensional hand function feature matrix X retains the original feature values, Z-score standardized values ​​and their original physical meanings of each dimension, without dimensionality reduction or compression; fields that do not have valid test action segments are marked as missing and prompted for retesting, and are not replaced with the value 0; X includes features used for Brunnstrom staging condition comparison in step 601, and may also include auxiliary evaluation features that do not participate in S_stage calculation.

[0135] Both the baseline database for healthy subjects and the quantitative rule base are version-managed. Longitudinal comparisons of the same patient are limited to the same equipment, action protocols, processing procedures, and database version; after a version update, old results are not directly mixed with new results. For continuous fields, Z-scores are calculated only when σ_health,f>0 and the distribution is expressed using mean-standard deviation representation; for discrete fields, σ_health,f=0, insufficient references, or unqualified data quality, Z-scores are not calculated, and the reasons are recorded.

[0136] In another exemplary embodiment of this application, step 105 specifically includes steps 601 to 604, wherein: Step 601: Based on the multi-dimensional hand function feature matrix, detect whether the conditions of various features of each Brunnstrom stage are met; among them, various features include three categories: kinematic features, mechanical features and electromyographic features.

[0137] Step 602, for each feature class of each Brunnstrom stage: If the required features for this type of feature used for Brunnstrom staging determination are complete, and the required features satisfy the corresponding feature conditions of Brunnstrom staging, then the indicator function of this type of feature takes the value of 1. If the required features for the Brunnstrom staging determination are complete, but at least one of the required features does not meet the corresponding feature conditions for the Brunnstrom staging, then the indicator function of the feature class is 0. If any required feature is missing from the category of features used for Brunnstrom staging determination, the output data will be incomplete and a retest prompt will be given. The indicator function value for the category of features will be determined based on the above situation after the required feature is found to be complete through retesting.

[0138] Among them, the various features required for each Brunnstrom phase division, as well as the required features among each feature, are pre-configured and their versions are recorded by the quantitative rule base.

[0139] Step 603: For each Brunnstrom stage, based on the indicator function value for each feature category, calculate the score for each Brunnstrom stage according to the following formula: S_stage =Σ_d(w_d×I(feature_d)); Where S_stage represents the score of the Brunnstrom stage, w_d represents the weight of the d-th feature corresponding to the Brunnstrom stage, d∈{1,2,3}, feature_d represents the d-th feature of the Brunnstrom stage, and I(feature_d) represents the value of the indicator function of the d-th feature of the Brunnstrom stage.

[0140] In this embodiment, the w_d corresponding to different Brunnstrom stages is different. Kinematic features are defined as Class 1 features, mechanical features as Class 2 features, and electromyographic features as Class 3 features. For any Brunnstrom stage s, the w_d in the formula calls the weight w_s,d of the d-th feature corresponding to that stage from the quantization rule base, and w_s,d≥0; when normalized weight configuration is used, Σ_d(w_s,d)=1 is satisfied.

[0141] Step 604: Based on the scores of each Brunnstrom stage, output the Brunnstrom stage assessment reference results for the stroke patient: If all features involved in the score calculation for any Brunnstrom phase are complete, and the scores for all Brunnstrom phases are 0, a message indicating that the current quantitative rule base phase conditions have not been met will be output.

[0142] If the highest and second-highest scores of all Brunnstrom periods correspond to adjacent Brunnstrom periods, and the difference between the highest and second-highest scores satisfies the transition judgment conditions pre-set in the quantitative rule base, and at least one pre-specified key feature exhibits a cross-period fulfillment state between the two adjacent Brunnstrom periods, then the transition period prompt for the two adjacent Brunnstrom periods is output, and the conflict feature (the key feature exhibiting a cross-period fulfillment state between the two adjacent Brunnstrom periods) is recorded. If the calibration of the transition judgment conditions is not completed, the transition period is not automatically output based on the score difference; only the scores and conflict features of the two adjacent Brunnstrom periods are output for manual review.

[0143] Key features refer to pre-specified judgment fields in the quantitative rule base used to distinguish a Brunnstrom stage from its adjacent stages, and are not determined by the weight of a single feature. The key features for each stage are determined from the aforementioned existing conditions: Stage I includes the maximum ROM of the MCP joints of fingers 2-5, the mean AUC, and the %RMS_MVIC standardized activation of RMS; Stage II includes the correlation coefficient ρ of the five-finger joints, the maximum K_stiff value, and CR; Stage III includes the maximum ROM of the MCP joints of fingers 2-5, ρ, the mean AUC, and CR; Stage IV includes the representative FII value and CR; Stage V includes Jerk_norm_mean, the success rate of compound grasping or pinching movements, and the electromyographic-motor onset time difference; Stage VI includes the relative health interval of Jerk_norm_mean, the relative health reference value of TAM_total, the relative health interval of the mean AUC and the maximum K_stiff value, and CR. Key features are only used for interpreting judgment conflicts and providing retesting prompts in the transition between adjacent stages and do not change the calculation of the three-category feature matching scores for S_stage.

[0144] If the above conditions are not met and only one Brunnstrom stage achieves the highest score, the Brunnstrom stage with the highest score will be output as the candidate stage.

[0145] If the above conditions are not met and two or more Brunnstrom stages achieve the highest score, or if the two Brunnstrom stages corresponding to the highest score are not adjacent, output a message indicating a conflict between the two Brunnstrom stages with the highest scores and prompt for retesting or manual review.

[0146] In another exemplary embodiment of this application, the conditions of various features of each Brunnstrom stage are stored in a quantitative rule base constructed based on the Brunnstrom recovery stage theory.

[0147] Wherein, for Brunnstrom Stage I (flaccid stage), the kinematic feature conditions include that in the active fist-clenching motion segment and the five-finger stretching motion segment of the stroke patient, the maximum value of the range of motion of the MCP joints of the 2nd to 5th fingers is less than the preset Stage I range of motion threshold θ_I,ROM; the mechanical feature conditions include that the mean AUC value < T1, where T1 represents the preset lower limit of the AUC threshold; the sEMG feature conditions include that the %RMS_MVIC normalized activation levels of the target flexor side and extensor side channels are both lower than the preset Stage I sEMG activation threshold θ_I,EMG. The mean AUC value shall be the arithmetic mean of the AUC of the passive stretching motions of all tested fingers (the 2nd to 5th fingers). In the standard test protocol, passive stretching tests are performed sequentially on the index finger, middle finger, ring finger, and little finger, therefore, the mean AUC value is the mean of the AUC of the index finger, middle finger, ring finger, and little finger during the passive stretching motions. The target muscle groups refer to the major flexor muscle groups and extensor muscle groups of the forearm, the flexor muscle groups refer to the regions where the flexor digitorum superficialis (FDS) and flexor digitorum profundus (FDP) are located, and the extensor muscle group refers to the region where the extensor digitorum communis (EDC) is located.

[0148] θ_I,ROM, T1 and θ_I,EMG are determined by calibration data using the same equipment, motion protocol and data processing flow, and are written into the quantitative rule base and the version is recorded after meeting the preset verification requirements; when the calibration is not completed, the corresponding indicators are output as continuous evidence fields, and no fixed values are used for staging determination.

[0149] For Brunnstrom Stage II (initial appearance of synergic movement), the kinematic feature conditions include that the five-finger synergic movement correlation coefficient ρ is greater than the preset Stage II correlation threshold θ_II,ρ; the mechanical feature conditions include that the mean AUC value ∈ [T1, T2], where T2 represents the preset upper limit of the AUC threshold, and the maximum value of the stiffness coefficient K_stiff is greater than the product of K_baseline and the preset multiplier λ_II,K, where K_baseline is the mean value of the stiffness coefficient K_stiff of healthy people in the same age group; the sEMG feature conditions include that CR is greater than the preset Stage II co-contraction ratio threshold θ_II,CR. The maximum K_stiff value shall take the maximum value of the passive stretching stiffness coefficients of all tested fingers (the 2nd to 5th fingers), this is because Brunnstrom Stage II determination focuses on whether spasticity (increased muscle tone) exists, and the maximum value can best reflect the degree of abnormal muscle tone.

[0150] θ_II,ρ, T1, T2, λ_II,K and θ_II,CR are determined by the calibration data of Stage II and adjacent stages using the same equipment, motion protocol and data processing flow, and are written into the quantitative rule base and the version is recorded after meeting the preset verification requirements; T1 and T2 are not directly determined by using the healthy mean value plus a fixed standard deviation multiple that has not been verified by the project. CR participates in the determination only when both two synchronous sEMG signals are valid.

[0151] For Brunnstrom stage III (spasticity and synkinesis phase), the kinematic characteristics are defined as follows: in the active fist clenching and finger extension segments of the stroke patient, the maximum range of motion of the MCP joints of the 2nd to 5th fingers is within the preset interval [θ_III,ROM,L, θ_III,ROM,U], and the correlation coefficient ρ of the five fingers' synkinesis is within the preset interval (θ_III,ρ,L, θ_III,ρ,U], that is, the five fingers' movements still exhibit synkinesis coupling characteristics; the mechanical characteristics are defined as follows: the mean AUC is within the preset interval [T_III,AUC,L, T_III,AUC,U], and the electromyographic characteristics are defined as follows: %RMS_MVIC is greater than the preset threshold θ_III,EMG, and CR is within the preset interval [θ_III,CR,L, θ_III,CR,U].

[0152] The aforementioned thresholds and ranges are determined by the calibration data of Phase II, Phase III, and Phase IV using the same equipment, action protocol, and data processing flow, and are written into the quantization rule base and recorded after the preset verification requirements are met; CR only participates in the judgment when both synchronous sEMG signals are valid.

[0153] For Brunnstrom stage IV (partial dissociation phase), the kinematic characteristics are determined by the following conditions: the representative value of the dissociation index FII_rep in the preset selective movement segment is greater than the preset stage IV dissociation threshold θ_IV,FII; the mechanical characteristics are determined by the following conditions: the mean AUC is less than the preset stage IV AUC threshold T_IV,AUC; and the electromyographic characteristics are determined by the following conditions: CR is located in the preset interval [θ_IV,CR,L, θ_IV,CR,U]. FII_rep is calculated from multiple valid repetitions of the same selective single-finger movement according to the aggregation method specified in step 503. Unless otherwise specified, the median is used, and the maximum value of a single movement is not directly used. FII is only used for the thumb-to-four-finger pinching and thumb-index-finger lateral pinching movements explicitly set in the movement standard protocol.

[0154] If the conditions for the kinematic characteristics of the Brunnstrom stage IV label are met, it indicates the presence of independent motion components, thus satisfying the kinematic separation condition.

[0155] For the Brunnstrom V phase label (separate motor development phase), the conditions for kinematic features include the arithmetic mean of the normalized Jerk values ​​of effective active movement segments of the same movement type, Jerk_norm_mean, which satisfies the V phase smoothness condition preset in the quantification rule base for this movement type; where, the smaller the Jerk_norm value, the smoother the movement trajectory under this definition; the conditions for mechanical features include: the success rate of compound grasping action is not lower than the preset threshold θ_V,success, where the success rate of compound grasping action is defined as: in a single test, multiple attempts are made on the same compound action, and each attempt is considered successful when the following success conditions are met simultaneously: (a) the complete movement trajectory of the action is completed; (b) the peak grip force F_peak,grip or F_peak,pinch reaches the preset threshold θ_V,force,a; (c) the hold time T_hold reaches the preset threshold θ_V,hold,a; the conditions for electromyographic features include: the electromyographic activation timing and the movement phase matching error Δt are located in the preset interval R_V,Δt, and CR is located in the preset interval R_V,CR.

[0156] For compound gripping or pinching actions that include a holding phase in the action protocol, when the corresponding action segment contains the action start point determined in step 3012, the target posture arrival time t_arr determined in step 3013, and the action end point, and the necessary kinematic and mechanical signals from the action start point to the action end point are complete, the complete motion trajectory of the action is determined. Within the time range from the target posture arrival time t_arr to the action end point, the stable holding time T_hold is determined as the duration of the longest continuous time interval where the relative angular velocity magnitude corresponding to the main driving finger does not exceed the termination threshold ω_end and the mechanical characteristics satisfy the longest continuous time interval. The longest continuous time interval is the longest continuous time interval of the mechanical signal holding condition, which is pre-configured by the action rules corresponding to the current action type.

[0157] If the start point of the action, the arrival time of the target posture (t_arr), or the end point of the action cannot be identified, or the required kinematic or mechanical features are incomplete, the action is not considered to have completed a complete trajectory, and T_hold is marked as missing. If it is impossible to determine whether a complete trajectory has been completed or T_hold is marked as missing, no success or failure determination is made for that attempt.

[0158] In a test, the success rate is calculated by comparing the number of successful movements among all similar movement segments to the total number of movements. Attempts deemed invalid due to incorrect movement labels, incomplete segments, or substandard quality of key signals are excluded from the denominator, and the reasons for invalidity are recorded. Attempts where data is missing, making it impossible to determine whether a complete motion trajectory has been completed or to calculate T_hold, are treated as invalid attempts and excluded from the denominator of the success rate, with the corresponding reasons for invalidity recorded. Attempts where the required signal is complete and of acceptable quality, and can meet the above success criteria, but fail to achieve any of the success criteria—complete motion trajectory, peak force, or T_hold—are judged as failed attempts and included in the denominator of the success rate.

[0159] The V-phase Jerk_norm condition, θ_V,success, θ_V,force,a, θ_V,hold,a, R_V,Δt, and R_V,CR are determined by the V-phase and adjacent phase calibration data using the same equipment, action protocol, and data processing flow. After meeting preset verification requirements, these are written into the quantization rule base and the version is recorded. Here, 'a' represents the current composite grip or pinch action type. The R_V,Δt used for V-phase determination is the R_Δt comparison interval configured for Brunnstrom V-phase in the quantization rule base. Before the corresponding calibration is completed, Jerk_norm, action success rate, peak force, T_hold, and Δt are output as continuous evidence fields and are not used for phase determination with fixed values. CR only participates in the determination when both synchronous sEMG signals are valid.

[0160] The error in the timing of electromyographic activation and the matching of motor phases is defined as: For any test action segment, for the electromyographic envelope signal EMG_env,c(t) of the target muscle group specified by the action rules, an activation threshold θ_EMG,on and a duration τ_EMG,on are set, and the muscle activation time t_EMG_on is detected; wherein t_EMG_on is the starting time of the time period when EMG_env,c(t) first continuously exceeds θ_EMG,on and continues to reach τ_EMG,on. For any test motion segment, the motion initiation time t_motion_on is detected based on the relative angular velocity modulus |ω_rel,i(t)| corresponding to the main driving finger of the current test motion obtained in step 209; the initiation time when the relative angular velocity modulus first continuously exceeds ω_start and reaches the first target time is t_motion_on. The time difference between the two is Δt = |t_EMG_on - t_motion_on|, which is the matching error between the electromyographic activation sequence and the motion phase.

[0161] When Δt falls within the pre-defined coordination comparison interval R_Δt in the quantization rule base, and this condition is met within the active test action type specified in the quantization rule base, it can be used as evidence field for electromyographic activation and motor phase coordination. The θ_EMG,on, τ_EMG,on, R_Δt, and the action type involved in the judgment are determined by calibration data using the same equipment, muscle group location, action protocol, and data processing flow. After meeting the pre-defined verification requirements, these values ​​are written into the quantization rule base and the version is recorded. Before calibration is complete, only continuous Δt values ​​are output, and fixed coordination judgments are not made based on these values.

[0162] For Brunnstrom Phase VI (coordinated movement recovery phase), the conditions for kinematic characteristics include: Jerk_norm_mean and TAM_total are located in the healthy reference intervals R_VI,Jerk and R_VI,TAM, respectively, which match the same movement type, equipment, and data processing flow; the conditions for mechanical characteristics include: the mean AUC and the maximum K_stiff are located in the healthy reference intervals R_VI,AUC and R_VI,K, respectively, under the same passive loading protocol; the conditions for electromyography characteristics include: CR is located in the preset Phase VI reference interval R_VI,CR.

[0163] The health reference values ​​are obtained from healthy subjects using the same movement types, joint composition, movement segment intervals, ROM calculation methods, and feature aggregation methods as the subjects being evaluated. The matching factors, statistical representation methods, and minimum sample requirements for the health reference intervals are determined by a pre-established statistical analysis scheme and recorded in the database version. Fixed percentages or mean ± standard deviation multiples that have not been validated by the project are not directly applied. CR is only used in the determination when both synchronous sEMG signals are valid.

[0164] In another exemplary embodiment of this application, the Brunnstrom staging feature conditions in the quantitative rule base are divided into two categories according to the comparison caliber: the first category is absolute physical threshold conditions, which refer to conditions that are compared using fixed thresholds with physical units such as angle and time, including ROM, electromyography-motor initiation time difference and stable holding time; the second category is dimensionless, proportional or health reference conditions, including correlation coefficient ρ, co-contraction rate CR, dissociated movement index FII, %RMS_MVIC, movement success rate, and conditions of AUC, K_stiff, Jerk_norm and TAM_total relative to matching health reference statistics or reference intervals.

[0165] When the system performs the judgment, the processing module compares features according to the feature fields, actions or joint objects, aggregation methods, comparison criteria, and comparison directions recorded for each phase rule: (1) For the absolute physical threshold condition: directly compare the original feature value of the corresponding field in the multidimensional hand function feature matrix X with the absolute threshold; optionally, when the health reference standard deviation is greater than 0 and the field is represented by mean-standard deviation, the original threshold is converted into Z-score boundary value using the same health reference parameters and then compared with the corresponding Z-score standardized value. (2) For dimensionless, proportional or health reference conditions: extract the corresponding original calculated value, proportional value or standardized value from the multidimensional hand function feature matrix X, and compare them according to the threshold, reference interval and comparison direction recorded in the quantification rule library.

[0166] The baseline database of healthy subjects was constructed using characteristic data of healthy subjects collected with the same equipment, standard testing protocols, and data processing workflows. Necessary matching factors, stratification methods, minimum sample requirements, types of statistics, and methods for calculating reference intervals were determined by a pre-defined statistical analysis protocol. Each updated database version was re-validated; the same version was used for longitudinal comparisons of the same patient, and results from older versions were not directly mixed with those from newer versions.

[0167] Taking the conversion of the stage I ROM threshold as an example: when the quantitative rule base has determined the stage I ROM threshold θ_I,ROM based on calibrated data, the original ROM feature values ​​of the corresponding action and joint of the patient can be directly compared with θ_I,ROM. Optionally, when σ_ROM>0 and the distribution is expressed as mean-standard deviation, the threshold is converted into a Z-score cutoff value according to z_th=(θ_I,ROM-μ_ROM) / σ_ROM, and the z_ROM of the corresponding ROM of the patient is calculated using the same μ_ROM and σ_ROM. The comparison of the original value and the comparison of the Z-score are the same way of implementing the determined threshold at different scales. The Z-score conversion itself does not constitute evidence that the threshold is effective for staging.

[0168] For the matching error Δt between the electromyographic activation timing and the motion phase, the staging rule comparison is performed only when the corresponding comparison interval R_Δt has been determined in the quantization rule base; the original value of Δt can be compared directly, or the determined boundary can be converted into a Z-score before comparison if the Z-score calculation conditions are met. Both implementations use the same actions, matching conditions, and database version.

[0169] The quantization rule base stores for each rule the feature fields, action or joint object, aggregation method, applicable data quality conditions, comparison direction, threshold or reference range, parameter source, verification status, and effective version. Specific thresholds not supported by calibration are not retained after Z-score transformation.

[0170] In another exemplary embodiment of this application, step 105 further includes: If the calculated score S_stage for all Brunnstrom stages is 0, it means that the stroke patient's various characteristics have not fallen within any of the preset stage threshold ranges. In addition to outputting a prompt that the stage threshold has not been reached, it is recommended to expand the baseline database or conduct a manual review.

[0171] In this embodiment of the application, the baseline database area includes a healthy subject baseline database and a Brunnstrom quantitative rule base. The healthy subject baseline database provides Z-score standardized reference mean and standard deviation, and the Brunnstrom quantitative rule base stores the conditional thresholds required for staging. Expanding the baseline database mainly refers to increasing the sample size of healthy subjects.

[0172] In another exemplary embodiment of this application, for ease of clinical use, step 105 further includes: For continuous fields configured in the quantitative rule base as normalizable, with acceptable data quality, and a health reference standard deviation σ_f > 0, the deviation_f of each dimension feature is calculated based on the health mean μ_f and standard deviation σ_f of each dimension feature given by the health control database: Deviation_f = (x_f -μ_f) / σ_f.

[0173] The Z-score of each feature within the healthy distribution is used as the deviation. The larger the absolute value of the deviation, the more significant the difference from the healthy state. Deviation is not calculated for discrete fields, σ_f=0, insufficient reference, or distributions where mean-standard deviation representation is not applicable; the reasons are recorded.

[0174] In another exemplary embodiment of this application, step 105 further includes: The output is a structured data report containing calculable deviations of features across various dimensions, the Brunnstrom staging matching score S_stage for stroke patients, Brunnstrom staging assessment reference results, evidence fields, reasons for missing data, retesting prompts, the version of the quantitative rule base, and the version of the health reference database. When a muscle tone grading mapping model with fixed parameters and version is enabled, T_MAS and the model version are also output; this field is not generated when it is not enabled. When the multi-dimensional hand function feature matrix X contains valid auxiliary evaluation features, the structured data report also outputs the corresponding auxiliary evaluation features, which may include auxiliary movement velocity indices ω_max, F_peak_resist, θ_catch, and hysteresis loop area, etc. The structured data report also outputs valid iEMG, %iEMG_MVIC, the corresponding deviation, and the change relative to the patient's previous assessment time points, and marks them as auxiliary evaluation features that do not participate in the S_stage calculation.

[0175] Because stroke patients' hand movements are often characterized by low speed and small angles, traditional IMUs are prone to low-frequency integral drift. Furthermore, fixed-threshold segmentation algorithms are prone to missegmentation during resting noise and the transition between the start and end of movements, leading to distortion in subsequent feature calculations. Therefore, this application embodiment uses adaptive complementary filtering to fuse the angle calculated by the IMU with the bending reference angle from the flexible bending sensor to reduce the impact of sensor drift on joint angle calculations. Based on the relative angular velocity modulus |ω_rel,i(t)| corresponding to the target finger or the primary driving finger pre-specified according to the current test action standard protocol, combined with the start threshold ω_start, end threshold ω_end, and duration constraints, automatic action cycle segmentation is performed to reduce baseline noise interference and obtain time-aligned test action segments. This improves the temporal consistency and reliability of low-speed, small-amplitude hand movement feature calculations.

[0176] Because Brunnstrom staging is a clinical qualitative description while sensor output is a continuous physical quantity, traditional methods rely on subjective scoring based on expert experience. Furthermore, the feature spaces of different stages (such as the synkinesis and spasticity peak in stages II / III, and separation and coordination in stages V / VI) highly overlap, lacking an automated and interpretable quantitative decision-making mechanism. Therefore, this application constructs a quantitative rule base and conflict resolution logic, introduces the Separate Movement Index (FII) to quantify independent movement ability, uses the weighted scoring formula S_stage=Σ_d(w_d×I(feature_d)) for multimodal evidence fusion, and sets transitional judgment conditions pre-configured, calibrated, and frozen versions from the quantitative rule base. Continuous engineering features are compared with the preset conditions in the quantitative rule base, and stage candidates are determined by combining FII and multimodal weighted scores. Transitional period prompts are output for cases where adjacent stage scores are close and key features are satisfied across stages. The evaluation results include deviations of features in each dimension for easy verification. This forms interpretable and verifiable matching reference information between clinical qualitative staging and engineering quantitative signals, providing interpretable automated decision-making.

[0177] Because there are significant individual differences in resting muscle tone, maximal voluntary contraction capacity, and fatigue levels among different stroke patients, simply using a fixed absolute threshold can lead to inaccurate cross-patient assessments and incomparable cross-batch data, making it difficult for the system to achieve long-term stable operation and cross-batch tracking in real rehabilitation scenarios. Therefore, this application's embodiments use %RMS_MVIC normalization and calculate Z-scores or determine relative thresholds based on health baseline statistical parameters corresponding to the matching conditions, test actions, joints, and characteristics of the corresponding subjects. These parameters include health reference parameters such as T1, T2, and K_baseline. Matching conditions and health baseline database versions are kept consistent across different assessment time points for the same patient. This processing reduces the impact of baseline differences on feature comparisons, facilitates objective and quantitative tracking of rehabilitation progress, and improves the consistency of longitudinal efficacy comparisons and deviation analysis of various dimensions of characteristics.

[0178] The method for evaluating the staged hand movement rehabilitation of stroke patients in this application simultaneously acquires kinematic signals, mechanical signals, and sEMG signals through a wearable data glove integrating multiple sensors. It preprocesses different modal signals, segments movement fragments, and extracts features. Based on statistical parameters matching the healthy baseline, it calculates corresponding standardized values ​​and organizes the original feature values ​​and standardized values ​​according to movement type and feature fields to form a multi-dimensional hand function feature matrix. Finally, based on the quantitative rule base of Brunnstrom's recovery staging theory, it uses a weighted scoring mechanism of kinematic, mechanical, and electromyographic features to output Brunnstrom staging, staging scores, and deviations of features in each dimension. It also outputs transition period prompts and conflict features when adjacent stage scores are close or key features cross-stage satisfaction.

[0179] In another exemplary embodiment of this application, the processing module 2 described above employs an embedded processor, which is used to execute the above-described method for evaluating the phased hand movement rehabilitation of stroke patients.

[0180] In this embodiment of the application, the architecture is suitable for portable scenarios such as home / community rehabilitation.

[0181] In another exemplary embodiment of this application, such as Figure 6 As shown, the above-mentioned processing module 2 includes a main control acquisition unit 21 and a host computer 22.

[0182] The main control acquisition unit 21 is used to acquire kinematic signals, mechanical signals and sEMG signals synchronously acquired by the data glove system during the standardized hand function test, and to preprocess the acquired kinematic signals, mechanical signals and sEMG signals.

[0183] The host computer 22 is used to: extract features from each test action segment based on the preprocessing results to obtain heterogeneous feature vectors for each test action segment; wherein the features include kinematic features, mechanical features and / or electromyographic features, and the heterogeneous feature vectors include kinematic heterogeneous feature vectors, mechanical heterogeneous feature vectors and / or electromyographic heterogeneous feature vectors; based on the heterogeneous feature vectors of all test action segments, perform multi-dimensional feature fusion and time alignment to obtain a multi-dimensional hand function feature matrix; based on the multi-dimensional hand function feature matrix, perform Brunnstrom staging to obtain Brunnstrom staging assessment reference results for stroke patients, which serve as reference information for rehabilitation assessment.

[0184] In this embodiment, the architecture is suitable for fixed environments such as hospital rehabilitation departments.

[0185] In another exemplary embodiment of this application, the host computer 22 is further configured to: before extracting features from each test action segment based on the preprocessing results to obtain the heterogeneous feature vector of each test action segment, use an automatic action cycle segmentation algorithm based on fusion kinematic threshold to segment the multimodal raw data stream to obtain each test action segment.

[0186] In this embodiment of the application, an automatic action cycle segmentation algorithm based on fusion kinematic thresholds is used to segment the action of the multimodal raw data stream. For details on this, please refer to the description of the above method embodiment, which will not be repeated here.

[0187] In another exemplary embodiment of this application, the host computer 22 is further used for: If the calculated score S_stage for all Brunnstrom stages is 0, it means that the stroke patient's various characteristics have not fallen within any of the preset stage threshold ranges. In addition to outputting a prompt that the stage threshold has not been reached, it is recommended to expand the baseline database or conduct a manual review.

[0188] In another exemplary embodiment of this application, the host computer 22 is further used for: For continuous fields configured in the quantitative rule base as normalizable, with acceptable data quality, and a health reference standard deviation σ_f > 0, the deviation_f of each dimension feature is calculated based on the health mean μ_f and standard deviation σ_f of each dimension feature given by the health control database: Deviation_f = (x_f -μ_f) / σ_f.

[0189] For discrete fields, σ_f=0, insufficient reference, or distributions where mean-standard deviation representation is not applicable, deviation is not calculated, and the reason is recorded.

[0190] In another exemplary embodiment of this application, the host computer 22 is further used for: The output is a structured data report containing calculable deviations of features across various dimensions, the Brunnstrom stage score S_stage for stroke patients, Brunnstrom stage assessment reference results, evidence fields, reasons for missing data, retesting prompts, the version of the quantitative rule base, and the version of the health reference database. When a muscle tone grading mapping model with fixed parameters and version is enabled, T_MAS and the model version are also output; this field is not generated when it is not enabled. When the multi-dimensional hand function feature matrix X contains valid auxiliary evaluation features, the structured data report also outputs the corresponding auxiliary evaluation features, which may include the auxiliary movement velocity index ω_max; the auxiliary evaluation features may also include F_peak_resist, θ_catch, and the area under the hysteresis loop.

[0191] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data related to a phased hand motor rehabilitation evaluation method for stroke patients. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a phased hand motor rehabilitation evaluation method for stroke patients.

[0192] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0193] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0194] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0195] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0196] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0197] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0198] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0199] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for evaluating the phased hand motor rehabilitation of stroke patients, characterized in that, An evaluation system for phased hand motor rehabilitation in stroke patients is provided. This system includes a data glove system and a processing module. The data glove system is worn on the stroke patient's hand during standardized hand function testing to collect kinematic, mechanical, and sEMG signals from the stroke patient's hand during the standardized hand function testing process. The method for phased hand motor rehabilitation evaluation in stroke patients includes: During the standardized hand function test, the data glove system synchronously collects kinematic signals, mechanical signals, and sEMG signals. Preprocessing is performed on the acquired kinematic signals, mechanical signals, and sEMG signals; Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each test action segment; wherein, the features include kinematic features, mechanical features and / or electromyographic features, and the heterogeneous feature vectors include kinematic heterogeneous feature vectors, mechanical heterogeneous feature vectors and / or electromyographic heterogeneous feature vectors; Based on the heterogeneous feature vectors of all test action segments, multi-dimensional feature fusion and time alignment are performed to obtain a multi-dimensional hand function feature matrix; Based on the multidimensional hand function feature matrix, Brunnstrom staging is performed to obtain Brunnstrom staging assessment reference results for stroke patients, which can be used as reference information for rehabilitation assessment.

2. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 1, characterized in that, The data glove system includes a first inertial measurement unit, five second inertial measurement units, five flexible bending sensors, thirteen flexible pressure sensors, and a surface electromyography (EMG) acquisition component. The first inertial measurement unit is located on the back of the hand, the five second inertial measurement units are correspondingly arranged on the fingertips, the five flexible bending sensors are correspondingly arranged along the back of the fingers, and the thirteen flexible pressure sensors are correspondingly embedded in the fingertips, palmar surface of the metacarpophalangeal joints, ulnar surface of the thumb, the thenar eminence, and hypothenar eminence. The surface EMG acquisition component includes a first bipolar surface electrode group, a second bipolar surface electrode group, a reference electrode, and flexor EMG acquisition channels and extensor EMG acquisition channels that respectively receive differential signals. The first bipolar surface electrode group is set on the skin surface corresponding to the flexor or extensor muscles of the forearm of a stroke patient, and the second bipolar surface electrode group is set on the skin surface corresponding to the extensor or extensor muscles of the forearm of a stroke patient, respectively, for detecting surface EMG signals at the corresponding locations. The two electromyography (EMG) acquisition channels synchronously acquire the raw surface EMG signals EMG_raw,F(t) on the flexor side and the raw surface EMG signals EMG_raw,E(t) on the extensor side based on the same sampling clock or unified timestamp, and share the reference electrode. The kinematic signals include triaxial angular velocity, triaxial acceleration, triaxial magnetic field and corresponding original attitude quaternions collected by the first inertial measurement unit and each second inertial measurement unit, as well as the original bending electrical signals collected by each flexible bending sensor. The raw mechanical signals include the raw pressure signals collected by each flexible pressure sensor; The raw sEMG signals include the raw surface electromyography (EMG) signals EMG_raw,F(t) on the flexor side and EMG_raw,E(t) on the extensor side, which are synchronously acquired by the flexor side EMG acquisition channel and the extensor side EMG acquisition channel during the same test action.

3. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 2, characterized in that, The preprocessing of the acquired kinematic signals, mechanical signals, and sEMG signals specifically includes: The original attitude quaternions collected by the first inertial measurement unit and each second inertial measurement unit are initialized with attitude and aligned with the local coordinate system. For each finger, the relative attitude quaternion of the corresponding second inertial measurement unit relative to the first inertial measurement unit is calculated based on the aligned quaternion of the first inertial measurement unit and the corresponding second inertial measurement unit. Each relative posture quaternion is converted into the overall Euler angle of the corresponding finger relative to the back of the hand, and the angle component corresponding to the finger flexion and extension movement is extracted as the IMU solution angle of the corresponding finger. An angle fusion is performed on the IMU-derived angle of each finger and the bending reference angle of the corresponding flexible bending sensor based on an adaptive complementary filter to obtain the overall fused angle of each finger; wherein, the bending reference angle is obtained by mapping the original bending electrical signal of each flexible bending sensor on each finger to the bending reference angle through factory calibration or multi-point attitude calibration. Based on the finger joint angle allocation parameters pre-calibrated and fixed for the subject or device model, the overall fusion angle of each finger is assigned to the corresponding finger joint to obtain the estimated fusion angle of each finger joint; wherein, the finger joints of the 2nd to 5th fingers include the metacarpophalangeal joint, proximal interphalangeal joint, and distal interphalangeal joint, and the finger joints of the thumb include the carpometacarpophalangeal joint, metacarpophalangeal joint, and interphalangeal joint; the estimated fusion angle of each finger joint is obtained based on the overall fusion angle allocation of the corresponding finger. Zero-point drift compensation and temperature drift calibration are performed on the raw pressure signals collected by each flexible pressure sensor. Based on the sampling rate, the effective bandwidth of the surface electromyography acquisition component, and the local power frequency, the two raw sEMG signals were subjected to bandpass filtering and power frequency notch filtering using the same parameters. The attitude of the three-axis angular velocities output by the first inertial measurement unit and each of the second inertial measurement units is aligned, and the three-axis angular velocities of each finger relative to the back of the hand are calculated based on the three-axis angular velocities after transformation to a unified coordinate system. Based on the triaxial angular velocities of each finger relative to the back of the hand, the angular velocity modulus of each finger relative to the back of the hand is calculated. The multimodal raw data stream is time-stamped according to a unified sampling clock, and the time-synchronized multimodal raw data stream is output, along with the corresponding overall fusion angle, the estimated fusion angle of each finger joint, and the triaxial angular velocity and angular velocity magnitude of each finger relative to the back of the hand. The multimodal raw data stream includes preprocessed kinematic signals, mechanical signals, and sEMG signals.

4. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 3, characterized in that, Before performing feature extraction on each test action segment based on the preprocessing results, the method further includes: An automatic action cycle segmentation algorithm based on fused kinematic thresholds is used to segment the action of the multimodal raw data stream: For each test action, the primary driving finger is determined; where, for test actions that require only one finger to move, the primary driving finger is the finger that needs to move when performing the test action; for test actions that require multiple fingers to move simultaneously, the primary driving finger is determined according to the provisions of the action standard protocol. The relative angular velocity modulus corresponding to the main driving finger of each test action is detected when the relative angular velocity modulus is greater than the preset starting threshold ω_start and the duration reaches the first target time. The moment when the relative angular velocity modulus first exceeds ω_start in each target time period is determined as the starting point of the corresponding action. For test actions that do not include a hold phase in the action protocol, detect the time period during which the relative angular velocity modulus corresponding to the main driving finger of the test action falls below the preset termination threshold ω_end and the duration reaches the second target time, and determine the moment when the relative angular velocity modulus first falls below ω_end in the time period as the action end point of the corresponding action. For test actions that include a hold phase in the action protocol, the moment when the relative angular velocity modulus corresponding to the main driving finger of the test action first falls below ω_end and satisfies the second target time is determined as the end point of the motion phase or the time when the target posture is reached t_arr. The hold phase is then retained, and the earliest of the test action end prompt, release start, or the end of the maximum hold window specified in the action protocol is determined as the end point of the test action. Based on the start and end points of each test action, the time-synchronized multimodal raw data stream is divided into independent data segments corresponding to each test action to obtain each test action segment.

5. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 4, characterized in that, The test action segments include active action segments and passive action segments. The active action segments include active fist clenching, active finger extension, active grasping, and active pinching. The passive action segments include passive finger extension. Based on the preprocessing results, feature extraction is performed on each test action segment to obtain heterogeneous feature vectors for each test action segment, specifically including: For each active motion segment, the joint range of motion of each finger joint, the total active range of motion of the five fingers, and the smoothness of the motion trajectory are extracted. Separation motion indices are extracted for selective motion segments pre-specified by the motion standard protocol. For active fist clenching segments and active five-finger extension segments, the correlation coefficient ρ of the common motion of the five fingers is calculated to obtain the kinematic heterogeneous feature vector of each active motion segment. Among them, the finger joints of the 2nd to 5th fingers include the metacarpophalangeal joint (MCP), the proximal interphalangeal joint (PIP), and the distal interphalangeal joint (DIP), and the finger joints of the thumb include the carpometacarpophalangeal joint (CMC), the metacarpophalangeal joint (MCP), and the interphalangeal joint (IP). For each active grasp segment and each active pinch segment, the maximum muscle strength index is extracted to obtain the mechanical heterogeneous feature vector of each active grasp segment and each active pinch segment; For each passive extension segment of a finger, muscle tone quantification index is extracted to obtain the mechanical heterogeneous feature vector of each passive extension segment of a finger; For each test action segment, the degree of muscle activation, the antagonist muscle synergistic contraction rate, and the degree of muscle fatigue are extracted. For active action segments, the time difference Δt=|t_EMG_on-t_motion_on| between the onset of electromyographic activation t_EMG_on and the onset of movement t_motion_on is further extracted to obtain the corresponding electromyographic heterogeneous feature vector.

6. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 5, characterized in that, For each active motion segment, the range of motion of each finger joint, the total active range of motion of the five fingers, and the smoothness of the motion trajectory are extracted. Separation motion indices are also extracted from selective motion segments pre-specified by the motion standard protocol. Specifically, this includes: For each active movement segment, the range of motion of each finger joint is calculated using the following formula: ROM_(i,j)=max(θ_(i,j)(t))-min(θ_(i,j)(t)); Where i represents the finger number, i∈{1,2,3,4,5}, representing the thumb, index finger, middle finger, ring finger, and little finger respectively; j represents the joint type of finger i, ROM_(i,j) represents the joint mobility of the finger joint j of finger i; θ_(i,j)(t) represents the estimated fusion angle of the middle finger joint j of finger i, and max(θ_(i,j)(t)) and min(θ_(i,j)(t)) are both obtained within t∈[t_start,t_end], where t_start and t_end represent the start and end points of the current active action segment; For fingers 2-5, the active range of motion TAM_finger,i of finger i is calculated using the following formula: TAM_finger,i=ROM_MCP,i+ROM_PIP,i+ROM_DIP,i; Wherein, ROM_MCP,i represents the range of motion of the metacarpophalangeal joint of finger i, ROM_PIP,i represents the range of motion of the proximal interphalangeal joint of finger i, and ROM_DIP,i represents the range of motion of the distal interphalangeal joint of finger i. The active range of motion (TAM_thumb) of the thumb is calculated using the following formula: TAM_thumb=ROM_CMC+ROM_MCP+ROM_IP; Wherein, ROM_CMC represents the range of motion of the carpometacarpal joint of the thumb, ROM_MCP represents the range of motion of the metacarpophalangeal joint of the thumb, and ROM_IP represents the range of motion of the interphalangeal joint of the thumb. For each active motion segment, the total active range of motion (TAM_total) of the five fingers is calculated using the following formula: ； For each active motion segment, the smoothness of the motion trajectory is characterized by a normalized Jerk value, calculated using the following formula: Jerk_norm=(T 5 / D 2 )·∫[d 3 θ_out,i(t) / dt 3 ] 2 dt; Where Jerk_norm represents the normalized Jerk value, T represents the duration of each active action segment, T=t_end-t_start, t_start and t_end represent the start and end points of the current active action segment, D represents the total change in the overall fusion angle of the main driving finger i within the current active action segment, D=max(θ_out,i(t))-min(θ_out,i(t)), θ_out,i(t) represents the overall fusion angle of the unique main driving finger i pre-specified for the current test action by the action standard protocol, and max(θ_out,i(t)) and min(θ_out,i(t)) are both obtained within t∈[t_start,t_end]. For selective motion segments pre-specified by the motion standard protocol, extract the separation motion index according to the following steps: Identify the non-target fingers to obtain a set N of non-target fingers; wherein, the non-target fingers refer to the fingers that should remain still when performing the test action of the selective action segment; Extract the angular velocity modulus |ω_rel,i(t)| of the main driving finger i relative to the back of the hand and the angular velocity modulus |ω_rel,k(t)| of each non-target finger k relative to the back of the hand, k∈N; If the number of effective sampling points for the primary driving finger i and the non-target finger k meets the requirements, and the standard deviations of |ω_rel,i(t)| and |ω_rel,k(t)| within the selected action segment are both greater than the preset fluctuation threshold, indicating that the primary driving finger i has effective movement, calculate the Pearson correlation coefficient r_inter,k of |ω_rel,i(t)| and |ω_rel,k(t)| to obtain the correlation coefficient for each non-target finger k: r_inter,k=Cov[|ω_rel,i(t)|,|ω_rel,k(t)|] / (σ_rel,i·σ_rel,k); Where σ_rel,i is the standard deviation of |ω_rel,i(t)| within the selective action segment, σ_rel,k is the standard deviation of |ω_rel,k(t)| within the selective action segment, the calculation interval is the start and end time of the selective action segment, σ_rel,i>0, σ_rel,k>0; r_inter,k represents the Pearson correlation coefficient between |ω_rel,i(t)| and |ω_rel,k(t)| within the selective action segment, k∈N; Cov[·] is the covariance operator; If σ_rel,i is greater than the preset fluctuation threshold, while σ_rel,k is not greater than the preset fluctuation threshold, and |ω_rel,k(t)| does not exceed the preset resting threshold within the selective action segment, then the corresponding r_inter,k is determined to be 0; Based on the correlation coefficient of the non-target finger of the selected motion segment, the separation motion index FII is determined according to the following formula: FII = 1 - max(|r_inter,k|),k∈N; If the main driving finger i has no valid movement, mark FII as missing and prompt for retest until the main driving finger i has valid movement; For the active fist-clenching segment and the active five-finger stretching segment, the five-finger co-movement correlation coefficient ρ is calculated according to the following steps: Extract the overall fusion angles of the five fingers, and calculate the Pearson correlation coefficient r_{i,k} of any two fingers among the five fingers according to the following formula: r_{i,k}=corr(θ_out,i(t),θ_out,k(t)); Wherein, r_{i,k} represents the Pearson correlation coefficient of the i-th finger and the k-th finger, 1≤i<k≤5, θ_out,i(t) represents the overall fusion angle of the i-th finger, θ_out,k(t) represents the overall fusion angle of the k-th finger, and corr() represents the statistical function for calculating the Pearson correlation coefficient of θ_out,i(t) and θ_out,k(t); Calculate the five-finger co-movement correlation coefficient ρ according to the following formula: ρ=(1 / 10)Σ|r{i,k}|.

7. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 5, characterized in that, The maximum muscle strength index includes the maximum grip strength quantified value F_peak,grip and the maximum pinch strength quantified value F_peak,pinch; The extracting the maximum muscle strength index for each active grasping segment and each active pinching segment specifically includes: For each active grasping segment and active pinching segment, calculate the total muscle strength F_total(t) according to the following formula: F_total(t) = ΣF_i(t); F_i(t) = k_i·P_i(t); Wherein, F_i(t) represents the calibrated contact force of the i-th flexible pressure sensor, k_i represents the calibration coefficient, and P_i(t) represents the original pressure signal collected by the i-th flexible pressure sensor; sum the contact forces F_i(t) of the relevant flexible pressure sensors pre-specified by the current action rule to obtain the total muscle strength F_total(t); Extract the peak value F_peak,grip of the total muscle strength F_total(t) of each active grasping segment as the maximum grip strength quantified value of the active grasping segment, and extract the peak value F_peak,pinch of the total muscle strength F_total(t) of each active pinching segment as the maximum pinch strength quantified value of the active pinching segment; The extracting the muscle tension quantified index for each finger passive stretching segment specifically includes: For the finger passive stretching segment of any finger, within the preset contact position, loading direction and angular velocity range, average the contact forces F_i(t) of the calibrated flexible pressure sensors that are consistent with the finger passive stretching loading direction of the any finger, or perform weighted summation according to the preset weights, to obtain the combined resistance F_resist(t) of the any finger; For any passive extension segment of a finger, define the non-negative passive extension angular displacement. (t) = |θ_MCP(t) - θ_MCP(t_start)|, where θ_MCP(t) is the estimated fusion angle of the MCP joint of any finger at time t, and θ_MCP(t_start) is the estimated fusion angle of the MCP joint of any finger at the start of the movement. With F_resist(t) as the x-axis and F_resist(t) as the y-axis, construct the drag-extension angular displacement curve F_resist(t). ); Based on the aforementioned resistance-extension angular displacement curve F_resist( ),according to Extract the area under the non-negative curve (AUC), where... _end represents the non-negative extension angular displacement at the end of the passive extension segment of the finger; and the resistance-extension angular displacement curve F_resist( In the linear ascending segment of ), F_resist is relative to The non-negative fitting slope is used as the stiffness coefficient K_stiff; The extracting the muscle activation degree, the antagonistic muscle co-contraction rate and the muscle fatigue degree for each test action segment specifically includes: For each test action segment, the preprocessed sEMG signals from the flexor side EMG acquisition channel F and the extensor side EMG acquisition channel E are subjected to full-wave rectification and low-pass envelope processing to obtain EMG_env,F(t) and EMG_env,E(t). For each path c∈{F,E}, the root mean square value of the integral electromyography (RMS_c) is calculated based on the electromyographic envelope EMG_env,c(t) according to the following formula: RMS_c = ; Where T = t_end - t_start represents the duration of each test action segment, t ∈ [t_start, t_end]; In the MVIC calibration test, an analysis window of the same duration as each test action segment is used to calculate RMS_MVIC,c, and the calculation is performed when the corresponding MVIC reference value is greater than the minimum effective value and the calibration signal quality is qualified. %RMS_MVIC,c=(RMS_c / RMS_MVIC,c)×100%. Where %RMS_MVIC,c represents the standardized muscle activation level; Within the same effective action segment, the electromyographic envelopes of the flexor-side electromyographic acquisition channel F and the extensor-side electromyographic acquisition channel E are normalized according to their respective effective reference values ​​to obtain E_F(t) and E_E(t). The co-contraction rate CR of the flexor-side and extensor-side muscle groups is then calculated according to the following formula: CR=[∫min(E_F(t),E_E(t))dt / ∫max(E_F(t),E_E(t))dt]×100%.

8. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 1, characterized in that, The Brunnstrom staging process, based on the multi-dimensional hand function feature matrix, yields reference results for Brunnstrom staging assessment of stroke patients, specifically including: Based on the multi-dimensional hand function feature matrix, it is determined whether the conditions of various features of each Brunnstrom stage are met; wherein, the various features include three categories: kinematic features, mechanical features, and electromyographic features. For each category of features in each Brunnstrom period: If the required features in the class features used for the Brunnstrom staging determination are complete and the required features satisfy the corresponding feature conditions of the Brunnstrom staging, then the indicator function of the class feature takes the value of 1; If the required features for the class feature used in the Brunnstrom staging determination are complete, but at least one of the required features does not meet the corresponding feature conditions of the Brunnstrom staging, then the indicator function of the class feature takes the value of 0. If any required feature is missing from the category feature used for Brunnstrom staging determination, the output data is incomplete and a retest prompt is given until the required feature is found to be complete in the category feature during retesting. Based on the above situation, the value of the indicator function of the category feature is determined. For each Brunnstrom stage, the score for that stage is calculated based on the indicator function value for each feature class, according to the following formula: S_stage =Σ_d(w_d×I(feature_d)); Where S_stage represents the score of the Brunnstrom stage, w_d represents the weight of the d-th feature of the Brunnstrom stage, feature_d represents the d-th feature of the Brunnstrom stage, and I(feature_d) represents the value of the indicator function of the d-th feature of the Brunnstrom stage. Based on the scores of each Brunnstrom stage, the Brunnstrom stage assessment reference results for the stroke patients are output, specifically including: If only one Brunnstrom period achieves the highest score, the Brunnstrom period with the highest score will be output as the candidate period.

9. The method for evaluating the staged hand motor rehabilitation of stroke patients according to claim 8, characterized in that, The method of outputting the Brunnstrom staging assessment reference results for stroke patients based on the scores of each Brunnstrom stage also includes: If all features involved in the score calculation for any Brunnstrom phase are complete, and the scores for all Brunnstrom phases are 0, a prompt will be output indicating that the current quantitative rule base phase conditions have not been met. If the highest and second-highest scores of all Brunnstrom periods correspond to adjacent Brunnstrom periods, and the difference between the highest and second-highest scores satisfies the transition judgment conditions pre-set in the quantitative rule base, and at least one pre-specified key feature appears in a cross-period fulfillment state between the two adjacent Brunnstrom periods, then the transition period prompt for the two adjacent Brunnstrom periods is output, and the conflict feature is recorded; if the calibration of the transition judgment conditions is not completed, the scores and conflict features of the two adjacent Brunnstrom periods are output for manual review. If the above conditions are not met and two or more Brunnstrom stages achieve the highest score, or if the two Brunnstrom stages corresponding to the highest score are not adjacent, output a message indicating a conflict between the two Brunnstrom stages with the highest scores and prompt for retesting or manual review.

10. A phased hand motor rehabilitation assessment system for stroke patients, characterized in that, The stroke patient phased hand motor rehabilitation assessment system includes: The data glove system is worn on the hands of stroke patients during standardized hand function testing to collect kinematic, mechanical, and sEMG signals from the stroke patient's hand during the standardized hand function testing process. The processing module is used to execute the phased hand movement rehabilitation evaluation method for stroke patients as described in any one of claims 1-9.

Images