Evaluation training method and system based on hand function combination evaluation training board
By combining multimodal sensors and artificial intelligence models, the difficulty of training tasks can be evaluated in real time and dynamically adjusted, which solves the problems of subjectivity and staticity in assessment and training in intelligent hand function rehabilitation, and achieves personalized and efficient rehabilitation training results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies in the field of intelligent hand function rehabilitation suffer from problems such as strong subjectivity in hand function assessment methods, lack of effective closed-loop feedback, and difficulty in achieving personalization and dynamic adjustment, resulting in poor efficiency and effectiveness of rehabilitation training.
A multimodal sensor array is used to collect patients' hand movement data in real time. The data is mapped to clinical assessment scores through an artificial intelligence model, and the difficulty of training tasks is dynamically adjusted based on the scores to form a personalized and adaptive assessment-training closed loop.
It achieves objectivity, personalization, and improved efficiency in hand function rehabilitation training, overcomes the subjectivity of traditional assessment and the static nature of training, and ensures the continuous effectiveness and safety of training tasks.
Smart Images
Figure CN121659043A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent rehabilitation equipment technology. More specifically, this invention relates to an assessment and training method and system based on a hand function combination assessment and training board. Background Technology
[0002] Hand function rehabilitation is a crucial step in restoring patients' daily living abilities, and the effectiveness of its assessment and training directly impacts rehabilitation outcomes. However, current technologies still face significant challenges in the field of intelligent hand function rehabilitation.
[0003] First, traditional methods for assessing hand function, such as clinical scales (e.g., the Fugl-Meyer Assessment Scale), rely heavily on the therapist's experience and subjective judgment. This approach makes it difficult to objectively and quantitatively measure a patient's fine motor control, resulting in assessment results that are highly subjective and inconsistent.
[0004] Secondly, the assessment and training phases are often disconnected, lacking an effective closed-loop feedback mechanism. Existing rehabilitation training usually follows a pre-set plan or is manually adjusted by the therapist, making it difficult to provide dynamic and personalized intervention based on the patient's real-time performance, fatigue level, and individual differences, thus limiting the efficiency and safety of the training.
[0005] While some rehabilitation devices may collect patients' basic movement or strength data through sensors (such as pressure sensors or inertial measurement units) and perform preliminary processing using simple digital filtering algorithms, these techniques are usually limited to data acquisition and preliminary display. They generally lack algorithmic models that can deeply fuse multimodal sensor data and objectively and quantitatively map it to standardized clinical assessment scales.
[0006] Therefore, existing technologies struggle to provide comprehensive, accurate, and clinically aligned objective assessments, and are unable to achieve intelligent and dynamic adjustments to training tasks based on these assessments, thus limiting the personalization and efficiency of rehabilitation training. Summary of the Invention
[0007] To address the aforementioned technical problems of low efficiency and poor effectiveness in hand function rehabilitation training, this invention provides solutions in the following aspects.
[0008] In a first aspect, the present invention provides an assessment and training method based on a hand function combination assessment and training board, comprising: acquiring real-time data from a multimodal sensor array on the hand function combination assessment and training board for a personalized training task generated for the patient; preprocessing the real-time data and extracting key performance features characterizing the quality of the patient's hand movements from it, wherein the key performance features include at least one of force features, time features, kinematic features, and stability features; inputting the key performance features into a preset artificial intelligence assessment model to obtain a quantitative assessment score corresponding to a clinical rehabilitation scale; and adaptively adjusting the difficulty parameters of the next round of training tasks based on a comparison between the quantitative assessment score and a preset performance threshold.
[0009] In one embodiment, the generation of the personalized training task includes: before the first training session, guiding the patient to complete preset standardized assessment actions and collecting corresponding sensor data to establish an initial capability baseline model; and matching and generating the training task from a task library based on the initial capability baseline model.
[0010] In one embodiment, the multimodal sensor array includes at least one of a pressure sensor, an inertial measurement unit, a touch sensor, and a rotary encoder.
[0011] In one embodiment, the force characteristics include the maximum force, average force, and force rise rate during the task; the kinematic characteristics include the smoothness of the motion trajectory and the frequency of jitter; and the stability characteristics include the standard deviation of the actual output force when maintaining a constant force.
[0012] In one embodiment, the artificial intelligence evaluation model is a support vector machine model, a random forest model, or a long short-term memory network model, and the training process of the artificial intelligence evaluation model includes minimizing the mean squared error between the predicted score and the clinical expert score.
[0013] In one embodiment, the artificial intelligence evaluation model includes a linear combination that maps multiple key performance features to predicted scores, calculated as follows:
[0014] in: This represents the predicted Fugl-Meyer assessment score. Indicates the first A key performance characteristic value extracted from sensor data, Indicates the first The weight coefficients corresponding to each feature, b is the bias term, and f(•) is an activation function or mapping function.
[0015] In one embodiment, the adaptive adjustment includes: increasing the difficulty of the next training task when the quantitative evaluation score is higher than a preset high-performance threshold; and reducing the difficulty of the next training task when the quantitative evaluation score is lower than a preset low-performance threshold, wherein the high-performance threshold is 90% to 95% of the full evaluation score; and the low-performance threshold is 50% to 60% of the full evaluation score.
[0016] In one embodiment, the difficulty parameters of the training task include: the magnitude of the target force, the target maintenance time, the allowed reaction time, or the range of motion.
[0017] In one embodiment, the method further includes: acquiring fatigue-related indicators from the key performance characteristics; and when the fatigue-related indicators are detected to be deteriorating in multiple consecutive training tasks, pausing training and issuing a rest suggestion to the patient.
[0018] In a second aspect, the present invention provides an evaluation and training system based on a hand function combination evaluation and training board, comprising: a processor; and a memory storing computer program instructions, which, when executed by the processor, implement an evaluation and training method based on a hand function combination evaluation and training board as described in one or more of the preceding embodiments.
[0019] The beneficial effects of this invention are as follows: This invention utilizes multimodal sensors to capture objective data from patients during training in real time, and then uses an artificial intelligence model to instantly convert this data into standardized clinical assessment scores. Based on these scores, the difficulty of the training tasks is dynamically and automatically adjusted. This method seamlessly integrates assessment and training, forming a personalized and adaptive closed loop of assessment-training-reassessment. It overcomes the shortcomings of traditional rehabilitation assessment being subjective and training being static, and significantly improves the objectivity, efficiency, and personalization of rehabilitation.
[0020] Furthermore, this invention also constructs an artificial intelligence model to map the objective feature vectors extracted in the previous step into quantitative assessment scores that are easy for clinicians to understand in real time, thereby realizing automated and standardized functional assessment during the training process and effectively improving the training efficiency of the intelligent hand function combination training board. Attached Figure Description
[0021] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating an evaluation and training method based on a hand function combination evaluation and training board according to an embodiment of the present invention; Figure 2 This is a simplified diagram of a device illustrating a hand function combination assessment training board according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the composition of an evaluation and training system based on a hand function combination evaluation and training board according to an embodiment of the present invention. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] This invention provides an assessment and training method based on a hand function combination assessment and training board. By digitizing the training process and modeling the assessment, a closed-loop feedback of assessment guiding training and training verifying assessment is realized, which significantly improves the objectivity, personalization and efficiency of rehabilitation training.
[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an evaluation training method 100 based on a hand function combination evaluation training board according to an embodiment of the present invention.
[0025] like Figure 1 As shown, in step S101, for the personalized training task generated for the patient, real-time data from the multimodal sensor array on the hand function combination assessment training board is acquired. The multimodal sensor array includes at least one of a pressure sensor, an inertial measurement unit, a touch sensor, and a rotary encoder. The generation of the personalized training task includes: before the first training, guiding the patient to complete preset standardized assessment actions and collecting corresponding sensor data to establish an initial capability baseline model. Based on the initial capability baseline model, the training task is matched and generated from the task library.
[0026] In step S102, the real-time data is preprocessed, and key performance features characterizing the quality of the patient's hand movements are extracted. These key performance features include at least one of the following: force characteristics, time characteristics, kinematic characteristics, and stability characteristics. In some embodiments, force characteristics include maximum force, average force, and rate of force rise during the task; kinematic characteristics include the smoothness of the movement trajectory and the frequency of jitter; and stability characteristics include the standard deviation of the actual output force when maintaining a constant force.
[0027] In one application scenario, the artificial intelligence evaluation model includes a linear combination that maps multiple key performance features to predicted scores, calculated as follows:
[0028] in: This represents the predicted Fugl-Meyer assessment score. Indicates the first A key performance characteristic value extracted from sensor data, Indicates the first The weight coefficients corresponding to each feature, b is the bias term, and f(•) is an activation function or mapping function.
[0029] In step S103, key performance characteristics are input into a pre-defined artificial intelligence assessment model to obtain a quantitative assessment score corresponding to the clinical rehabilitation scale. This artificial intelligence assessment model is a support vector machine model, a random forest model, or a long short-term memory network model. The training process of the artificial intelligence assessment model includes minimizing the mean squared error between the predicted score and the clinical expert score.
[0030] In step S104, the difficulty parameters of the next training task are adaptively adjusted based on the comparison between the quantitative evaluation score and a preset performance threshold. In some embodiments, the adaptive adjustment can be performed as follows: when the quantitative evaluation score is higher than a preset high-performance threshold, the difficulty of the next training task is increased; when the quantitative evaluation score is lower than a preset low-performance threshold, the difficulty of the next training task is decreased. The high-performance threshold is 90% to 95% of the full evaluation score, and the low-performance threshold is 50% to 60% of the full evaluation score. The difficulty parameters of the training task include: the magnitude of the target force, the target maintenance time, the allowed reaction time, or the range of motion.
[0031] Furthermore, the difficulty of the training task can be adjusted by setting a criterion that the quantitative evaluation score is repeatedly higher or lower than a preset performance threshold. For example, if the task completion rate is >90% for three consecutive times, the difficulty level is increased by 1.
[0032] In addition, fatigue-related indicators from key performance characteristics can be acquired during training. When these fatigue-related indicators show a worsening trend across multiple consecutive training tasks, training is paused and a rest recommendation is issued to the patient.
[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention.
[0034] Figure 2 This is a simplified diagram of a device illustrating a hand function combination evaluation training board according to an embodiment of the present invention.
[0035] like Figure 2As shown, in the embodiments of the present invention, the hand function combination assessment training board is made of high-strength, lightweight materials (such as engineering plastics and carbon fiber), and its form can be flat, three-dimensional or modular assembly type.
[0036] The board integrates multiple functional areas, corresponding to different hand function training processes, specifically: Pressure Sensing Area: Embedded with a high-precision pressure film sensor for gripping and pressing pressure training and evaluation. Touch Slide Area: Employs capacitive or resistive touch strips for evaluating finger coordination and control precision. Button and Switch Area: Arranges physical buttons and switches of different sizes, shapes, and pressing pressures to simulate everyday operations. Pinch Module: Provides cylinders and particles of different sizes, with an integrated miniature force sensor for fine-grasping function evaluation. Rotation Module: An encoder-integrated knob for evaluating wrist rotation function and flexibility.
[0037] Data Acquisition and Processing Unit: Main Control Chip (MCU): Employs a high-performance embedded processor (such as the ARM Cortex-M series) responsible for basic sensor data acquisition, preprocessing, and transmission. The MCU is equipped with a data processing module that implements digital filtering (such as Kalman filtering) to eliminate sensor noise. It normalizes the raw data to eliminate individual differences and environmental interference. The feature extraction module calculates key feature values in real time from the preprocessed data, such as: force features: maximum force, average force, force rise rate; time features: reaction time, total task completion time, motion interval time; kinematic features: motion trajectory smoothness, jitter frequency, range of motion; stability features: stability error of force maintenance. Sensor Module: Includes a pressure sensor array, IMU (Inertial Measurement Unit), touch sensor, etc., responsible for raw data acquisition. Communication Module: Integrates a Wi-Fi / Bluetooth module to wirelessly transmit the acquired data to a host computer (such as a computer, tablet) or the cloud.
[0038] Terminal display and interaction devices: Using tablets, touchscreens, or VR / AR headsets as display terminals, these devices present training task instructions, animation guidance, and real-time feedback results. Based on instructions from the decision engine, new training game scenarios are retrieved from the task library or dynamically generated (e.g., requiring the user to pop a balloon on the screen with a specific force or light up different colored lights in sequence), and then rendered on the display terminal.
[0039] The solution of the present invention will be further described below with specific steps. An exemplary process of an evaluation and training closed-loop method based on a smart hand function combination evaluation and training board provided by an embodiment of the present invention may include the following steps: S1: Establish a baseline model of the patient's initial capabilities.
[0040] In this embodiment, when a stroke patient with hand dysfunction (e.g., in Brunnstrom stage III-IV) uses the intelligent hand function combination assessment training board of this system for the first time, the system guides them to complete a set of preset standardized assessment action sequences. The perception layer of this assessment training board integrates a multimodal sensor array, including but not limited to: Pressure sensor array: Distributed and embedded under the grip area and press button, used to accurately measure the force, duration of force, and rate of change of force when the user performs tasks such as gripping and pressing.
[0041] Inertial Measurement Unit (IMU): It can be built into the training board or worn as a wearable finger sleeve to collect real-time data on the acceleration, angular velocity and attitude of the hand or training board in three-dimensional space.
[0042] Touch sensors: such as capacitive touch bars, are used to detect finger contact, position, and swipe trajectory.
[0043] Rotary encoder: Integrated into the knob module, used to quantify the wrist's rotation angle, speed, and direction.
[0044] During the execution of standard movements, raw data streams from the aforementioned sensor array are collected in real time. This data constitutes an objective, multi-dimensional portrayal of the patient's initial hand function. Subsequently, the collected data undergoes preliminary processing to establish the patient's personal ability baseline profile. This profile not only includes traditional information such as muscle strength and joint range of motion, but also quantitative indicators of fine motor control abilities such as movement smoothness, coordination, and stability.
[0045] S2: Generate and execute personalized initial training tasks.
[0046] Based on the patient's baseline ability model established in step S1, the system's decision engine module will match and generate a personalized training task with an initial difficulty from the task library. For example, if the baseline assessment shows that the patient's maximum grip strength is 3 kg, but the stability of force control is poor, the system may generate a task: "Please use the affected hand to press the designated area with a force of 2 kg and hold for 3 seconds."
[0047] Task instructions are presented to the patient in the form of visual animation or voice through a terminal display device (such as a tablet or VR headset) connected to the training board. After watching the demonstration, the patient attempts to perform the task in the corresponding functional area of the training board. During the execution, the sensor array of the perception layer continuously collects real-time training data.
[0048] S3: Real-time data preprocessing and key feature extraction.
[0049] During the patient's training task, the main control unit (MCU) preprocesses the high-frequency raw sensor data in real time to eliminate noise and extract meaningful kinematic and dynamic information.
[0050] Data preprocessing: As a preferred approach, digital filtering algorithms, such as Kalman filters, can be used to filter the raw data from the IMU and pressure sensors to eliminate environmental noise and random errors inherent in the sensors. Simultaneously, the data is normalized to eliminate the influence of dimensional differences caused by individual variations.
[0051] Key Feature Extraction: The preprocessed data stream is used to calculate a series of key performance feature values in real time, which form the basis for subsequent intelligent evaluation. Exemplary features include: Force characteristics: such as maximum force, average force, and rate of force rise during the task.
[0052] Time characteristics (Ftime): such as reaction time (Treaction), total task completion time (Ttotal), and action interval (Tinterval).
[0053] Kinematics: such as smoothness of motion trajectory, jitter, and range of motion (ROM) calculated from IMU data.
[0054] Stability characteristics (Fstability): such as the standard deviation or coefficient of variation of the actual output force when a constant force is required.
[0055] S4: Intelligent assessment and scale mapping based on artificial intelligence models.
[0056] The key feature vectors extracted in step S3 are input in real time into a pre-trained artificial intelligence (AI) assessment model. This model maps objective sensor data to scores on widely used clinical rehabilitation assessment scales (such as the Fugl-Meyer Assessment Scale (FMA-UE) for the upper limbs and the Wolf Motor Function Test (WMFT)). This AI assessment model can be deployed in two ways: in the cloud and at the edge. Specifically, when deployed in the cloud, the feature data can be uploaded to a cloud server and analyzed using a pre-trained deep learning model (such as CNN or LSTM) or a machine learning model (such as SVM or random forest), returning assessment scores and diagnostic results. When deployed at the edge, a lightweight model can be deployed on a local processor or host computer to achieve low-latency real-time assessment while protecting user privacy.
[0057] In this embodiment, the AI model can be a regression model, such as a Support Vector Machine (SVM), Random Forest, or a lightweight Long Short-Term Memory (LSTM) network. The model is constructed based on the fact that each rating item on a clinical scale corresponds to a specific quality of movement, which can be quantified using a combination of one or more features from sensor data. For example, the "coordination / speed" rating on the Fugl-Meyer scale is highly correlated with trajectory smoothness in kinematic features and task completion time in temporal features.
[0058] The model outputs a quantitative evaluation score; for example, the overall score for task completion is 95%. Furthermore, a quantitative relational model can be constructed to map multi-dimensional feature vectors to predicted scores on clinical scales. This is based on the premise that the total clinical score can be viewed as a weighted combination of various performance quality features. An exemplary linear mapping formula is as follows:
[0059] in: This represents the predicted Fugl-Meyer assessment score. Indicates the first Key performance characteristics extracted from sensor data, such as maximum force Fmax or motion smoothness. Indicates the first The weights for each feature are obtained by training the model on a dataset containing a large number of paired "sensor feature-clinical expert rating" samples. b is the bias term. f(•) is an activation function or mapping function used to ensure that the output score is within the valid range of the scale.
[0060] As can be seen from the above formula, when the characteristic values reflecting good athletic ability (such as Fmax and Smoothness) increase, if their corresponding weights... A positive result will lead to a higher predicted clinical score. The corresponding improvement reflects an objective improvement in the patient's hand function in the scenario.
[0061] As a specific implementation method, a random forest model can be used. During the training phase, a large amount of patient training data and their Fugl-Meyer scores provided by experienced rehabilitation therapists are collected to form the training set. The goal of model training is to minimize the mean squared error between the predicted score and the actual score. Preferably, the hyperparameters of the random forest model can be set as follows: the number of decision trees is 100-200, and the maximum depth is 10-20. Through this model, the system can output an objective evaluation result with a high within-group correlation coefficient (ICC ≥ 0.90).
[0062] S5: Decision-making and dynamic adjustment of training tasks based on evaluation results.
[0063] The system's built-in adaptive decision engine makes real-time decisions based on the evaluation results output in step S4 and dynamically adjusts the difficulty of the next training task.
[0064] Decision Logic: This engine can be based on a pre-defined "if-then" rule base. As a preferred approach, the performance score threshold of this rule base can be parameterized. For example: Good performance: If the assessment score is higher than a preset high-performance threshold Thigh (preferably 90%), the system determines that the current task difficulty is too low. When the score is higher than Thigh, it indicates that the patient can easily complete the task, and the difficulty needs to be increased to maintain the challenge of training.
[0065] Poor performance: If the assessment score is below a preset low performance threshold Tlow (preferably 60%), the system determines that the task is too difficult or the patient has not mastered the key points of the action.
[0066] Fatigue: If the system detects specific fatigue characteristics (e.g., a significant increase in jitter frequency in kinematic characteristics over several consecutive tasks, or a continuous deterioration in stability characteristics), the system will determine that the patient may be fatigued, even if the score is acceptable.
[0067] Difficulty Adjustment: Based on the decision results, the system automatically adjusts the parameters of the next task. For example, increasing difficulty: increasing the required pressure from 2kg to 2.5kg, or shortening the allowed reaction time. Decreasing difficulty: decreasing the required pressure from 2kg to 1.5kg, extending the required duration to 2 seconds, and providing more detailed visual guidance.
[0068] Suggest a break: A pop-up message appears saying "You seem tired, take a 30-second break!" and plays relaxing music.
[0069] In this way, the training intensity is always dynamically maintained within the user's optimal challenge range, ensuring the effectiveness and safety of the training.
[0070] S6: Report generation, data synchronization, and closed-loop iteration.
[0071] After each training session, the system automatically generates an instant report, which is displayed to the patient via a terminal device. The report includes completed tasks, average score, a progress curve comparing the previous training session to the previous one, and visual charts such as fatigue index. This data is recorded and synchronized to a cloud server or host computer for remote monitoring and analysis by rehabilitation therapists.
[0072] The patient can then begin a new round of tasks adjusted from step S5, and the system will repeat steps S2 to S6. This "assessment-training-reassessment" cycle is repeated continuously in a single training session, forming a complete, personalized adaptive training loop.
[0073] In conjunction with the above embodiments, this invention combines objective sensor data with clinical assessment standards through hardware coupling, algorithm mapping, and closed-loop feedback, achieving accurate assessment of patients' hand function and personalized, adaptive rehabilitation training. This method not only overcomes the subjectivity and lag of traditional rehabilitation assessments but also ensures the continuous effectiveness of training tasks through a closed-loop adjustment mechanism, thus potentially significantly improving rehabilitation outcomes and efficiency.
[0074] Figure 3 This is a schematic diagram illustrating the composition of an evaluation and training system based on a hand function combination evaluation and training board according to an embodiment of the present invention.
[0075] This invention also provides an assessment and training system based on a hand function combination assessment and training board. For example... Figure 3 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an evaluation and training method based on a hand function combination evaluation and training board as described above.
[0076] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.
[0077] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.
[0078] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.
Claims
1. An assessment and training method based on a hand function combination assessment and training board, characterized in that, include: For the personalized training tasks generated for the patient, real-time data from the multimodal sensor array on the hand function combination assessment training board is obtained; The real-time data is preprocessed, and key performance features characterizing the quality of the patient's hand movements are extracted from it. The key performance features include at least one of the following: force features, time features, kinematic features, and stability features. The key performance characteristics are input into a preset artificial intelligence evaluation model to obtain a quantitative evaluation score corresponding to the clinical rehabilitation scale. Based on the comparison between the quantitative evaluation score and the preset performance threshold, the difficulty parameters of the next training task are adaptively adjusted.
2. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, The generation of the personalized training task includes: Before the first training session, patients are guided to complete pre-set standardized assessment actions, and corresponding sensor data is collected to establish an initial capability baseline model. The training task is generated by matching the task library based on the initial capability baseline model.
3. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, The multimodal sensor array includes at least one of a pressure sensor, an inertial measurement unit, a touch sensor, and a rotary encoder.
4. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, The force characteristics include the maximum force, average force, and rate of force rise during the task; Kinematic characteristics include the smoothness of the motion trajectory and the frequency of jitter; Stability characteristics include the standard deviation of the actual output force when maintaining a constant force.
5. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, The artificial intelligence evaluation model is a support vector machine model, a random forest model, or a long short-term memory network model. The training process of the artificial intelligence evaluation model includes minimizing the mean square error between the predicted score and the clinical expert score.
6. The assessment and training method based on the hand function combination assessment and training board according to claim 1 or 5, characterized in that, The artificial intelligence evaluation model includes a linear combination that maps multiple key performance features to predicted scores, and its calculation method is as follows: in: This represents the predicted Fugl-Meyer assessment score. Indicates the first A key performance characteristic value extracted from sensor data, Indicates the first The weight coefficients corresponding to each feature, b is the bias term, and f(•) is an activation function or mapping function.
7. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, The adaptive adjustment includes: When the quantitative evaluation score is higher than the preset high-performance threshold, the difficulty of the next round of training tasks is increased; When the quantitative evaluation score is lower than the preset low performance threshold, the difficulty of the next round of training task is reduced, wherein the high performance threshold is 90% to 95% of the full evaluation score; and the low performance threshold is 50% to 60% of the full evaluation score.
8. The assessment and training method based on the hand function combination assessment and training board according to claim 7, characterized in that, The difficulty parameters of the training task include: the magnitude of the target force, the target maintenance time, the allowable reaction time, or the range of motion.
9. The assessment and training method based on the hand function combination assessment and training board according to claim 1, characterized in that, Also includes: Obtain fatigue-related indicators from the key performance characteristics; When the fatigue-related indicators are detected to be worsening in multiple consecutive training tasks, training is suspended and the patient is advised to rest.
10. An assessment and training system based on a hand function combination assessment and training board, characterized in that, include: processor; A memory storing computer program instructions that, when executed by the processor, implement an evaluation and training method based on a hand function combination evaluation and training board according to any one of claims 1-9.