A stroke home hand training system based on seamless care model

Abstract

The application discloses a stroke home hand training system based on a seamless nursing model, and particularly relates to the technical field of medical data management, and is used for solving the problem that the stroke rehabilitation training at home lacks seamless nursing chain support, cannot be connected with hospital standard rehabilitation data, and results in poor training effect, poor compliance and high readmission rate, and comprises an end effector, a data acquisition module, a standard matching module, a recommended parameter generation module, a remote monitoring feedback module, a progress tracking evaluation module, a man-machine interaction module and a data storage analysis module; the application optimizes the stroke home hand training through a seamless nursing chain and an intelligent system, and generates task parameter recommendation data.

Description

Technical Field

[0001] This invention relates to the field of medical data management technology, and more specifically, to a home-based hand training system for stroke patients based on a seamless transfer care model. Background Technology

[0002] The article 1 ("The efficacy of the seamless transfer of care model to apply for the patients with cerebral apoplexy in China") introduces the effect of the Seamless Transfer of Care Model (STCM) on improving the readmission rate of stroke patients. It evaluates the model from three aspects: quality of life, compliance and readmission rate. It constructs a seamless care chain between the hospital, community and home, and reduces the difference between the home care effect and the hospital care effect. It is to carry out upper limb, lower limb and hand motor rehabilitation training for patients in the home environment, including hand movement, and follow the principle of gradual and orderly development of rehabilitation exercise. In the article 2 ("Tian Anni, Yang Jing, Sun Jing, et al. Meta-analysis of the effect of rehabilitation robot assisted training on the recovery of hand motor function in stroke patients at different stages [J / OL]. Chinese General Practice, 1-11[2]"), ...3 ("The efficacy of the seamless transfer of care model to apply for the patients with cerebral apoplexy in China"), introduces the effect of the seamless transfer of care model (STCM) on improving the readmission rate of stroke patients. It evaluates the model from three aspects: quality of life, compliance and readmission rate. It constructs a seamless care chain between the hospital, community and home, and reduces the difference between the home care effect and the hospital care effect. It is to carry out upper limb, lower limb and hand motor rehabilitation training for patients in the home environment, including hand movement, and follow the principle of gradual and orderly development of rehabilitation exercise. It is to carry out rehabilitation exercise training in the home environment, including upper limb, lower limb and hand movement, including hand movement, and follow the principle of gradual and orderly development of rehabilitation exercise. It is to carry out rehabilitation exercise training in the home environment, including upper limb, lower limb and hand movement, including upper limb, lower limb and hand movement, including upper limb, lower The article [024-11-07] discusses the training results of rehabilitation robots for stroke patients at different stages based on literature search. The meta-analysis results show that in acute and subacute stroke patients, rehabilitation robot-assisted training is beneficial to the recovery of fine motor function of the hand, especially grasping score and grip strength. However, in the existing home care of acute and subacute stroke patients, there is no system for optimizing hand rehabilitation training based on seamless transfer care model and rehabilitation robot. It is impossible to provide recommended parameters for the implementation of rehabilitation training robots for hand rehabilitation training of acute and subacute stroke patients based on data from the seamless care chain of hospitals, communities and homes. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, this invention provides a home-based hand training system for stroke patients based on a seamless transfer care model. Through a seamless care chain and intelligent optimization of home-based hand training for stroke patients, recommended values ​​for task parameters are generated.

[0004] In Reference 1, an analysis of nursing data from a group of stroke patients hospitalized in my country, comparing the STCM (Self-Transfer Care) group and a control group, showed that the seamless transfer nursing model improved the quality of life of stroke patients, increased medication and dietary adherence, and reduced readmission rates. Reference 2, through keyword-based literature analysis, indicated that rehabilitation robot-assisted training is beneficial for the recovery of upper limb and hand functions in acute and subacute stroke patients. Reference 2 also pointed out that post-stroke hand dysfunction specifically manifests as flexion contractures, inability to extend interphalangeal and metacarpophalangeal joints, and inability to perform fine motor movements such as grasping, opposition, and finger opposition. Hand grip strength and pinching strength are important indicators of hand function; decreased muscle strength is one of the main manifestations of motor dysfunction in stroke patients, affecting not only fine motor skills but also daily living abilities. Based on these two references, the technical solution of this invention is proposed.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention relates to a home-based hand training system for stroke patients based on a seamless transfer care model. The system's application background is the construction of a seamless care chain for stroke patients across hospitals, communities, and homes, and a meta-analysis of publicly available literature on rehabilitation robot training for acute and subacute stroke patients. The system includes an end effector, a data acquisition module, a standard matching module, a recommended parameter generation module, a remote monitoring and feedback module, a progress tracking and evaluation module, a human-computer interaction module, and a data storage and analysis module. The data acquisition module is connected to the end effector and the standard matching module at both ends. The standard matching module is connected to the recommended parameter generation module, which in turn is connected to both the remote monitoring and feedback module and the human-computer interaction module. The human-computer interaction module is also connected to both the remote monitoring and feedback module and the progress tracking and evaluation module. The system is interconnected with the end effector, data acquisition module, standard matching module, recommended parameter generation module, remote monitoring feedback module, progress tracking and evaluation module, and human-computer interaction module. The standard matching module maps the baseline points of the patient's hand measurement model to the baseline points of the standard model in space, so that the baseline points in the home training data match the baseline points of the standard model. By comparing the home training data of stroke patients' hands with the standard rehabilitation model data in the hospital, the accuracy of home hand training is judged based on feature similarity analysis. The recommended parameter generation module, based on the analysis results of the standard matching module, combined with the patient's training data and data development trends, as well as the patient's stroke hand dysfunction assessment report, predicts and generates recommended control parameters for each training task based on a machine learning model.

[0007] As a further aspect of the present invention, the rehabilitation training movements provided by the end effector for stroke patients include, but are not limited to, clenching a fist, extending all five fingers, extending the thumb, flexing the thumb, extending the index finger, flexing the index finger, extending the middle finger, flexing the middle finger, pinching with both fingertips, pinching with multiple fingertips, lateral pinching, spherical grasping, and cylindrical grasping.

[0008] As a further embodiment of the present invention, the data acquisition module includes image acquisition station 1, image acquisition station 2, and sensor data acquisition station. Image acquisition station 1 and image acquisition station 2 each include a motion image acquisition camera and a depth camera. The sensor data acquisition station includes a force sensor, an angle sensor, a position sensor, and an electromyography (EMG) sensor. The electrodes of the EMG sensor are located in the thenar eminence, hypothenar eminence, extensor digitorum group, and flexor digitorum group, respectively. The key points of the images acquired by the image acquisition camera include the wrist joint key point, the thumb key point, the key points of non-thumb fingers, and the center key point of the palm. The depth camera acquires the depth position and angle of the metacarpophalangeal joints, proximal interphalangeal joints, and distal phalangeal joints, the depth relationship between the thumb and non-thumb fingers, the vertical depth distance between the center point of the palm and the fingertip, the depth distance and angle from the wrist to the finger, and the depth difference value between the fingers, and tracks the depth trajectory of each finger joint during the training process.

[0009] As a further aspect of the present invention, the feature similarity analysis of the standard matching module is performed using data collected from image acquisition station #1, image acquisition station #2, and sensor data acquisition station. This includes multi-station fusion similarity analysis based on flexion contracture, multi-station fusion similarity analysis based on limited interphalangeal joint extension, and multi-station fusion similarity analysis based on insufficient grip and pinching strength.

[0010] As a further aspect of the present invention, in the standard matching module, the multi-station fusion similarity analysis based on flexion contracture is obtained by weighted fusion of the first angle similarity term, the first depth similarity term, and the first electromyographic signal similarity term in the patient's flexion contracture rehabilitation training. The formula for the multi-station fusion similarity analysis based on flexion contracture is as follows:

[0011] S q =α1·S angle +α2·S sg +α3·S EMG

[0012] In the formula: S q To achieve multi-station fusion similarity based on flexion contracture, S angleFor the first angular similarity term, ideal joint angles of the hand and fingers are extracted from the standard hospital rehabilitation model, and angles are set in the 3D simulation. The azimuth and roll angles of the simulation target are used to adjust the posture angles of the simulation model to maximize the angular similarity between the simulation image and the real image. The weighted sum of the cosine similarity between the angle vectors of the home hand training model obtained from measurement stations #1 and #2 and the angle vectors of the standard hospital model is used as the value of the first angular similarity term, S. sg For the first depth similarity term, the depth information of each key point obtained from measurement stations #1 and #2 is compared with the 3D depth information of the hospital's rehabilitation model. The simulated image region is used as a template, and a matching search is performed on the depth image of the actual image. By adjusting the depth angle and position parameters of the simulated target pair, the sum of the maximum values ​​of the overlapping pixel ratio of the two images from the two measurement stations is taken as the value of the first depth similarity term, S. EMG The first electromyographic signal similarity term is calculated by collecting electromyographic signals of the finger extensor and flexor muscle groups from the electromyographic sensor and comparing them with the electromyographic data in the hospital's standard model. The sum of the cosine similarity of the measured electromyographic signal vectors under the set action and those of the hospital's standard model is used as the value of the first electromyographic signal similarity term. α1, α2, and α3 are the weighting factors of the first angle similarity term, the first depth similarity term, and the first electromyographic signal similarity term, respectively, and are determined based on the flexion contracture assessment of stroke patients.

[0013] As a further aspect of the present invention, in the standard matching module, the multi-station fusion similarity based on limited interphalangeal joint extension is analyzed by weighting the second depth similarity term from the fingertip to the center of the palm, the depth and position similarity term of the finger joint, and the second electromyographic signal similarity term during the patient's rehabilitation training for limited interphalangeal joint extension. The multi-station fusion similarity analysis formula based on limited interphalangeal joint extension is as follows:

[0014] S zs =β1·S sd +β2·S sw +β3·S EMG ′

[0015] In the formula: S zs To achieve multi-station fusion similarity based on interphalangeal joint extension limitation, S d For the second depth similarity term, the sum of the ratios of the Euclidean distances between the fingertips and palm centers in the measured 3D models obtained from stations 1 and 2 and the hospital's standard model is used as the value of the second depth similarity term, S. sw For the depth and position similarity of the finger joints, the sum of the weighted sums of the Euclidean distance ratios between the metacarpophalangeal joint and the proximal and distal phalangeal joints, based on the measured 3D models from stations #1 and #2 and the hospital's standard model, is used as the numerical value for the depth and position similarity of the finger joints. EMGThe second electromyographic signal similarity term is the sum of the electromyographic signal ratios of the finger extensor muscle group in the standard surface model of the same hospital, which is measured as the second electromyographic signal similarity term. β1, β2, and β3 are the weighting factors of the second depth similarity term, the depth and position similarity term of the finger joint, and the second electromyographic signal similarity term, respectively, which are determined according to the finger joint limitation assessment of stroke patients.

[0016] As a further aspect of this invention, in the standard matching module, the multi-station fusion similarity analysis based on insufficient grip and pinch strength is performed by weighting the similarity items of pinch strength, grip strength, depth changes of fingers in grasping and pinching movements, and third electromyographic signal similarity items of the thenar and hypothenar muscles during the patient's rehabilitation training for insufficient grip and pinch strength. The formula for the multi-station fusion similarity analysis based on insufficient grip and pinch strength is as follows:

[0017] S wn =γ1·S n +γ2·S w +γ3·S g +γ4·S EMG "

[0018] In the formula: S wn To achieve multi-site fusion similarity based on insufficient grip and pinching strength, S n For the pinching force similarity term, the product of the measured pinching force and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting is divided by the product of the standard pinching force in the hospital standard model and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting. The resulting ratio is used as the value of the pinching force similarity term, S. w For the grip strength similarity term, the product of the measured grip strength and the duration of maintaining the measured grip strength within a set percentage of the standard grip strength is divided by the product of the standard grip strength and the duration of maintaining the standard grip strength within a set percentage in the hospital standard model. The resulting ratio is used as the value of the grip strength similarity term, S. g To determine the similarity term of finger depth changes during grasping and pinching actions, dynamic time warping analysis is used to analyze the depth change trajectory of the fingertip during these actions. A nonlinear comparison is then performed with a standard hospital model. The obtained depth trajectory matching degree is used as the numerical value of the similarity term of finger depth changes during grasping and pinching actions, S. EMG "The third electromyographic similarity term is defined as the cosine similarity between the measured electromyographic signals of the thenar and hypothenar muscles and the electromyographic signals of the thenar and hypothenar muscles in the hospital's standard model. "

[0019] As a further aspect of this invention, the recommended parameter generation module, based on flexion contracture rehabilitation training tasks, interphalangeal joint extension limitation rehabilitation training tasks, and insufficient grip and pinch strength rehabilitation training tasks, filters task-related feature data through task-driven gating 1 combined with task type features. This data is then input into an expert network, which includes expert 1, expert 2, and expert 3. Expert 1 analyzes angle and depth-related data to generate first angle similarity items, first depth similarity items, second depth similarity items, finger joint depth and position similarity items, and depth change similarity items, generating image similarity features. Expert 2 processes electromyographic (EMG) signal data, comparing the cosine similarity between measured EMG data and a hospital standard model to obtain first EMG signal similarity items, second EMG signal similarity items, and third EMG similarity items, generating EMG signal-related similarity features. Expert 3 analyzes grip and pinch strength data to obtain pinch strength similarity items and grip strength similarity items. Similarity features related to grip strength and pinch strength are generated. Each expert passes the extracted features to the commentator network. The commentator network evaluates the feature similarity of each expert's output according to the target standard model for each task, and obtains multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. Then, through task-driven gating 2, the features of stroke patient assessment reports are weighted and fused with the multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. The weighted fused sub-features are then aggregated to determine the priority of flexion contracture rehabilitation training tasks, interphalangeal joint extension limitation rehabilitation training tasks, and insufficient grip and pinch strength rehabilitation training tasks. According to the priority, the machine learning model generates the regulation parameters of the rehabilitation training tasks in sequence.

[0020] As a further aspect of the present invention, the recommended adjustment parameters for the rehabilitation training task generated by the recommended parameter generation module include:

[0021] Recommended parameters for flexion contracture training tasks: target angle range of interphalangeal joints, target flexion and extension angles of the wrist, frequency of flexion contracture training, hold time, movement speed, and resistance level;

[0022] Recommended adjustment parameters for limited interphalangeal joint extension: target depth range from the metacarpophalangeal joint to the fingertip, vertical depth distance range from the fingertip to the center of the palm, frequency of finger extension training, duration of holding at the maximum extension angle, resistance during finger extension, and speed of switching from flexion to extension.

[0023] Parameters for adjusting the rehabilitation training task for insufficient grip and pinch strength: target value of grip strength, target value of pinch strength, trajectory of change in depth of movement between them, training flatness, holding time at the set grip and pinch strength levels, and execution speed of grasping and pinching movements.

[0024] To address the technical problems raised in the background section, this invention proposes the following systemic technical effects: This invention utilizes task-driven gating filtering features, expert network-extracted feature similarity, commentator network-assessed multi-site fusion similarity, and weighted fusion features based on patient assessment reports to generate recommended parameters. This dynamically adjusts the priority and parameters of rehabilitation training tasks for flexion contractures, limited interphalangeal joint extension, and insufficient grip and pinch strength, thereby providing recommended control parameters for the execution of end effector training tasks. By combining a seamless care chain built from hospitals, communities, and homes, it solves the problem of inaccurately connecting with standard hospital rehabilitation data in existing home care. Attached Figure Description

[0025] Figure 1 This is a flowchart of a home-based hand training system for stroke patients based on a seamless transfer nursing model, according to the present invention.

[0026] Figure 2 This is a schematic diagram of the structure of a home-based hand training system for stroke patients based on a seamless transfer nursing model, according to the present invention. Figure 3 This is a comparative analysis table of hand grip strength between the control group and the group using the rehabilitation training robot in Reference 2; Figure 4 This is a system algorithm analysis diagram of a home-based hand training system for stroke patients based on a seamless transfer nursing model, as described in this invention. Detailed Implementation

[0027] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, not all of it. Based on the content of this invention, all other technical solutions obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0028] In Reference 1, an analysis of nursing data from a group of stroke patients hospitalized in my country, comparing the STCM (Self-Transfer Care) group and a control group, showed that the seamless transfer nursing model improved the quality of life of stroke patients, increased medication and dietary adherence, and reduced readmission rates. Reference 2, through keyword-based literature analysis, indicated that rehabilitation robot-assisted training is beneficial for the recovery of upper limb and hand functions in acute and subacute stroke patients. Reference 2 also pointed out that post-stroke hand dysfunction specifically manifests as flexion contractures, inability to extend interphalangeal and metacarpophalangeal joints, and inability to perform fine motor movements such as grasping, opposition, and finger opposition. Hand grip strength and pinching strength are important indicators of hand function; decreased muscle strength is one of the main manifestations of motor dysfunction in stroke patients, affecting not only fine motor skills but also daily living abilities. Based on these two references, the technical solution of this invention is proposed.

[0029] To achieve the above objectives, the present invention provides the following technical solution:

[0030] like Figures 1 to 4 As shown, this invention proposes a home-based hand training system for stroke patients based on a seamless transfer care model. The application background is the seamless care chain constructed by the seamless transfer care model for stroke patients in hospitals, communities, and homes, and the meta-analysis results of rehabilitation robot training for acute and subacute stroke patients based on published literature. The system includes an end effector, a data acquisition module, a standard matching module, a recommended parameter generation module, a remote monitoring feedback module, a progress tracking and evaluation module, a human-computer interaction module, and a data storage and analysis module. The data acquisition module is connected to the end effector and the standard matching module at both ends. The standard matching module is connected to the recommended parameter generation module, which is also connected to the remote monitoring feedback module and the human-computer interaction module. The human-computer interaction module is connected to the remote monitoring feedback module and the progress tracking and evaluation module. The estimation module is interconnected with the data storage and analysis module, which is also connected to the end effector, data acquisition module, standard matching module, recommended parameter generation module, remote monitoring and feedback module, progress tracking and evaluation module, and human-computer interaction module. The standard matching module maps the baseline points of the patient's hand measurement model to the baseline points of the standard model in space, so that the baseline points in the home training data match the baseline points of the standard model. By comparing the home training data of stroke patients' hands with the standard rehabilitation model data in the hospital, the accuracy of home hand training is judged based on feature similarity analysis. The recommended parameter generation module, based on the analysis results of the standard matching module, combined with the patient's training data and data development trends, as well as the patient's stroke hand dysfunction assessment report, predicts and generates recommended control parameters for each training task based on a machine learning model.

[0031] It should be noted that the rehabilitation training movements provided by the end effector for stroke patients include, but are not limited to, clenching the fist, extending all five fingers, extending the thumb, flexing the thumb, extending the index finger, flexing the index finger, extending the middle finger, flexing the middle finger, pinching with both fingertips, pinching with multiple fingertips, lateral pinching, ball grip, and cylindrical grip.

[0032] It should be noted that the data acquisition module includes image acquisition station #1, image acquisition station #2, and sensor data acquisition station. Image acquisition stations #1 and #2 both include motion image acquisition cameras and depth cameras. The sensor data acquisition station includes force sensors, angle sensors, position sensors, and electromyography (EMG) sensors. The electrodes of the EMG sensors are located in the thenar eminence, hypothenar eminence, extensor digitorum group, and flexor digitorum group, respectively. The key points captured by the image acquisition cameras include the wrist joint key point, thumb key point, non-thumb finger key points, and palm center key point. The depth camera acquires the depth position and angle of the metacarpophalangeal joints, proximal interphalangeal joints, and distal phalangeal joints, the depth relationship between the thumb and non-thumb fingers, the vertical depth distance between the palm center point and the fingertip, the depth distance and angle from the wrist to the finger, the depth difference value between fingers, and tracks the depth trajectory of each finger joint during training.

[0033] A single image acquisition station typically only captures motion data of the patient's hand from one angle, which can lead to issues such as occlusion of key points or incomplete depth information. Two image acquisition stations acquire images of the patient's hand from different angles (e.g., front and side views) to comprehensively cover the motion trajectories of key parts such as the wrist joint and fingers, avoiding information loss in single-angle acquisition and improving the spatial accuracy of depth and position data.

[0034] It should be noted that the feature similarity analysis of the standard matching module is based on data collected from image acquisition station #1, image acquisition station #2, and sensor data acquisition station. The analysis includes multi-station fusion similarity analysis based on flexion contracture, multi-station fusion similarity analysis based on limited interphalangeal joint extension, and multi-station fusion similarity analysis based on insufficient grip and pinching strength.

[0035] Flexion contractures prevent the fingers and wrists from extending naturally, severely affecting the patient's grasping and releasing functions, limiting daily living abilities (such as picking up objects and dressing). Prolonged flexion increases muscle tension, causing pain and exacerbating functional degeneration. Unrelieved flexion contractures lead to insufficient motivation for rehabilitation, affecting training effectiveness. Limited interphalangeal joint extension affects fine motor skills, such as button operation, writing, and holding objects. Long-term limited extension leads to joint deformities or further functional degeneration, affecting the patient's overall hand function recovery. Limited finger extension hinders patients' daily activities, reducing their independence and quality of life. Insufficient grip strength prevents patients from steadily grasping or lifting objects, affecting basic daily living abilities (such as holding a cup and opening a door). Insufficient pinch strength affects patients' ability to perform fine motor tasks (such as picking up coins and using chopsticks). Insufficient hand strength significantly reduces patients' independence and increases their dependence on others.

[0036] like Figure 1 and Figure 2As shown in Reference 1, the readmission frequency (p<0.05) and compliance (p<0.05) of the STCM group were lower than those of the control group. Community health centers used various different models to manage these individuals. Therefore, establishing a seamless care chain between hospitals, communities, and homes avoids unnecessary repeat examinations, shortens treatment time, reduces costs, and facilitates personal care for stroke patients.

[0037] Therefore, the system proposed in this invention collects standard rehabilitation training data of patients in hospitals through a data acquisition module, while community health centers collect data using portable devices to regularly update patients' rehabilitation training data. In home settings, it collects patients' hand rehabilitation training data and dynamically uploads it to a data storage and analysis module for analysis by hospitals, communities, and families, enabling cross-scenario data sharing. The analysis results are fed back to a recommendation parameter generation module and a remote monitoring feedback module, while the analysis results from the standard matching module are used for access and viewing by community service personnel and hospitals. This reflects the seamless nursing chain built by the seamless transfer model between families, communities, and hospitals.

[0038] like Figure 3 As shown in Reference 2, a comparative analysis of hand grip strength between the control group and the group using the rehabilitation training robot was conducted. The results indicated that hand grip strength was assessed in 208 stroke patients. Due to significant heterogeneity among the included studies, a random-effects model was used. The hand grip strength score in the experimental group was significantly higher than that in the control group. After excluding all studies, no further sources of heterogeneity were found. Ultimately, the rehabilitation training robot was found to improve upper limb and hand training in patients with acute and subacute stroke. Therefore, this invention addresses the automatic generation of recommended training parameters for rehabilitation training end effectors for conditions such as flexion contractures, limited interphalangeal joint extension, and insufficient grip and pinch strength, as proposed in the literature.

[0039] It should be noted that in the standard matching module, the multi-station fusion similarity analysis based on flexion contracture is obtained by weighted fusion of the first angle similarity, the first depth similarity, and the first electromyographic signal similarity in the patient's flexion contracture rehabilitation training. The formula for multi-station fusion similarity analysis based on flexion contracture is as follows:

[0040] S q =α1·S angle +α2·S sg +α3·S EMG

[0041] In the formula: S q To achieve multi-station fusion similarity based on flexion contracture, S angleFor the first angular similarity term, ideal joint angles of the hand and fingers are extracted from the standard hospital rehabilitation model, and angles are set in the 3D simulation. The azimuth and roll angles of the simulation target are used to adjust the posture angles of the simulation model to maximize the angular similarity between the simulation image and the real image. The weighted sum of the cosine similarity between the angle vectors of the home hand training model obtained from measurement stations #1 and #2 and the angle vectors of the standard hospital model is used as the value of the first angular similarity term, S. sg For the first depth similarity term, the depth information of each key point obtained from measurement stations #1 and #2 is compared with the 3D depth information of the hospital's rehabilitation model. The simulated image region is used as a template, and a matching search is performed on the depth image of the actual image. By adjusting the depth angle and position parameters of the simulated target pair, the sum of the maximum values ​​of the overlapping pixel ratio of the two images from the two measurement stations is taken as the value of the first depth similarity term, S. EMG The first electromyographic signal similarity term is calculated by collecting electromyographic signals of the finger extensor and flexor muscle groups from the electromyographic sensor and comparing them with the electromyographic data in the hospital's standard model. The sum of the cosine similarity of the measured electromyographic signal vectors under the set action and those of the hospital's standard model is used as the value of the first electromyographic signal similarity term. α1, α2, and α3 are the weighting factors of the first angle similarity term, the first depth similarity term, and the first electromyographic signal similarity term, respectively, and are determined based on the flexion contracture assessment of stroke patients.

[0042] Multi-station fusion similarity analysis based on flexion contracture integrates three aspects: angle, depth, and electromyographic signal. It comprehensively reflects the flexion contracture status of the hand in stroke patients through multidimensional assessment. Combining multi-station data improves the accuracy of analysis, and the importance of features is dynamically adjusted using weight factors, providing a data foundation for the recommendation parameter generation module.

[0043] It should be noted that in the standard matching module, the multi-station fusion similarity based on limited interphalangeal joint extension is analyzed by weighting the second depth similarity item from the fingertip to the center of the palm, the depth and position similarity item of the finger joint, and the second electromyographic signal similarity item during the patient's rehabilitation training for limited interphalangeal joint extension. The formula for multi-station fusion similarity analysis based on limited interphalangeal joint extension is as follows:

[0044] S zs =β1·S d +β2·S sw +β3·S EMG ′

[0045] In the formula: S zs To achieve multi-station fusion similarity based on interphalangeal joint extension limitation, S d For the second depth similarity term, the sum of the ratios of the Euclidean distances between the fingertips and palm centers in the measured 3D models obtained from stations 1 and 2 and the hospital's standard model is used as the value of the second depth similarity term, S.sw For the depth and position similarity of the finger joints, the sum of the weighted sums of the Euclidean distance ratios between the metacarpophalangeal joint and the proximal and distal phalangeal joints, based on the measured 3D models from stations #1 and #2 and the hospital's standard model, is used as the numerical value for the depth and position similarity of the finger joints. ENG The second electromyographic signal similarity term is the sum of the electromyographic signal ratios of the finger extensor muscle group in the standard surface model of the same hospital, which is measured as the second electromyographic signal similarity term. β1, β2, and β3 are the weighting factors of the second depth similarity term, the depth and position similarity term of the finger joint, and the second electromyographic signal similarity term, respectively, which are determined according to the finger joint limitation assessment of stroke patients.

[0046] Based on multi-station fusion similarity analysis of interphalangeal joint extension limitation, this method integrates second depth similarity terms, finger joint depth and position similarity terms, and second electromyographic signal similarity terms. It utilizes multi-station data to provide a comprehensive feature evaluation basis and dynamically adjusts the importance of features by combining weight factors. This provides a high-precision feature comparison and data analysis basis for the system's standard matching module, recommendation parameter generation module, and data storage and analysis module, supporting cross-scenario data management and dynamic optimization.

[0047] It should be noted that in the standard matching module, the multi-station fusion similarity analysis based on insufficient grip and pinch strength is performed by weighting the similarity items of pinch strength, grip strength, depth changes of fingers in grasping and pinching movements, and third electromyographic signal similarity items of the thenar and hypothenar muscles during the patient's rehabilitation training for insufficient grip and pinch strength. The formula for multi-station fusion similarity analysis based on insufficient grip and pinch strength is as follows:

[0048] S wn =γ1·S n +γ2·S w +γ3·S g +γ4·S EMG "

[0049] In the formula: S wn To achieve multi-site fusion similarity based on insufficient grip and pinching strength, S n For the pinching force similarity term, the product of the measured pinching force and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting is divided by the product of the standard pinching force in the hospital standard model and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting. The resulting ratio is used as the value of the pinching force similarity term, S. w For the grip strength similarity term, the product of the measured grip strength and the duration of maintaining the measured grip strength within a set percentage of the standard grip strength is divided by the product of the standard grip strength and the duration of maintaining the standard grip strength within a set percentage in the hospital standard model. The resulting ratio is used as the value of the grip strength similarity term, S. gTo determine the similarity term of finger depth changes during grasping and pinching actions, dynamic time warping analysis is used to analyze the depth change trajectory of the fingertip during these actions. A nonlinear comparison is then performed with a standard hospital model. The obtained depth trajectory matching degree is used as the numerical value of the similarity term of finger depth changes during grasping and pinching actions, S. EMG "The third electromyographic similarity term is defined as the cosine similarity between the measured electromyographic signals of the thenar and hypothenar muscles and the electromyographic signals of the thenar and hypothenar muscles in the hospital's standard model. "

[0050] Multi-station fusion similarity analysis based on insufficient grip and pinch strength integrates similarity items for pinch strength, grip strength, finger depth changes, and electromyography signals. Utilizing dynamic time warping and nonlinear comparison methods, it performs unified analysis on data collected from multiple stations, dynamically adjusting feature weights to comprehensively assess the correlation and matching degree between multidimensional data. This provides accurate feature comparison results and a reliable data foundation for the standard matching module, recommendation parameter generation module, and data storage and analysis module. It also supports cross-scenario data management optimization, ensuring the consistency and accuracy of data sharing across different scenarios.

[0051] Dynamic time warping is a method for measuring the similarity of two time series, even if they exhibit non-linear deformations along the time axis. It is suitable for comparing sequences of different durations or speeds, such as analyzing the depth change trajectory of fingers during grasping and pinching movements. This method compares the depth change trajectories of finger grasping and pinching movements collected from measurement stations #1 and #2, resolving time misalignment issues caused by differences in acquisition angle and hand speed. Non-linear alignment is a method for comparing two datasets through non-linear transformations. It can adapt to data deformations in space, time, or scale, improving the accuracy and robustness of data matching. This method is used to compare electromyographic signals, depth, and position data, resolving bias issues caused by differences in data distribution or inconsistent acquisition precision from different stations. By standardizing, aligning, and fusing data collected from different measurement stations, biases between multi-station data are eliminated, enabling unified analysis of multi-dimensional features. The implementation process is as follows:

[0052] Data standardization: Normalize data collected from multiple stations, mapping the collected values ​​from different measurement stations to the same feature range;

[0053] Multi-station data alignment: Dynamic time warping (DTW) is used to align time-related data, and non-linear alignment methods are used to adjust the spatial distribution of multidimensional features (such as depth, location, and electromyography signals).

[0054] Weighted fusion: Based on the importance of each feature, weights are assigned to achieve weighted fusion and generate multi-site fusion similarity;

[0055] Output unified analysis results: Provide unified feature similarity results, such as pinch force similarity, grip force similarity, finger depth change similarity, and electromyographic signal similarity, to provide high-quality data support for subsequent modules.

[0056] It should be noted that, as Figure 3 As shown, the recommended parameter generation module proposed in this invention is based on flexion contracture rehabilitation training tasks, interphalangeal joint extension limitation rehabilitation training tasks, and grip and pinch strength deficiency rehabilitation training tasks. It filters task-related feature data through task-driven gating 1 combined with task type features, and inputs this data into an expert network. The expert network includes expert 1, expert 2, and expert 3. Expert 1 analyzes angle and depth-related data to generate first angle similarity items, first depth similarity items, second depth similarity items, finger joint depth and position similarity items, and depth change similarity items, generating image similarity features. Expert 2 processes electromyographic signal data, compares the cosine similarity between measured electromyographic data and hospital standard models, obtains first electromyographic signal similarity items, second electromyographic signal similarity items, and third electromyographic similarity items, generating electromyographic signal-related similarity features. Expert 3 analyzes grip and pinch strength data, obtains pinch strength similarity items and grip strength similarity items, generating... For grip strength and pinch strength related similar features, each expert transmits the extracted features to the commentator network. The commentator network evaluates the feature similarity of each expert's output according to the target standard model for each task, and obtains multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. Then, through task-driven gating 2, the features of stroke patient assessment reports are weighted and fused with the multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. The weighted fused sub-features are then aggregated to determine the priority of flexion contracture rehabilitation training tasks, interphalangeal joint extension limitation rehabilitation training tasks, and insufficient grip and pinch strength rehabilitation training tasks. According to the priority, the machine learning model generates the regulation parameters of the rehabilitation training tasks in sequence.

[0057] The recommended parameter generation module filters task-related feature data through task-driven gating, generates multi-dimensional feature similarity through expert network analysis, and performs weighted fusion with multi-site fusion similarity and patient assessment reports through commentator network. This dynamically adjusts the priority of training tasks and generates precise task control parameters in sequence, ensuring high consistency between feature data and the target model. This provides a scientific basis for personalized task parameter optimization, while improving data analysis accuracy, task allocation efficiency, and the relevance of parameter generation, enabling cross-task data management and dynamic optimization.

[0058] It should be noted that the recommended adjustment parameters for the rehabilitation training task generated by the recommended parameter generation module include:

[0059] Recommended parameters for flexion contracture training tasks: target angle range of interphalangeal joints, target flexion and extension angles of the wrist, frequency of flexion contracture training, hold time, movement speed, and resistance level;

[0060] Recommended adjustment parameters for limited interphalangeal joint extension: target depth range from the metacarpophalangeal joint to the fingertip, vertical depth distance range from the fingertip to the center of the palm, frequency of finger extension training, duration of holding at the maximum extension angle, resistance during finger extension, and speed of switching from flexion to extension.

[0061] Parameters for adjusting the rehabilitation training task for insufficient grip and pinch strength: target value of grip strength, target value of pinch strength, trajectory of change in depth of movement between them, training flatness, holding time at the set grip and pinch strength levels, and execution speed of grasping and pinching movements.

[0062] This invention utilizes task-driven gating filtering features, expert network-extracted feature similarity, commentator network-assessed multi-site fusion similarity, and weighted fusion features based on patient assessment reports to generate recommended parameters. It dynamically adjusts the priority and parameters of rehabilitation training tasks for flexion contractures, limited interphalangeal joint extension, and insufficient grip and pinch strength, thereby providing recommended control parameters for the execution of training tasks for the end effector. By combining a seamless care chain built from hospitals, communities, and homes, it solves the problem in existing home care that it cannot accurately connect with standard hospital rehabilitation data and cannot provide control guidance parameters for the end effector of rehabilitation training.

[0063] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0064] In conclusion, the above description is merely a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A home-based hand training system for stroke patients based on a seamless transfer care model, the application background being the seamless care chain constructed by the seamless transfer care model for stroke patients in hospitals, communities, and homes, and the meta-analysis results of rehabilitation robot training for acute and subacute stroke patients based on publicly available literature, characterized in that... The system includes an end effector, a data acquisition module, a standard matching module, a recommended parameter generation module, a remote monitoring feedback module, a progress tracking and evaluation module, a human-machine interaction module, and a data storage and analysis module. The data acquisition module is connected to the end effector and the standard matching module at both ends. The standard matching module is connected to the recommended parameter generation module. The recommended parameter generation module is connected to both the remote monitoring feedback module and the human-machine interaction module. The human-machine interaction module is connected to both the remote monitoring feedback module and the progress tracking and evaluation module. The data storage and analysis module is connected to the end effector, the data acquisition module, the standard matching module, the recommended parameter generation module, and the remote monitoring feedback module. The progress tracking and assessment module and the human-computer interaction module are interconnected. The standard matching module maps the baseline points of the patient's hand measurement model to the baseline points of the standard model in space, so that the baseline points in the home training data match the baseline points of the standard model. By comparing the home training data of stroke patients' hands with the standard rehabilitation model data in the hospital, the accuracy of home hand training is judged based on feature similarity analysis. The recommended parameter generation module, based on the analysis results of the standard matching module, combined with the patient's training data and data development trends, as well as the patient's stroke hand dysfunction assessment report, predicts and generates recommended control parameters for each training task based on the machine learning model.

2. The home-based hand training system for stroke patients based on a seamless transfer care model according to claim 1, characterized in that, The rehabilitation exercises provided by the end effector for stroke patients include, but are not limited to, clenching the fist, extending all five fingers, extending the thumb, flexing the thumb, extending the index finger, flexing the index finger, extending the middle finger, flexing the middle finger, pinching with both fingertips, pinching with multiple fingertips, lateral pinching, ball grip, and cylindrical grip.

3. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 2, characterized in that, The data acquisition module includes image acquisition station #1, image acquisition station #2, and sensor data acquisition station. Image acquisition stations #1 and #2 both include motion image acquisition cameras and depth cameras. The sensor data acquisition station includes force sensors, angle sensors, position sensors, and electromyography (EMG) sensors. The electrodes of the EMG sensors are located in the thenar eminence, hypothenar eminence, extensor digitorum group, and flexor digitorum group, respectively. The key points captured by the image acquisition cameras include the wrist joint key point, thumb key point, non-thumb finger key points, and palm center key point. The depth camera acquires the depth position and angle of the metacarpophalangeal joints, proximal interphalangeal joints, and distal phalangeal joints, the depth relationship between the thumb and non-thumb fingers, the vertical depth distance between the palm center point and the fingertip, the depth distance and angle from the wrist to the finger, the depth difference value between fingers, and tracks the depth trajectory of each finger joint during training.

4. A home-based hand training system for stroke patients based on a seamless transfer care model, as described in claim 3, is characterized in that... The feature similarity analysis of the standard matching module is conducted using data collected from image acquisition station #1, image acquisition station #2, and sensor data acquisition station. The analysis includes multi-station fusion similarity analysis based on flexion contracture, multi-station fusion similarity analysis based on limited interphalangeal joint extension, and multi-station fusion similarity analysis based on insufficient grip and pinching strength.

5. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 4, characterized in that, In the standard matching module, the multi-station fusion similarity analysis based on flexion contracture is obtained by weighted fusion of the first angle similarity, first depth similarity, and first electromyographic signal similarity in the patient's flexion contracture rehabilitation training. The formula for multi-station fusion similarity analysis based on flexion contracture is as follows: S q =α1·S angle +α2·S sg +α3·S EMG In the formula: S q To achieve multi-station fusion similarity based on flexion contracture, S angle For the first angular similarity term, ideal joint angles of the hand and fingers are extracted from the standard hospital rehabilitation model, and angles are set in the 3D simulation. The azimuth and roll angles of the simulation target are used to adjust the posture angles of the simulation model to maximize the angular similarity between the simulation image and the real image. The weighted sum of the cosine similarity between the angle vectors of the home hand training model obtained from measurement stations #1 and #2 and the angle vectors of the standard hospital model is used as the value of the first angular similarity term, S. sg For the first depth similarity term, the depth information of each key point obtained from measurement stations #1 and #2 is compared with the 3D depth information of the hospital's rehabilitation model. The simulated image region is used as a template, and a matching search is performed on the depth image of the actual image. By adjusting the depth angle and position parameters of the simulated target pair, the sum of the maximum values ​​of the overlapping pixel ratio of the two images from the two measurement stations is taken as the value of the first depth similarity term, S. EMG The first electromyographic signal similarity term is calculated by collecting electromyographic signals of the finger extensor and flexor muscle groups from the electromyographic sensor and comparing them with the electromyographic data in the hospital's standard model. The sum of the cosine similarity of the measured electromyographic signal vectors under the set action and those of the hospital's standard model is used as the value of the first electromyographic signal similarity term. α1, α2, and α3 are the weighting factors of the first angle similarity term, the first depth similarity term, and the first electromyographic signal similarity term, respectively, and are determined based on the flexion contracture assessment of stroke patients.

6. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 5, characterized in that, In the standard matching module, the multi-station fusion similarity based on limited interphalangeal joint extension is analyzed by weighting the second depth similarity term from the fingertip to the center of the palm, the depth and position similarity term of the finger joint, and the second electromyographic signal similarity term during the patient's rehabilitation training for limited interphalangeal joint extension. The formula for multi-station fusion similarity analysis based on limited interphalangeal joint extension is as follows: S zs =β1·S d +β2·S sw +β3·S EMG ′ In the formula: S zs To achieve multi-station fusion similarity based on interphalangeal joint extension limitation, S d For the second depth similarity term, the sum of the ratios of the Euclidean distances between the fingertips and palm centers in the measured 3D models obtained from stations 1 and 2 and the hospital's standard model is used as the value of the second depth similarity term, S. sw For the depth and position similarity of the finger joints, the sum of the weighted sums of the Euclidean distance ratios between the metacarpophalangeal joint and the proximal and distal phalangeal joints, based on the measured 3D models from stations #1 and #2 and the hospital's standard model, is used as the numerical value for the depth and position similarity of the finger joints. EMG The second electromyographic signal similarity term is the sum of the electromyographic signal ratios of the finger extensor muscle group in the standard surface model of the same hospital, which is measured as the second electromyographic signal similarity term. β1, β2, and β3 are the weighting factors of the second depth similarity term, the depth and position similarity term of the finger joint, and the second electromyographic signal similarity term, respectively, which are determined according to the finger joint limitation assessment of stroke patients.

7. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 6, characterized in that, In the standard matching module, the multi-station fusion similarity analysis based on insufficient grip and pinch strength is performed by weighting the similarity items of pinch strength, grip strength, depth changes of fingers in grasping and pinching movements, and third electromyographic signal similarity items of the thenar and hypothenar muscles during the rehabilitation training for patients with insufficient grip and pinch strength. The formula for multi-station fusion similarity analysis based on insufficient grip and pinch strength is as follows: S wn =γ1·S n +γ2·S w +γ3·S g +γ4·S EMG ″ In the formula: S wn To achieve multi-site fusion similarity based on insufficient grip and pinching strength, S n For the pinching force similarity term, the product of the measured pinching force and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting is divided by the product of the standard pinching force in the hospital standard model and the duration for which the measured pinching force is maintained within a certain percentage of the standard pinching force setting. The resulting ratio is used as the value of the pinching force similarity term, S. w For the grip strength similarity term, the product of the measured grip strength and the duration of maintaining the measured grip strength within a set percentage of the standard grip strength is divided by the product of the standard grip strength and the duration of maintaining the standard grip strength within a set percentage in the hospital standard model. The resulting ratio is used as the value of the grip strength similarity term, S. g To determine the similarity term of finger depth changes during grasping and pinching actions, dynamic time warping analysis is used to analyze the depth change trajectory of the fingertip during these actions. A nonlinear comparison is then performed with a standard hospital model. The obtained depth trajectory matching degree is used as the numerical value of the similarity term of finger depth changes during grasping and pinching actions, S. EMG "The third electromyographic similarity term is defined as the cosine similarity between the measured electromyographic signals of the thenar and hypothenar muscles and the electromyographic signals of the thenar and hypothenar muscles in the hospital's standard model. " 8. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 7, characterized in that, The recommended parameter generation module is based on rehabilitation training tasks for flexion contracture, limited interphalangeal joint extension, and insufficient grip and pinch strength. It filters task-related feature data using task-driven gating 1 combined with task type features, and inputs this data into an expert network. The expert network includes Expert 1, Expert 2, and Expert 3. Expert 1 analyzes angle and depth-related data, generating first angle similarity items, first depth similarity items, second depth similarity items, finger joint depth and position similarity items, and depth change similarity items, thus generating image similarity features. Expert 2 processes electromyographic (EMG) signal data, comparing the cosine similarity between measured EMG data and a hospital standard model, obtaining first EMG signal similarity items, second EMG signal similarity items, and third EMG similarity items, generating EMG signal-related similarity features. Expert 3 analyzes grip and pinch strength data, obtaining pinch strength similarity items and grip strength similarity items, generating grip and pinch strength-related similarity features. Each expert transmits the extracted features to the commentator network. The commentator network evaluates the feature similarity of each expert's output according to the target standard model for each task, and obtains the multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. Then, through task-driven gating 2, the features of the stroke patient assessment report are weighted and fused with the multi-station fusion similarity based on flexion contracture, interphalangeal joint extension limitation, and insufficient grip and pinch strength. The weighted fused sub-features are then aggregated to determine the priority of the flexion contracture rehabilitation training task, interphalangeal joint extension limitation rehabilitation training task, and insufficient grip and pinch strength rehabilitation training task. According to the priority, the machine learning model generates the regulation parameters of the rehabilitation training tasks in sequence.

9. A home-based hand training system for stroke patients based on a seamless transfer care model according to claim 7, characterized in that, The recommended adjustment parameters for the rehabilitation training task generated by the recommended parameter generation module include: Recommended parameters for flexion contracture training tasks: target angle range of interphalangeal joints, target flexion and extension angles of the wrist, frequency of flexion contracture training, hold time, movement speed, and resistance level; Recommended adjustment parameters for limited interphalangeal joint extension: target depth range from the metacarpophalangeal joint to the fingertip, vertical depth distance range from the fingertip to the center of the palm, frequency of finger extension training, duration of holding at the maximum extension angle, resistance during finger extension, and speed of switching from flexion to extension. Parameters for adjusting the rehabilitation training task for insufficient grip and pinch strength: target value of grip strength, target value of pinch strength, trajectory of change in depth of movement between them, training flatness, holding time at the set grip and pinch strength levels, and execution speed of grasping and pinching movements.

Images