Rehabilitation training system and method for upper limbs of apoplexy patient

Abstract

The invention discloses an upper limb rehabilitation training system and method for a stroke patient. Relates to the technical field of medical rehabilitation. According to the stroke upper limb rehabilitation training system and method, the limbs are driven by a machine to do repeated movement, a nervous system for controlling limb movement is stimulated and reconstructed, and therefore the movement function of the limbs is recovered, and the rehabilitation training effect of the limbs is improved. On the basis of an existing motion tracking and feedback technology, the system and the method are improved, real-time data analysis is carried out based on a sensor, a brain-computer interface technology and a cloud platform, influences of operation environment information and motion performance are considered in individualized requirements of patients, individual differences and environment variables are comprehensively considered in the rehabilitation process, and the rehabilitation effect is improved. A patient can perform rehabilitation training at any place, a rehabilitation physician, the patient and family members can check the rehabilitation condition of the patient at any time, and the precise training system and method suitable for various environments and changeable rehabilitation requirements are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical rehabilitation, and in particular to an upper limb rehabilitation training system and method for stroke patients. Background Art

[0002] Stroke is a cerebral blood circulation disorder caused by acute rupture or blockage of cerebral blood vessels. The clinical manifestation is hemiplegia. Stroke often leads to upper limb dysfunction, which has a serious impact on the patient's quality of life. Rehabilitation training for stroke patients is based on the theory of neuroplasticity. Traditional upper limb rehabilitation training usually relies on the guidance of physical therapists or rehabilitation physicians, and the training process is relatively simple, which makes it difficult to stimulate patients' active participation and long-term persistence. Although there are some rehabilitation devices, most of them still need to be improved in terms of intelligence, personalization and interactivity.

[0003] In stroke rehabilitation training, the process of upper limb function recovery usually relies on various physical therapies and sports training systems. In fact, due to the limited motor ability and diverse rehabilitation needs of stroke patients, it is difficult to achieve completely personalized and efficient training. In the process of rehabilitation training, the differences in patients' physical conditions and motor abilities will inevitably lead to different degrees of training effect differences, which makes the rehabilitation system have a certain degree of uneven effect and mismatch of training intensity. This uneven effect will seriously affect the overall performance of the system in cases where the comprehensiveness and consistency of the rehabilitation effect are affected.

[0004] Traditional rehabilitation systems can cover a variety of scenarios and situations. Their main purpose is to quickly adapt to the specific needs of patients during rehabilitation training and ensure the safety and coordination of the rehabilitation process. However, in the actual rehabilitation training scenario, if the training intensity and content cannot be adjusted in real time, it will have a serious negative impact on the patient's recovery effect, thereby affecting the real-time, accuracy and reliability of the overall rehabilitation training, and may even affect the patient's long-term rehabilitation effect. Therefore, it is very important to ensure the flexibility and adaptability of the rehabilitation system. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide an upper limb rehabilitation training for stroke patients. The technical problem to be solved is that traditional upper limb rehabilitation training methods for stroke patients are mainly based on standardized movements and preset rehabilitation plans. However, there are large errors in the application of actual clinical scenarios, which may affect the evaluation of rehabilitation effects, especially when the patient's motor ability may change at any time and the patient needs to go to a designated hospital for offline rehabilitation. The present invention specifically solves the professional requirements of the existing rehabilitation training for the rehabilitation location, and the patient needs to be trained in a special rehabilitation hospital. The present invention provides a stroke upper limb rehabilitation training system and method. The present invention drives the limbs to perform repetitive movements through a machine, stimulates and reconstructs the nervous system that controls the movement of the limbs, thereby restoring the motor function of the limbs. On the basis of the existing motion tracking and feedback technology, improvements are made to the system and methods, based on real-time data analysis of sensors, brain-computer interface technology and cloud platforms, and the impact of operating environment information and motor performance is considered in the individualized needs of patients. Individual differences and environmental variables are comprehensively considered in the rehabilitation process. Patients can choose any location for rehabilitation training, and rehabilitation physicians, patients and their families can also check the patient's rehabilitation status at any time, providing a precise training system and method that is adaptable to various environments and changing rehabilitation needs.

[0006] The purpose of the present invention is mainly achieved through the following technical solutions: The present invention provides an upper limb rehabilitation training system for stroke patients, comprising: The patient terminal, cloud platform server and rehabilitation physician terminal are connected in sequence.

[0007] The patient terminal includes rehabilitation training equipment, a virtual reality platform, a virtual reality platform and a patient interaction platform; the rehabilitation training equipment and the virtual reality platform are electrically connected to the central control device, and the rehabilitation training equipment and the virtual reality platform receive action signals from the central control device and execute them.

[0008] The virtual reality platform consists of a VR headset worn on the patient's head.

[0009] Rehabilitation training equipment includes mechanical gloves, adjustable support devices, electroencephalogram equipment, neuromuscular electrical stimulation equipment and data acquisition devices.

[0010] The mechanical glove includes a sensor group, an actuator and a motor; the adjustable support device supports the training of the arm at different angles and heights through an electric adjustment mechanism, and the adjustable support device is installed with a sensor group; the sensor group collects the height, angle and strength of the patient's upper limb movements to form motion data.

[0011] EEG equipment uses brain-computer interface (BCI) technology to perform signal encoding, signal acquisition, signal decoding, and signal output in sequence.

[0012] Signal encoding refers to the process in which the patient performs internal thinking activities in the brain according to the instructions displayed on the VR headset. The brain sends out the intention to control the movement of the upper limbs according to the instructions displayed on the VR headset, which will produce various physiological reactions, and the reactions are manifested in the form of EEG signals.

[0013] EEG signal is a kind of physiological electrical signal. In the signal acquisition process, an EEG signal acquisition system is used to collect and store EEG signals.

[0014] Signal decoding is the process of preprocessing, feature extraction and classification of EEG signals collected during the signal acquisition process to identify the brain's control intention as the recognition result.

[0015] The recognition result is the patient's actual control instruction, which is used as EEG data and transmitted to an external data acquisition device. The data acquisition device is electrically connected to the central control device, and the corresponding instruction action execution signal is sent to the mechanical gloves and adjustable support device of the rehabilitation training equipment through the central control device, and the mechanical gloves and adjustable support device perform the corresponding action; the neuromuscular electrical stimulation equipment includes an electrode sheet, a power module and a communication module. The electrode sheet is attached to the patient's upper limb muscles. The communication module receives the control signal of the central control device, and the electrode sheet discharges accordingly. By electrically stimulating the upper limb muscles of the patient corresponding to the indicated action, the speed of upper limb motor nerve reconstruction is accelerated; the electrode sheet and the communication module are electrically connected to the power module respectively, and the power module supplies power to the electrode sheet and the communication module.

[0016] The cloud platform server includes a data storage system and a data analysis system.

[0017] The data storage system includes a patient history database that records patient movement data, a rehabilitation plan database that records patient rehabilitation plan data, and a healthy volunteer database that records healthy volunteer movement data. The movement data includes data on hand position, movement amplitude, and force output for completing indicated movements.

[0018] The data analysis system includes a preamplifier circuit, a motion information extraction unit and a similarity measurement unit which are electrically connected in sequence; the motion data in the patient history database is amplified by the preamplifier circuit, and the amplified data is sent to the motion information extraction unit to extract the feature vector of the motion data through a deep learning model, and the feature vector is calculated by the similarity measurement unit to obtain the difference value between the patient's skeleton motion feature vector and the healthy volunteer's feature vector; a threshold is set, and when the difference value is less than the set threshold, it is judged that the action meets the requirements.

[0019] The central control device detects that the data collected by the data acquisition device is abnormal, triggers the alarm device of the patient terminal and the alarm device of the rehabilitation physician terminal, notifies the rehabilitation physician to provide timely feedback, and sends a signal to the central control device to stop the rehabilitation training equipment and the virtual reality platform, thereby controlling the rehabilitation training equipment and the virtual reality platform to stop working.

[0020] The rehabilitation physician checks the patient's historical database through the rehabilitation physician terminal, formulates a rehabilitation plan based on the patient's rehabilitation data, transmits the rehabilitation plan data to the rehabilitation plan database, and controls the patient terminal to execute the indicated actions of the rehabilitation plan.

[0021] The specific operation steps of a stroke patient upper limb rehabilitation training method applicable to the system of the present invention are as follows: Step S1: After the patient completes the device configuration and wears it correctly, the patient starts the rehabilitation training device for rehabilitation training, and conducts scenario-based rehabilitation training in the virtual reality platform. The rehabilitation training device works accordingly to accelerate rehabilitation, and at the same time synchronizes the patient's training data to the data storage system of the cloud platform server.

[0022] Step S2: The rehabilitation physician views the patient's rehabilitation data on the computer of the rehabilitation physician terminal through the cloud platform server, and the rehabilitation physician formulates a training plan. The central control device receives the rehabilitation physician's plan and forwards it to the rehabilitation training equipment and virtual reality platform of the patient terminal.

[0023] Step S3: The cloud platform server realizes data storage and interaction. Patients view rehabilitation data and conduct social interactions through mobile terminals. Rehabilitation physicians and patients communicate online through the cloud platform to discuss rehabilitation progress and next steps, and provide remote guidance and support.

[0024] Step S4: At the end of the training, the training data is backed up to the cloud platform server for subsequent analysis and historical record review, and a training summary report is generated and reviewed.

[0025] A further optimization plan is that the signal acquisition of the EEG device adopts the EEG signal module in the brain-computer interface (BCI) technology. The EEG signal module is a non-invasive EEG signal acquisition technology. The EEG signal module is in the form of an EEG cap. There are electrodes distributed on the surface of the EEG cap. The electrodes are made of metal. The metal electrodes are in contact with the human brain scalp. According to the principle of current conduction, the EEG signal is transmitted to the EEG signal acquisition device for amplification and conversion. The conductivity of the metal can collect tiny changes in EEG signals on the scalp surface. When the patient completes signal encoding by sending action instructions to the brain according to the instructions of the VR headset, the EEG signal module captures, collects and stores the EEG signal, and then decodes the signal; signal decoding includes signal preprocessing, feature extraction and classification; a large number of EEG signals of known categories are collected, their power spectral density PSD matrices are calculated respectively, and the corresponding PSD matrix set is constructed according to the category. By establishing the PSD matrix, the conversion of the EEG signal from Euclidean space to Riemann space is completed; for the EEG signal collected by the signal acquisition EEG signal module, its PSD matrix is ​​calculated, and the Riemann distance between the PSD matrix and the PSD matrix in the set of all known category PSD matrices is measured; finally, the K nearest neighbor classification algorithm is used to determine the category to which the EEG signal of unknown category belongs, so as to identify the intention of brain control as the recognition result, and output the corresponding action signal to the data analysis system; collecting a large number of EEG signals of known categories is to collect the EEG signals generated by a large number of volunteers completing corresponding actions according to specified instructions.

[0026] A further optimization plan is that the sensors embedded in the mechanical gloves include force sensors, displacement sensors, angle sensors and acceleration sensors. Four sensors form a group, and each group of sensors is embedded in the fingers, palms and wrists to monitor the hand's range of motion, strength and frequency in real time.

[0027] Motors and actuators are used to achieve active feedback on hand movements. The motors and actuators simulate resistance of different intensities according to the environmental requirements of the virtual reality platform to provide targeted sports training.

[0028] The motors and actuators simulate resistance according to the settings of the data analysis system, and the data analysis system processes the sensor data in real time, controls the operation of the motors, adjusts the training intensity and mode, and adjusts the feedback intensity according to the sensor feedback.

[0029] A further optimization plan is that the electric adjustment mechanism of the adjustable support device adopts a servo motor to achieve height and angle adjustment of the adjustable support device, and the servo motor accurately adjusts the posture of the adjustable support device through the data analysis system; a sensor group is installed on the adjustable support device to detect and record the angle, displacement and acceleration of the adjustable support device to ensure stability and consistency during training; the adjustable support device is electrically connected to the data analysis system to realize automatic control and adjustment of the electric adjustment mechanism of the adjustable support device, or remote control through the rehabilitation physician terminal.

[0030] A further optimization scheme is that the motion information extraction unit convolves the motion data to identify the feature vector of the patient's skeleton movement, which specifically includes: Construct an undirected skeleton graph based on the human skeleton, bring the enlarged data into the skeleton graph and encode it into a matrix; N Represents the number of joint points, that is,N The skeleton of joint points is represented as an undirected skeleton graph ,in Represents all the vertices of this skeleton graph, that is, all the joint points in the skeleton; E The edges of the skeleton graph represent the physical connections between the joints, and the dependencies between the joints are encoded as a shape of The adjacency matrix of ; The feature vector is extracted through the ST-GCN spatiotemporal graph convolutional network model of the deep learning model; the ST-GCN spatiotemporal graph convolutional network model defines the implementation of graph convolution in skeleton data, so that the human skeleton action sequence is represented as ,in C IN Indicates the number of input channels, T Indicates the number of skeleton graph frames, N Represents the number of joints. The implementation process of graph convolution based on skeleton data can be expressed as follows: For the input skeleton data, the traditional graph convolution uses weights W Transform features and then pass a ij polymerization v i The features of adjacent vertices are used to update the i The output feature of each joint is , in represents the normalized adjacency matrix; represents a normalized diagonal matrix; represents the weight matrix, C OUT Indicates the number of output channels; L Initialize the matrix to a matrix with all elements set to 1; Indicates calculation of the Hadamard product.

[0031] A further optimization scheme is that the similarity measurement unit uses the Euclidean distance to calculate the corresponding similarity score to obtain the difference between the patient's skeleton motion data and the healthy volunteer data. Under the specified action, the patient's skeleton motion feature vector is , the corresponding healthy volunteer feature vector is , calculate the Euclidean distance of the feature vector ,The similarity score is calculated by taking the weighted average of the Euclidean distance of all feature vectors to obtain the difference value between the feature vector of the patient’s skeleton motion and that of the healthy volunteer.

[0032] The further optimization plan is to develop the virtual training environment of the virtual reality platform using the Unity3D game engine to develop virtual training scenes; create a variety of training tasks, including virtual object grabbing, moving and placing, simulating real-world actions and environments; and equip it with a VR headset connected to a mechanical glove; the VR headset is used to display virtual scenes, and the mechanical gloves are used for patients to control virtual objects with their movements to complete tasks.

[0033] The real-time feedback of the virtual reality platform also includes motion tracking. Motion tracking integrates motion tracking technology in the VR system. The built-in tracker of the glove and the sensor embedded in the glove capture the patient's hand movement in real time. The motion data of the hand movement is calculated and processed in real time by the cloud platform server, and feedback is generated in the virtual training environment to guide the patient to perform correct movements. Instant visual and auditory feedback is provided based on the patient's motion data, including guidance prompts, progress display and encouraging sounds to complete real-time feedback motion correction.

[0034] A further optimization plan is that the patient interaction platform uses mobile applications on mobile terminals as the patient human-computer interaction interface, including social functions, adaptive feedback and data viewing functions.

[0035] The mobile application provides a patient interface for patients to view training plans, progress reports, and receive training guidance.

[0036] Social features allow patients to communicate with other recovering people.

[0037] The adaptive feedback device automatically adjusts the training intensity and difficulty according to the real-time motion data. Patients can change the rehabilitation plan and choose different training movements through the mobile application, and rehabilitation physicians can change the rehabilitation plan through the physician terminal. The data viewing function can view the data in the data storage system.

[0038] The patient interaction platform is developed using the ReactNative framework, integrating API interfaces to achieve connection with the cloud platform server; and designs an intuitive patient interface and patient experience.

[0039] A further optimization scheme is that step S1 specifically includes: Step S11: The patient starts the VR headset, wears the VR headset, and enters the virtual training environment; the patient practices the virtual training scene in the virtual reality platform; the VR headset provides real-time feedback movement correction, and helps the patient correct movements through visual and auditory feedback.

[0040] Step S12: In the EEG device, the patient performs thinking activities according to the tasks in the virtual training environment. The EEG device encodes the control signals sent by the brain and performs signal encoding. The EEG device collects and decodes signals, extracts the brain's control intention for upper limb movements, and converts them into instructions, and transmits the signals to the central control device. The central control device adjusts the movements of the rehabilitation training equipment according to the instructions to complete the movement control.

[0041] Step S13: The rehabilitation training equipment is controlled by the command of the central control device to perform operations; the mechanical gloves monitor the hand movement range, strength and frequency through sensors, and adjust the training intensity according to the feedback; the adjustable support device electrically adjusts the angle, displacement and acceleration of the arm movement according to the training needs; neuromuscular electrical stimulation, the electrode sheet electrically stimulates the patient's upper limbs to promote motor nerve reconstruction and accelerate the rehabilitation process.

[0042] Step S14: While the patient is undergoing rehabilitation training, the data storage system records the patient's movement data during the training process, including hand position, range of motion and force output; the data analysis system uses artificial intelligence algorithms to analyze the training data, evaluate the rehabilitation progress, and generate personalized training reports and adjustment suggestions; if abnormal data occurs, when the data is monitored to be abnormal, the alarm device is triggered to notify the rehabilitation physician and control the equipment to stop working.

[0043] Through the upper limb rehabilitation training system and method for stroke patients of the present invention, patients can carry out personalized training at home and reduce their dependence on physical therapists. Virtual reality and intelligent feedback enhance the fun and effectiveness of training, while the data analysis system can be synchronized to the hospital in real time through the cloud platform and analyze the patient's training, helping rehabilitation physicians and patients to adjust rehabilitation strategies in real time and improve rehabilitation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 A system diagram of an upper limb rehabilitation training system and method for stroke patients according to the present invention; Figure 2 The present invention discloses an electroencephalogram (EEG) equipment workflow of an upper limb rehabilitation training system and method for stroke patients. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0046] Traditional stroke upper limb rehabilitation training methods are mainly based on standardized movements and preset rehabilitation plans. However, there are large errors in actual clinical scenarios, which may affect the evaluation of rehabilitation effects, especially in scenarios where the patient's motor ability changes dramatically and the rehabilitation progress is uneven; in view of this, the present invention provides the following embodiments to solve the above technical problems.

[0047] like Figure 1 As shown, in specific embodiment 1: A stroke patient upper limb rehabilitation training system includes a patient terminal 100, a cloud platform server 200 and a rehabilitation physician terminal 300 which are connected in sequence.

[0048] The patient terminal 100 includes a central control device 101, a rehabilitation training device 110, a virtual reality platform 120 and a patient interaction platform 130; the rehabilitation training device 110 and the virtual reality platform 120 are electrically connected to the central control device 101, and the rehabilitation training device 110 and the virtual reality platform 120 receive action signals from the central control device 101 and execute them; the patient interaction platform 130 and the data analysis system 200 are connected through a network based on the mobile device of the patient terminal to exchange data.

[0049] The rehabilitation training device 110 includes a mechanical glove 111 , an adjustable supporting device 112 , an electroencephalographic device 114 , a neuromuscular electrical stimulation device 115 and a data acquisition device 116 .

[0050] The mechanical glove 111 includes a sensor group, an actuator and a motor, which can provide motion feedback of different intensities, and monitor the hand motion range, strength and frequency in real time through sensors; the adjustable support device 112 supports arm training at different angles and heights through an electric adjustment mechanism, and the adjustable support device 112 is installed with a sensor group; the sensor group collects the height, angle and strength of the patient's upper limb movements to form motion data.

[0051] The sensors embedded in the mechanical glove 111 include force sensors, displacement sensors, angle sensors and acceleration sensors. Four sensors form a group, and each sensor group is embedded in the fingers, palms and wrists to monitor the hand's motion range, strength and frequency in real time.

[0052] The motors and actuators are used to realize active feedback of hand movements. The motors and actuators simulate resistances of different intensities according to the environmental requirements of the virtual reality platform 120 to provide targeted sports training.

[0053] The motors and actuators simulate resistance according to the settings of the data analysis system 220, and the data analysis system 220 processes the sensor data in real time, controls the operation of the motors, adjusts the training intensity and mode, and adjusts the feedback intensity according to the sensor feedback.

[0054] The electric adjustment mechanism of the adjustable support device 112 adopts a servo motor to realize the height and angle adjustment of the adjustable support device 112. The servo motor can accurately adjust the posture of the adjustable support device 112 through the data analysis system 220; a sensor group is installed on the adjustable support device 112 to detect and record the angle, displacement and acceleration of the adjustable support device 112 to ensure the stability and consistency of the patient's movement during training; the adjustable support device 112 is electrically connected to the data analysis system 220 to realize automatic control and adjustment of the electric adjustment mechanism of the adjustable support device 112, and can also be remotely controlled by the rehabilitation physician terminal 310.

[0055] The control system can coordinate with the feedback system of the mechanical glove 111 to adapt to the dynamic needs of the patient.

[0056] like Figure 2 ,Workflow of EEG equipment,EEG equipment 114 adopts brain-computer interface BCI technology to perform signal encoding, signal acquisition, signal decoding, and signal output in sequence.

[0057] EEG device 114 is a special passive training tool. At the first International BCI Research Conference, researchers gave the following definition of brain-computer interface: "Brain-computer interface is a communication system that does not rely on the normal output pathways of the brain consisting of the peripheral nervous system and muscles". Brain-computer interface uses engineering technology to establish a new way of direct external information exchange and control between the human brain and computers or other electromechanical devices. Using brain-computer interface BCI technology, by bypassing the patient's peripheral muscle nerve tissue, a neural pathway from the brain to the limbs is simulated in vitro. The system captures the intention signals sent by the brain, and performs signal recognition and analysis processing, and finally converts the EEG signals into machine language to control the rehabilitation vehicle to drive the limbs to complete the corresponding rehabilitation training movements. This training method allows patients to control limb movements completely according to their own intentions, and through scientific cycle training, gradually rebuild the neural pathways in the body, thereby achieving effective control and control of the limbs.

[0058] Signal coding means that the patient will display the virtual training environment on the VR headset and perform the indicated movements according to the rehabilitation plan set by the rehabilitation physician. The VR headset will also provide real-time feedback and movement corrections, and voice reminders and corrections will be made based on the patient's movements. When the patient thinks about the indicated movements in his brain, the brain sends out the intention to control the upper limb movements according to the instructions of the VR headset, which will produce various physiological reactions, which are manifested in the form of EEG signals.

[0059] EEG signal is a kind of physiological electrical signal. In the signal acquisition process, an EEG signal acquisition system is used to collect and store EEG signals.

[0060] Signal decoding is the process of preprocessing, feature extraction and classification of EEG signals collected during the signal acquisition process to identify the brain's control intention as the recognition result.

[0061] The recognition result is the patient's actual control instruction, which is transmitted as EEG data to the external data acquisition device 116. The data analysis system 220 performs corresponding actions on the mechanical glove 111 and the adjustable support device 112 of the rehabilitation training equipment 110 to control the equipment, thereby enabling the patient to control the equipment through brain waves.

[0062] In this implementation, the signal acquisition of the EEG device adopts the EEG signal module in the brain-computer interface BCI technology. The EEG signal module is a non-invasive EEG signal acquisition technology. The EEG signal acquisition device can be purchased in combination with the patient's economic situation. It does not require surgery, has little physical harm, and is easy to install. The EEG signal module is in the form of an EEG cap. Electrodes are distributed on the surface of the EEG cap. The electrodes are made of metal. The metal electrodes are in contact with the human brain scalp. According to the principle of current conduction, the EEG signals are transmitted to the EEG signal acquisition device for amplification and conversion. The conductivity of the metal can collect tiny EEG signal changes on the scalp surface. The present invention adopts Neuracle's 32-channel EEG signal acquisition system with a sampling frequency of 250Hz and 32 electrodes, which can record spontaneous activities of the human brain without any time delay.

[0063] When the patient completes signal encoding by issuing action instructions to the brain according to the instructions of the VR headset, the EEG signal module captures, collects and stores the EEG signal and then decodes the signal; signal decoding includes signal preprocessing, feature extraction and classification; a large number of EEG signals of known categories are collected, their power spectral density PSD matrices are calculated respectively, and the corresponding PSD matrix set is constructed according to the category, and by establishing the PSD matrix, the conversion of the EEG signal from Euclidean space to Riemann space is completed; for the EEG signal collected by the signal acquisition EEG signal module, its PSD matrix is ​​calculated, and the Riemann distance between the PSD matrix and the PSD matrix in the PSD matrix set of all known categories is measured; finally, the K nearest neighbor classification algorithm is used to determine the category to which the EEG signal of unknown category belongs, so as to identify the intention of brain control as the recognition result, and output the corresponding action signal to the data analysis system 220.

[0064] Collecting a large number of EEG signals of known categories means collecting EEG signals generated by a large number of volunteers completing corresponding actions according to specified instructions. The algorithm realizes the conversion of EEG signals from Euclidean space to Riemannian space through the PSD matrix. Since the PSD matrix can reflect the characteristics of EEG signals in different frequency components and show the interaction between different channels, this method has significant advantages over traditional Euclidean space algorithms in feature extraction, thereby ensuring the accuracy of subsequent pattern classification.

[0065] The neuromuscular electrical stimulation device 115 includes an electrode sheet, a power module and a communication module. In this embodiment, the power module of the neuromuscular electrical stimulation device 115 is powered by a lithium battery, and the communication module uses Bluetooth communication.

[0066] The electrodes of the neuromuscular electrical stimulation device 115 are attached to the patient's upper limb muscles, and the communication module receives the control signal of the central control device 101, and the electrodes discharge accordingly, thereby electrically stimulating the upper limb muscles of the patient corresponding to the indicated action, thereby accelerating the reconstruction of the upper limb motor nerves; the electrodes and the communication module are electrically connected to the power module respectively, and the power module supplies power to the electrodes and the communication module; the device drives the palm of the stroke patient to perform various movements, thereby alleviating the patient's hand curling symptoms and performing rehabilitation treatment on the hand's motor function.

[0067] The data acquisition device 116 also includes a heart rate sensor and a blood oxygen sensor, which are used to collect the patient's real-time heart rate and blood oxygen conditions as physical condition data. When the data analysis system 220 determines that the physical condition data is abnormal, including abnormal data collection such as heart rate cessation, rapid changes, and rapid changes in blood oxygen, the data analysis system 220 immediately determines the abnormality and sends a stop signal to the central control device 101 to the rehabilitation training device 110 and the virtual reality platform 120. The central control device 101 sends a stop instruction to the rehabilitation training device 110 and the virtual reality platform 120, and the rehabilitation training device 110 and the virtual reality platform 120 stop working, and trigger the alarm device 140 of the patient terminal 100 and the alarm device 320 of the rehabilitation physician end 300, notifying the rehabilitation physician and the patient and their family members to respond in time. The rehabilitation physician can also directly control the rehabilitation training device 110 and the virtual reality platform 120 to stop working through the computer 310 of the rehabilitation physician end 300, and the patient and his family members can also directly stop the system from working through the patient terminal 100 mobile device.

[0068] The virtual reality platform 120 includes a VR headset worn on the patient's head; the VR headset displays a virtual training environment and real-time feedback motion correction; the virtual training environment provides simulated rehabilitation scenarios, such as grasping and moving virtual objects, to enhance the fun and sense of participation in training; the real-time feedback motion correction guides the patient to perform correct movements by monitoring the patient's movements and providing instant feedback.

[0069] The virtual training environment of the virtual reality platform 120 also includes developing a virtual training environment, using the Unity3D game engine to develop virtual training scenes; creating a variety of training tasks, including grabbing, moving and placing virtual objects, simulating actions and environments in the real world; considering the patient's rehabilitation needs when designing scenes, ensuring that the virtual training environment is challenging and can motivate patients. It is also equipped with a VR headset and connected to the mechanical glove 111. The VR headset is used to display virtual scenes, and the mechanical glove 111 is used for patients to control virtual objects to complete tasks.

[0070] The real-time feedback of the virtual reality platform 120 also includes motion tracking. Motion tracking integrates motion tracking technology in the VR system. The built-in tracker of the glove and the sensor embedded in the glove capture the patient's hand movement in real time. The motion data of the hand movement is calculated and processed in real time by the data analysis system 220, and feedback is generated in the virtual training environment to guide the patient to perform correct movements. Instant visual and auditory feedback is provided according to the patient's motion data, including guidance prompts, progress display and encouraging sounds to complete real-time feedback motion correction.

[0071] The patient interaction platform 130 uses a mobile terminal as a patient human-computer interaction interface, including social functions, adaptive feedback and data viewing functions.

[0072] The mobile application provides a patient interface for patients to view training plans, progress reports, and receive training guidance.

[0073] The social function allows patients to share experiences with other rehabilitators, encourage each other, and improve the continuity and motivation of training.

[0074] The adaptive feedback device automatically adjusts the training intensity and difficulty according to the real-time motion data. The patient can change the training intensity and difficulty by himself through the mobile application, and the rehabilitation physician can change the training intensity and difficulty through the physician terminal and select different training movements. The data viewing function views the data in the data storage system 210, and tracks and analyzes the patient's rehabilitation status in real time. At the same time, the rehabilitation physician conducts analysis and confirmation, generates a scientific training plan and evaluation report, and feeds back to the patient.

[0075] The patient interaction platform 130 is developed using the ReactNative framework and integrates an API interface to achieve connection with the cloud platform server 200; an intuitive patient interface and patient experience are designed to ensure that patients can easily operate and obtain information, and patient training and support are provided to help patients become familiar with how to use the application; social functions are integrated into the mobile application, such as community forums, real-time chat and experience sharing functions, allowing patients to interact with other rehabilitators, share experiences and encouragement, and improve the continuity and motivation of training.

[0076] And set up incentive mechanisms, design and implement incentive mechanisms, such as point systems, reward badges and challenges, to encourage patients to actively participate in rehabilitation training and communicate with others.

[0077] The cloud platform server 200 includes a data storage system 210 and a data analysis system 220 .

[0078] The data storage system 210 includes a patient history database 211 that records patient movement data, a rehabilitation program database 212 that records patient rehabilitation program data, and a healthy volunteer database 213 that records healthy volunteer movement data, the movement data including data on hand position, movement amplitude, and force output for completing indicated movements.

[0079] The data storage system 210 includes a data acquisition device 116, which collects EEG data, physical condition data and motion data. The motion data collected from the mechanical glove 111 and the support frame is collected through sensors, including hand position, movement amplitude, and force output information. The physical condition data is collected through the patient physical condition monitoring device 113, and the EEG data is collected through the EEG device 114. The data storage system 210 uses the collected data to establish a database to store the patient's training data as the patient history database 211 to ensure the security and integrity of the data. The patient history database 211 is transmitted to the cloud platform processor in real time.

[0080] The data storage system 210 is electrically connected to the data analysis system 220 .

[0081] The data analysis system 220 uses the deep learning model of artificial intelligence to analyze the training data, evaluate the patient's rehabilitation progress, and generate a training report and adjustment suggestions. The data analysis system 220 includes a preamplifier circuit 221, a motion information extraction unit 222, and a similarity measurement unit 223 that are electrically connected in sequence; the motion data in the patient history database 211 is amplified by the preamplifier circuit 221, and the amplified data is sent to the motion information extraction unit 222 to extract the feature vector of the motion data through the deep learning model, and the feature vector is calculated by the similarity measurement unit 223 to obtain the difference between the patient's skeleton motion feature vector and the healthy volunteer's feature vector; a threshold is set, and when the difference value is less than the set threshold, the action is judged to meet the requirements, and the next rehabilitation action can be performed. The rehabilitation plan is a series of rehabilitation action groups arranged by the rehabilitation physician at the beginning.

[0082] The motion information extraction unit 222 of the data analysis system 220 convolves the motion data to identify the feature vector of the patient's skeleton motion, specifically including: Improve the classification ability of skeleton-based models on ambiguous samples, since the actions are very similar and can be easily misclassified. The feature extraction capability using graph convolution plays a crucial role and determines the information transfer between joints.

[0083] Construct an undirected skeleton graph based on the human skeleton, bring the enlarged data into the skeleton graph and encode it into a matrix; N Represents the number of joint points, that is, N The skeleton of joint points is represented as an undirected skeleton graph ,in Represents all the vertices of this skeleton graph, that is, all the joint points in the skeleton; E The edges of the skeleton graph represent the physical connections between the joints, and the dependencies between the joints are encoded as a shape of The adjacency matrix of .

[0084] The feature vector is extracted through the ST-GCN spatiotemporal graph convolutional network model of the deep learning model; the ST-GCN spatiotemporal graph convolutional network model defines the implementation of graph convolution in skeleton data, so that the human skeleton action sequence is represented as ,in C IN Indicates the number of input channels, T Indicates the number of skeleton graph frames, N Indicates the number of joints.

[0085] The implementation process of graph convolution based on skeleton data can be expressed as follows: For the input skeleton data, the traditional graph convolution uses weights W Transform features and then pass a ij polymerization v i The features of adjacent vertices are used to update the i The output feature of each joint is ,in represents the normalized adjacency matrix; represents the normalized diagonal matrix; let to avoid There are blank lines in the represents the weight matrix, C OUT Indicates the number of output channels; L Initialize the matrix to a matrix with all elements set to 1; Indicates calculation of the Hadamard product.

[0086] The similarity measurement unit 223 uses the Euclidean distance to calculate the corresponding similarity score to obtain the difference between the patient skeleton motion data and the healthy volunteer data. Under the specified action, the patient skeleton motion feature vector is , the corresponding healthy volunteer feature vector is , calculate the Euclidean distance of the feature vector ,The similarity score is calculated by taking the weighted average of the Euclidean distance of all feature vectors to obtain the difference value between the feature vector of the patient’s skeleton motion and that of the healthy volunteer.

[0087] According to the data analysis results, the movement quality of the patient's rehabilitation plan is directly converted into numerical values, which is clearer and more understandable, helping rehabilitation physicians to specify more personalized rehabilitation plans in real time, and using artificial intelligence to learn the rehabilitation plans, rehabilitation progress, and training effects made by the rehabilitation physician each time. The big data and artificial intelligence technology of the cloud platform server 200 are used to track and analyze the patient's rehabilitation status in real time, and a training report can be presented in the form of charts and text, and provided to the rehabilitation physician. At the same time, it is analyzed and confirmed by the rehabilitation physician to generate a scientific training plan and evaluation report, which is fed back to the patient.

[0088] The cloud platform server 200 terminal is located in the hospital and synchronizes the patient's rehabilitation data to the rehabilitation physician of the corresponding hospital in real time. The rehabilitation physician views the real-time data of the patient's historical database 211 through the computer 310 of the rehabilitation physician terminal 300, formulates a rehabilitation plan based on the patient's rehabilitation data, transmits the rehabilitation plan data to the rehabilitation plan database 212 of the data analysis system 200, and controls the actions of the rehabilitation training equipment 110 and the virtual reality platform 120 of the patient terminal 100 through the data analysis system 220.

[0089] The rehabilitation physician can also directly control the rehabilitation training equipment 110 of the patient terminal 100 and instruct the movements of the virtual reality platform 120 through the cloud platform server 200; the rehabilitation physician connects with the patient through the cloud platform and communicates with the patient in real time for online exchanges.

[0090] A method for an upper limb rehabilitation training system for stroke patients applicable to this embodiment, the specific operation steps of the method are as follows: Step S1: After the patient completes the device configuration and wears it correctly, the patient starts the rehabilitation training device for rehabilitation training, and conducts scenario-based rehabilitation training in the virtual reality platform. The rehabilitation training device works accordingly to accelerate rehabilitation, and at the same time synchronizes the patient's training data to the data storage system of the cloud platform server.

[0091] First, an upper limb rehabilitation training system for stroke patients is inspected and set up to ensure that the rehabilitation training equipment, virtual reality platform and patient interaction platform of the patient terminal are connected normally, check the power supply and connection status of the central control device to ensure its normal operation, ensure that the cloud platform server runs stably and is well connected with the patient interaction platform.

[0092] Complete the equipment configuration and wearing of the patient at the patient terminal; the patient wears the rehabilitation training equipment, installs and adjusts the mechanical gloves and adjustable support device, wears the EEG cap of the EEG device on the head, and sticks the electrodes of the neuromuscular electrical stimulation device on the muscles to ensure that the rehabilitation training equipment is suitable for the patient's upper limbs; the patient configures the virtual reality platform, wears the VR headset, ensures that his vision is clear, and the headset is electrically connected to the central control device; check the patient interaction platform to ensure that the mobile application and social functions are operating normally, and the patient can view and manage the training data through the platform.

[0093] Step S11: The patient starts the VR headset, wears the VR headset, and enters the virtual training environment; the patient practices virtual training scenarios in the virtual reality platform, such as grasping and moving virtual objects; the VR headset provides real-time feedback action correction, and helps the patient correct actions through visual and auditory feedback.

[0094] Step S12: In the EEG device, the patient performs thinking activities according to the tasks in the virtual training environment. The EEG device encodes the control signals sent by the brain and performs signal encoding. The EEG device collects and decodes signals, extracts the brain's control intention for upper limb movements, and converts them into instructions, and transmits the signals to the central control device. The central control device adjusts the movements of the rehabilitation training equipment according to the instructions to complete the movement control.

[0095] Step S13: The rehabilitation training equipment is controlled by the command of the central control device to perform operations; the mechanical gloves monitor the hand movement range, strength and frequency through sensors, and adjust the training intensity according to the feedback; the adjustable support device electrically adjusts the angle, height and acceleration of the arm movement according to the training needs; neuromuscular electrical stimulation, the electrode sheet electrically stimulates the patient's upper limbs to promote motor nerve reconstruction and accelerate the rehabilitation process.

[0096] Step S14: While the patient is undergoing rehabilitation training, the data storage system records the patient's movement data during the training process, including hand position, range of motion and force output; the data analysis system uses artificial intelligence algorithms to analyze the training data, evaluate the rehabilitation progress, and generate personalized training reports and adjustment suggestions; if abnormal data occurs, when the data is monitored to be abnormal, the alarm device is triggered to notify the rehabilitation physician and control the equipment to stop working.

[0097] Step S2: The rehabilitation physician views the patient's rehabilitation data on the computer of the rehabilitation physician terminal through the cloud platform server, and the rehabilitation physician formulates a training plan. The central control device receives the rehabilitation physician's plan and forwards it to the rehabilitation training equipment and virtual reality platform of the patient terminal.

[0098] During the rehabilitation training process, the rehabilitation physician views the patient's rehabilitation data through the cloud platform server and develops a personalized training plan; the central control device receives the rehabilitation physician's plan and forwards it to the rehabilitation training equipment and virtual reality platform on the patient's terminal.

[0099] Step S3: The cloud platform server realizes data storage and interaction. Patients view rehabilitation data and conduct social interactions through mobile terminals. Rehabilitation physicians and patients communicate online through the cloud platform to discuss rehabilitation progress and next steps, and provide remote guidance and support.

[0100] The cloud platform server synchronizes the patient's real-time data to the hospital's rehabilitation physician; the patient views the rehabilitation data through the patient interaction platform, and interacts socially to obtain psychological support; physician communication and plan adjustment, the rehabilitation physician adjusts the training plan based on the real-time data and analysis results through the rehabilitation physician terminal, and transmits the new plan data to the central control device through the cloud platform. The patient terminal device adjusts the actions of the rehabilitation training equipment and virtual reality platform according to the new plan through the central control device; and the rehabilitation physician and the patient communicate online through the cloud platform to discuss rehabilitation progress and next steps, and provide remote guidance and support.

[0101] Step S4: At the end of the training, the training data is backed up to the cloud platform server for subsequent analysis and historical record review, and a training summary report is generated and reviewed.

[0102] At the end of the training, the central control device will stop the rehabilitation training equipment and virtual reality platform, check the status of all equipment, ensure that there is no damage, and perform necessary cleaning and maintenance; at the end of the training, data backup and summary will be performed, and the training data will be backed up to the cloud platform server for subsequent analysis and historical record review, and a training summary report will be generated and reviewed to provide a reference for the next training.

[0103] Through the implementation of the above steps, rehabilitation training equipment, virtual reality platform, central control device and patient interaction platform can be integrated into a comprehensive upper limb rehabilitation training system for stroke patients, and real-time interaction between patients and rehabilitation physicians can be achieved through the cloud platform server. The system will provide personalized and highly interactive rehabilitation training programs, effectively improving the rehabilitation effect and enthusiasm of patients.

[0104] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A stroke patient upper limb rehabilitation training system, characterized in that: It comprises a patient terminal (100), a cloud platform server (200) and a rehabilitation physician terminal (300) which are connected in sequence; The patient terminal (100) comprises a central control device (101), a rehabilitation training device (110), a virtual reality platform (120) and a patient interaction platform (130); the rehabilitation training device (110) and the virtual reality platform (120) are electrically connected to the central control device (101), and the rehabilitation training device (110) and the virtual reality platform (120) receive action signals sent by the central control device (101) and execute them; The virtual reality platform (120) includes a VR head display worn on the patient's head; The rehabilitation training device (110) comprises a mechanical glove (111), an adjustable support device (112), an electroencephalogram device (114), a neuromuscular electrical stimulation device (115), and a data acquisition device (116); The mechanical glove (111) comprises a sensor group, an actuator and a motor; the adjustable support device (112) supports arm training at different angles and heights through an electric adjustment mechanism, and the adjustable support device (112) is installed with a sensor group; the sensor group collects the height, angle and strength of the patient's upper limb movements to form motion data; The EEG device (114) uses brain-computer interface (BCI) technology to sequentially perform signal encoding, signal acquisition, signal decoding, and signal output; The signal encoding is that when the patient performs internal thinking activities in the brain according to the instructions displayed by the VR headset, the brain sends out the intention to control the upper limb movement according to the instructions of the VR headset, which will produce various physiological reactions, and the reactions are reflected in the form of EEG signals; The EEG signal is a physiological electrical signal, and an EEG signal acquisition system is used to collect and store the EEG signal during the signal acquisition process; The signal decoding is to perform signal preprocessing, feature extraction and classification on the EEG signals collected during the signal collection process, so as to identify the intention of brain control as the recognition result; The recognition result is the actual control instruction of the patient, which is used as EEG data and transmitted to an external data acquisition device (116). The data acquisition device (116) is electrically connected to the central control device (101), and sends a corresponding instruction action execution signal to the mechanical glove (111) and the adjustable support device (112) of the rehabilitation training device (110) through the central control device (101), so that the mechanical glove (111) and the adjustable support device (112) perform the corresponding action. The neuromuscular electrical stimulation device (115) comprises an electrode sheet, a power module and a communication module. The electrode sheet is attached to the upper limb muscles of the patient. The communication module receives the control signal of the central control device (101), and the electrode sheet discharges accordingly. By electrically stimulating the upper limb muscles of the patient corresponding to the instruction action, the speed of upper limb motor nerve reconstruction is accelerated. The electrode sheet and the communication module are respectively electrically connected to the power module, and the power module supplies power to the electrode sheet and the communication module. The cloud platform server (200) comprises a data storage system (210) and a data analysis system (220); The data storage system (210) includes a patient history database (211) recording patient movement data, a rehabilitation program database (212) recording patient rehabilitation program data, and a healthy volunteer database (213) recording healthy volunteer movement data, wherein the movement data includes data on hand position, movement amplitude, and force output for completing an indicated movement; The data analysis system (220) comprises a preamplifier circuit (221), a motion information extraction unit (222) and a similarity measurement unit (223) which are electrically connected in sequence; the motion data in the patient history database (211) are amplified by the preamplifier circuit (221), the amplified data are sent to the motion information extraction unit (222), a feature vector is extracted from the motion data by a deep learning model, and the feature vector is calculated by the similarity measurement unit (223) to obtain a difference value between the patient's skeleton motion feature vector and the healthy volunteer's feature vector; a threshold is set, and when the difference value is less than the set threshold, it is judged that the motion meets the requirement; The rehabilitation physician checks the patient history database (211) through the rehabilitation physician terminal (300), formulates a rehabilitation plan based on the patient rehabilitation data, transmits the rehabilitation plan data to the rehabilitation plan database (212), and controls the patient terminal (100) to execute the indicated actions of the rehabilitation plan.

2. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The signal acquisition of the electroencephalogram (114) device adopts the EEG signal module in the brain-computer interface (BCI) technology. The EEG signal module is a non-invasive electroencephalogram (EEG) signal acquisition technology. The EEG signal module is in the form of an electroencephalogram (EEG) cap. Electrodes are distributed on the surface of the EEG cap. The electrodes are made of metal. The metal electrodes are in contact with the human brain scalp. According to the principle of current conduction, the electroencephalogram (EEG) signal is transmitted to the electroencephalogram (EEG) signal acquisition device for amplification and conversion. The electrical conductivity of the metal can acquire tiny changes in the electroencephalogram (EEG) signal on the scalp surface. When the patient completes signal encoding by sending an action command to the brain according to the instructions of the VR headset, the EEG signal module captures, collects and stores the EEG signal and then decodes the signal; the signal decoding includes signal preprocessing, feature extraction and classification; a large number of EEG signals of known categories are collected, their power spectrum density PSD matrices are calculated respectively, and a corresponding PSD matrix set is constructed according to the category. By establishing the PSD matrix, the conversion of the EEG signal from the Euclidean space to the Riemann space is completed; for the EEG signal collected by the signal acquisition EEG signal module, its PSD matrix is ​​calculated, and the Riemann distance between the PSD matrix and the PSD matrix in the PSD matrix set of all known categories is measured; finally, the K nearest neighbor classification algorithm is used to determine the category to which the EEG signal of the unknown category belongs, so as to identify the intention of brain control as the recognition result, and output the corresponding action signal to the data analysis system (220); the collection of a large number of EEG signals of known categories is to collect a large number of EEG signals generated by volunteers completing corresponding actions according to specified instructions.

3. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The sensors embedded in the mechanical glove (111) include a force sensor, a displacement sensor, an angle sensor and an acceleration sensor, with four sensors forming a group. Each group of sensors is embedded in a finger, a palm and a wrist to monitor the range of motion, strength and frequency of the hand in real time. The motor and actuator are used to realize active feedback of hand movement, and the motor and actuator simulate resistance of different intensities according to the environmental requirements of the virtual reality platform (120) to provide targeted sports training; The motor and the actuator simulate resistance according to the settings of the data analysis system (220), and the data analysis system (220) processes sensor data in real time, controls the operation of the motor, adjusts the training intensity and mode, and adjusts the feedback intensity according to sensor feedback.

4. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The electric adjustment mechanism of the adjustable support device (112) uses a servo motor to achieve height and angle adjustment of the adjustable support device (112), and the servo motor accurately adjusts the posture of the adjustable support device (112) through the data analysis system (220); a sensor group is installed on the adjustable support device (112) to detect and record the angle, displacement and acceleration of the adjustable support device (112) to ensure stability and consistency during training; the adjustable support device (112) is electrically connected to the data analysis system (220) to achieve automatic control and adjustment of the electric adjustment mechanism of the adjustable support device (112), or remote control through a rehabilitation physician terminal (310).

5. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The motion information extraction unit (222) convolves the motion data to identify the feature vector of the patient's skeleton movement, specifically including: Construct an undirected skeleton graph based on the human skeleton, bring the enlarged data into the skeleton graph and encode it into a matrix; N Represents the number of joint points, that is, N The skeleton of joint points is represented as an undirected skeleton graph ,in Represents all the vertices of this skeleton graph, that is, all the joint points in the skeleton; E The edges of the skeleton graph represent the physical connections between the joints, and the dependencies between the joints are encoded as a shape of The adjacency matrix of ; The feature vector is extracted through the ST-GCN spatiotemporal graph convolutional network model of the deep learning model; the ST-GCN spatiotemporal graph convolutional network model defines the implementation of graph convolution in skeleton data, so that the human skeleton action sequence is represented as ,in C IN Indicates the number of input channels, T Indicates the number of skeleton graph frames, N Indicates the number of joints, The implementation process of graph convolution based on skeleton data can be expressed as follows: For the input skeleton data, the traditional graph convolution uses weights W Transform features and then pass a ij polymerization v i The features of adjacent vertices are used to update the i The output feature of each joint is , in represents the normalized adjacency matrix; represents a normalized diagonal matrix; represents the weight matrix, C OUT Indicates the number of output channels; L Initialize the matrix to a matrix with all elements set to 1; Indicates calculation of the Hadamard product.

6. An upper limb rehabilitation training system for stroke patients according to any one of claims 1 or 5, characterized in that: The similarity measurement unit (223) uses Euclidean distance to calculate the corresponding similarity score to obtain the difference between the patient skeleton motion data and the healthy volunteer data. Under the specified action, the patient skeleton motion feature vector is: , the corresponding healthy volunteer feature vector is , calculate the Euclidean distance of the feature vector ,The similarity score is calculated by taking the weighted average of the Euclidean distance of all feature vectors to obtain the difference value between the feature vector of the patient’s skeleton motion and that of the healthy volunteer.

7. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The virtual training environment of the virtual reality platform (120) is developed by using the Unity3D game engine to develop virtual training scenes; creating a variety of training tasks, including grabbing, moving and placing virtual objects, simulating actions and environments in the real world; and being equipped with a VR head display, which is connected to a mechanical glove (111); the VR head display is used to display the virtual scene, and the mechanical glove (111) is used for the patient to control the virtual object through action to complete the task; The real-time feedback of the virtual reality platform (120) further includes motion tracking, wherein the motion tracking technology is integrated into the VR system, and the tracker built into the glove and the sensor embedded in the glove capture the patient's hand movement in real time; the motion data of the hand movement is calculated and processed in real time by the cloud platform server (200), and feedback is generated in the virtual training environment to guide the patient to perform correct movements, and instant visual and auditory feedback is provided according to the patient's motion data, including guidance prompts, progress display and encouraging voice to complete the real-time feedback motion correction.

8. The upper limb rehabilitation training system for stroke patients according to claim 1, characterized in that: The patient interaction platform (130) uses a mobile application of a mobile terminal as a patient human-computer interaction interface, including social functions, adaptive feedback and data viewing functions; The mobile application provides a patient interface for patients to view training plans, progress reports, and receive training guidance; The social function allows patients to communicate with other recovering people; The adaptive feedback device automatically adjusts the training intensity and difficulty according to the real-time motion data. The patient can change the rehabilitation plan by himself through the mobile application, and the rehabilitation physician can change the rehabilitation plan through the physician terminal and select different training movements. The data viewing function views the data in the data storage system (210); The patient interaction platform (130) is developed using the ReactNative framework and integrated with an API interface to achieve connection with the cloud platform server (200); an intuitive patient interface and patient experience are designed.

9. A method for upper limb rehabilitation training for stroke patients, characterized in that: The method uses an upper limb rehabilitation training system for stroke patients according to any one of claims 1 to 8. The specific operation steps of the upper limb rehabilitation training method for stroke are as follows: Step S1: The patient completes the device configuration and wears it correctly. The patient starts the rehabilitation training device for rehabilitation training. The patient conducts scenario-based rehabilitation training in the virtual reality platform. The rehabilitation training device works accordingly to accelerate rehabilitation. At the same time, the patient's training data is synchronized to the data storage system of the cloud platform server. Step S2: The rehabilitation physician views the patient's rehabilitation data on the computer of the rehabilitation physician terminal through the cloud platform server, and the rehabilitation physician formulates a training plan. The central control device receives the rehabilitation physician's plan and forwards it to the rehabilitation training equipment and virtual reality platform of the patient terminal; Step S3: The cloud platform server realizes data storage and interaction. The patient views the rehabilitation data through the mobile terminal and conducts social interaction. The rehabilitation physician and the patient communicate online through the cloud platform to discuss the rehabilitation progress and next steps, and provide remote guidance and support. Step S4: At the end of the training, the training data is backed up to the cloud platform server for subsequent analysis and historical record review, and a training summary report is generated and reviewed.

10. The upper limb rehabilitation training method for stroke patients according to claim 9, characterized in that: The step S1 specifically includes: Step S11: The patient starts the VR headset, wears the VR headset, and enters the virtual training environment; the patient practices the virtual training scene in the virtual reality platform; the VR headset provides real-time feedback action correction, and helps the patient correct the action through visual and auditory feedback; Step S12: In the EEG device, the patient performs thinking activities according to the tasks in the virtual training environment, and the EEG device encodes the control signals sent by the brain and performs signal encoding; the EEG device collects and decodes the signals, extracts the brain's control intention for upper limb movement, and converts it into instructions, and transmits the signals to the central control device, which adjusts the movements of the rehabilitation training device according to the instructions to complete the movement control; Step S13: The rehabilitation training equipment is controlled by the command of the central control device to perform actions; the mechanical gloves monitor the range, strength and frequency of hand movements through sensors, and adjust the training intensity according to feedback; the adjustable support device electrically adjusts the angle, displacement and acceleration of the arm movement according to training needs; the neuromuscular electrical stimulation, the electrode sheet electrically stimulates the patient's upper limbs to promote motor nerve reconstruction and accelerate the rehabilitation process; Step S14: While the patient is undergoing rehabilitation training, the data storage system records the patient's movement data during the training process, including hand position, range of motion and force output; the data analysis system uses artificial intelligence algorithms to analyze the training data, evaluate the rehabilitation progress, and generate personalized training reports and adjustment suggestions; if abnormal data occurs, when the data is monitored to be abnormal, the alarm device is triggered to notify the rehabilitation physician and control the equipment to stop working.

Images