Exercise rehabilitation system and method based on layered stimulation navigation and closed-loop brain-computer interface

Abstract

The invention belongs to the cross technical field of rehabilitation engineering and neural engineering, and relates to an exercise rehabilitation system and method based on layered stimulation navigation and a closed-loop brain-computer interface, and the exercise rehabilitation system comprises a layered stimulation navigation module, a brain-computer interaction module, a multi-channel functional electrical stimulation module and an exercise feedback module. The layered stimulation navigation module outputs different stimulation schemes for the superficial layer muscle stimulation area and the deep layer muscle stimulation area; the brain-computer interaction module is used for outputting a decoding result capable of representing the motion intention of the convalescent; the multi-channel functional electrical stimulation module calls a stimulation scheme according to a decoding result, loads an injection current to a working electrode determined by the stimulation scheme, implements electrical stimulation and generates a motion response; the motion feedback module collects and fuses the multi-mode feedback signals under the motion response, generates a stimulation parameter adjustment signal used for adjusting a stimulation scheme executed in the previous rehabilitation period, and outputs a neural feedback signal used for guiding the rehabilitation person to adjust the motion imagination.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of rehabilitation engineering and neural engineering, and in particular to a motor rehabilitation system and method based on layered stimulation navigation and closed-loop brain-computer interface. Background Art

[0002] Stroke is a syndrome of acute neurological deficits caused by cerebrovascular disease, including cerebral infarction and cerebral hemorrhage. Characterized by high morbidity, recurrence, disability, mortality, and economic burden, it has become the leading cause of death and disability among adults in my country. Epidemiological survey data released by the Chinese Stroke Society indicate that there are currently 28 million stroke patients in my country, with 2 million new cases each year. Over 80% of stroke survivors experience upper limb motor dysfunction, severely impacting their quality of life and ability to perform daily activities. Therefore, the reconstruction and rehabilitation of fine motor function in the hands of stroke patients has become an important research area in the field of neurorehabilitation.

[0003] Currently, rehabilitation techniques for hand function recovery in stroke patients primarily include traditional rehabilitation methods based on occupational therapy, acupuncture, or medication, robotic-assisted rehabilitation using pneumatic gloves or exoskeletons, and traditional electrical stimulation rehabilitation using electronic acupuncture devices. However, these methods still have the following shortcomings in restoring fine motor function in the hand: (1) Traditional rehabilitation methods mainly rely on rehabilitation therapists for auxiliary treatment, resulting in low patient participation and slow rehabilitation process. Although they can relieve spasticity to a certain extent, they are difficult to effectively activate deep neural pathways and have limited effects on the recovery of fine movements. (2) Robot-assisted rehabilitation technology mainly focuses on training gross movements such as overall hand grasping, making it difficult to achieve precise training of fingers or knuckles, and the movement accuracy is limited, especially for patients with severe motor dysfunction. (3) Traditional electrical stimulation rehabilitation technology mainly targets large muscle groups, which can reduce hand muscle tension and improve joint mobility. However, it is limited by the stimulation depth and is difficult to accurately activate deep muscles, resulting in insufficient recovery of fine motor skills. (4) The above rehabilitation technologies are mostly open-loop controlled with fixed stimulation patterns. Patients cannot control them independently. They lack real-time adjustment mechanisms and personalized rehabilitation plans, and cannot establish neural circuits, which can easily lead to problems such as muscle fatigue and inaccurate movement patterns.

[0004] The achievement of fine motor skills in the hand relies on the coordinated action of the superficial and deep forearm muscles and the forearm nerves. The superficial forearm muscles are primarily responsible for wrist flexion and extension and basic finger movements, providing mechanical support and stability for fine manipulation. The deep muscles dominate the independent movement of the fingers, fine-tuning force, and rotational control, and are the core execution units of fine motor skills. The forearm nervous system is responsible for the transmission and regulation of motor commands. Traditional rehabilitation methods are unable to effectively coordinate the remodeling of these muscle groups and neural networks, resulting in limited recovery of patients' fine motor function.

[0005] Currently, the BCI-FES closed-loop rehabilitation system, which combines the brain-computer interface (BCI) with functional electrical stimulation (FES), has become an important research direction in the field of neurorehabilitation. This technology integrates BCI and FES to collect real-time electrical signals from the patient's brain activity, decode movement intentions, and automatically trigger precise electrical stimulation, forming a closed-loop feedback loop to enhance neural plasticity and accelerate the recovery of fine motor function in the hands. However, existing FES devices generally suffer from the following problems: (1) The number of stimulation channels is small, making it difficult to cover multiple key muscle groups in the hand; (2) Existing electrodes mainly rely on anatomical structure or experience for placement, lacking personalized adaptation; they lack dynamic switching during exercise, affecting the precision of movement; (3) The stimulation frequency is limited, and it mainly acts on the surface of the muscle, making it difficult to activate the deep muscle groups, affecting the neural regulation effect; (4) The lack of an efficient real-time feedback mechanism in the closed-loop system affects the accuracy of stimulation and individualized regulation.

[0006] Therefore, there is an urgent need to propose a sports rehabilitation system and method that can cover multiple key muscle groups of the hand, has a high degree of personalized adaptation, can activate deep muscle groups, and has an efficient real-time feedback mechanism. Summary of the Invention

[0007] The purpose of the present invention is to address the shortcomings of existing FES devices, such as the small number of stimulation channels, lack of personalized adaptation, difficulty in activating deep muscle groups, and lack of an efficient real-time feedback mechanism. A sports rehabilitation system and method based on layered stimulation navigation and closed-loop brain-computer interface is proposed to construct a bidirectional constant current drive structure with "PWM+Wilson current mirror+H-bridge" as the core. Multiple constant current sources, Wilson current mirrors and H-bridge modules can be expanded as needed to form a flexible and scalable multi-channel electrical stimulation framework that supports high-precision, high-synchronization and multi-muscle neuromuscular electrical stimulation applications. In addition, a layered precision stimulation mechanism of shallow low-frequency stimulation and deep coherent electrical stimulation is established. With the synergistic effect of timing control, frequency control and current amplitude control, the activation rate of deep muscles can be improved and precise stimulation control can be achieved.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a motor rehabilitation system based on layered stimulation navigation and a closed-loop brain-computer interface, comprising: a layered stimulation navigation module that receives image data and outputs a stimulation plan through modeling, tissue segmentation and annotation, stimulation area setting, and optimization algorithm solution; the stimulation area includes a superficial muscle stimulation area and a deep muscle stimulation area; the stimulation plan includes working electrodes to be activated in corresponding stimulation areas and an injection current corresponding to each working electrode; The brain-computer interaction module guides the patient to generate motor imagery, collects, pre-processes, and decodes the EEG signals generated by motor imagery, and outputs decoding results that represent motor intentions; The multi-channel functional electrical stimulation module receives and calls the stimulation scheme according to the decoding results, injects current into the working electrodes determined by the stimulation scheme, implements electrical stimulation and generates a motor response; The motion feedback module collects multimodal feedback signals under motion response and fuses the multimodal feedback signals to generate stimulation parameter adjustment signals and neural feedback signals; among them, the stimulation parameter adjustment signal is sent to the multi-channel functional electrical stimulation module to adjust the stimulation plan executed in the previous rehabilitation cycle; the neural feedback signal is sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

[0009] As a possible implementation method, the hierarchical stimulation navigation module includes a tissue structure image acquisition unit, a tissue segmentation and annotation unit, a modeling unit, a stimulation area setting unit, and a solution unit; Wherein, the tissue structure image acquisition unit obtains image data of the recovered person; The tissue segmentation and annotation unit combines anatomical structures and uses deep learning methods to segment and annotate the skin, fat, and muscle in the image data to obtain a two-dimensional annotation set; The modeling unit uses a three-dimensional voxel interpolation algorithm to perform voxel interpolation to establish a three-dimensional model; The stimulation area setting unit, based on the established three-dimensional model, assigns electrical characteristic values ​​corresponding to the frequency to the superficial muscle stimulation area and the deep muscle stimulation area. The electrical characteristic values ​​are equal to the electrical characteristic values ​​of human tissue at the corresponding frequency. The finite element method is used to solve the unit electric field distribution when a single pair of electrodes is activated. According to rehabilitation needs or anatomical positioning, the superficial muscle stimulation area and the deep muscle stimulation area are divided in the three-dimensional model based on the unit electric field distribution, and the desired effective electric field threshold is set accordingly. The solving unit uses an optimization algorithm and takes the effective electric field threshold as input to solve the initial stimulation plan, and continuously optimizes the electrode layout and injection current through simulation iteration to finally obtain the optimal stimulation plan.

[0010] As a possible implementation method, in the stimulation scheme, low-frequency electrical stimulation is used for the superficial muscle stimulation area, and coherent electrical stimulation is used for the deep muscle stimulation area.

[0011] As a possible implementation method, two pairs of working electrodes located at different positions around the target area constitute the stimulation conditions of coherent electrical stimulation; the stimulation mechanism of coherent electrical stimulation is: high-frequency bidirectional rectangular wave current is applied to the two pairs of working electrodes, and the high-frequency bidirectional rectangular wave current is driven by a set of high-frequency signals with a preset frequency difference; the high-frequency bidirectional rectangular wave current undergoes coherent interference in the deep muscle stimulation area, forming a low-frequency envelope electric field with a preset frequency difference.

[0012] As a possible implementation, a pair of electrodes corresponds to one stimulation channel; the stimulation circuit corresponding to one stimulation channel includes a symmetrically arranged programmable constant current source, a Wilson current mirror, and an H-bridge topology; The main control chip determines the analog voltage signal output by the peripheral port based on the injection current corresponding to each working electrode provided by the stimulation scheme; the analog voltage signal is input to the input port of the programmable constant current source, and after voltage division, a reference voltage signal is obtained to drive the operational amplifier and transistor. Driven by the reference voltage signal, the operational amplifier and transistor form a closed-loop feedback control; based on the virtual short and virtual open characteristics of the operational amplifier, the reference voltage signal is converted into the stimulation current output by the programmable constant current source through a resistor connected in series with the transistor emitter; The input port of the Wilson current mirror is connected to the output port of the programmable constant current source. After replicating and isolating the stimulation current output by the programmable constant current source, it is input into the H-bridge topology structure to achieve control of the stimulation current loaded on the working electrode.

[0013] As a possible implementation method, the Wilson current mirror is integrated between the upper and lower arms of the H-bridge topology structure, and the pulse width modulation signal output by the main control chip is used to realize the time-sharing conduction control of the upper and lower arms, forming a bipolar constant current output that alternates between forward and reverse directions, thereby forming a high-frequency bidirectional rectangular wave current loaded on the working electrode.

[0014] As a possible implementation method, the brain-computer interaction module includes a virtual animation presentation unit, an EEG device, a preprocessing unit, and a motion intention decoding unit; The virtual animation presentation unit is used to present virtual animation to guide the rehabilitator to observe the virtual animation and generate movement imagination; the virtual animation is generated based on the movement imagination task paradigm guided by movement observation; EEG equipment, at least for collecting EEG signals generated during motor imagery; A preprocessing unit, used for preprocessing EEG signals; The movement intention decoding unit uses the common space pattern algorithm to extract the EEG features of the EEG signal, and uses the support vector machine to classify the EEG features to realize the decoding of the movement intention and obtain the decoding result.

[0015] As a possible implementation method, the motion rehabilitation system is applied to the rehabilitation of fine motor skills of the upper limbs; the motion feedback module includes: an inertial sensor, a bending sensor, a flexible film sensor, a multimodal data extraction unit, a multimodal data fusion unit, and a signal decomposition unit; Among them, there are multiple inertial sensors, which are installed on the back of the hand and the fingertips of the five fingers to obtain acceleration and angular velocity. A flexible bending sensor is configured along the extension direction of each finger to obtain the finger bending angle. The posture angle data is determined based on the acceleration, angular velocity and bending angle. The flexible film sensor is attached to the inside of the finger to collect surface pressure data between the fingers or between the fingers and the rehabilitation device. The multimodal data extraction unit receives the posture angle data and the surface pressure data, performs normalization processing, extracts the joint bending angle value from the posture angle data to characterize the posture characteristics of the hand movement, and extracts the pressure value from the surface pressure data to characterize the finger pressure characteristics; The multimodal data fusion unit configures weights for the joint bending angle and pressure values, and obtains the initial fusion data through weighted summation. The multimodal data fusion unit is also configured with a linear state transfer model, which uses the linear state transfer model to take the initial fusion data as input, recursively obtains the optimal estimate of the fusion result, and obtains the multimodal feedback signal based on the optimal estimate. The signal decomposition unit generates a stimulation parameter adjustment signal and a neural feedback signal based on the multimodal feedback signal.

[0016] In a second aspect, the present invention provides a motor rehabilitation method based on layered stimulation navigation and closed-loop brain-computer interface, comprising the following steps: Develop a stimulation plan for each stimulation area, including superficial muscle stimulation areas and deep muscle stimulation areas. The stimulation plan includes the working electrodes that need to be activated in the corresponding stimulation area and the injection current corresponding to each working electrode; Guide the rehabilitator to generate motor imagery, collect, pre-process, and decode the EEG signals generated by motor imagery to output decoding results representing motor intention; Receive and call the stimulation plan according to the decoding result, load the working electrode determined by the stimulation plan with injection current, implement electrical stimulation and generate movement response; Multimodal feedback signals under movement response are collected and fused to generate stimulation parameter adjustment signals and neural feedback signals; the stimulation parameter adjustment signals are sent to the multi-channel functional electrical stimulation module to adjust the stimulation scheme executed in the previous rehabilitation cycle; the neural feedback signals are sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The proposed motor rehabilitation system, based on layered stimulation navigation and a closed-loop brain-computer interface, utilizes a bidirectional constant-current drive architecture centered around a "PWM + Wilson current mirror + H-bridge" system. Multiple constant-current sources, Wilson current mirrors, and H-bridge modules can be expanded as needed, forming a flexible and scalable multi-channel electrical stimulation framework that supports high-precision, high-synchronization, and multi-muscle neuromuscular electrical stimulation applications.

[0018] 2. The sports rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface proposed in the present invention uses deep learning methods to segment and label the skin, fat, and muscles in image data, and then divides the superficial muscle stimulation area and the deep muscle stimulation area, and establishes a layered precise stimulation mechanism of superficial low-frequency stimulation and deep coherent electrical stimulation. Combined with the synergistic effect of timing control, frequency control, and current amplitude control, it can improve the activation rate of deep muscles and achieve precise stimulation control.

[0019] 3. The proposed motor rehabilitation system, based on layered stimulation navigation and a closed-loop brain-computer interface, establishes an inner feedback loop of "real-time perception-feedback-closed-loop execution." When the fusion results indicate that the actual movement does not meet the expected movement according to EEG decoding, the electrical stimulation parameters are adjusted dynamically and promptly. Based on the data fusion results, the system automatically generates a reward and punishment feedback mechanism and displays it to the patient. It also establishes an outer feedback loop of "long-term learning-optimization closed loop," continuously improving the optimal rehabilitation strategy through progressive reinforcement learning.

[0020] 4. The sports rehabilitation system and method based on layered stimulation navigation and closed-loop brain-computer interface proposed in the present invention automatically switches the electrodes and current parameters during exercise according to the joint bending angle value and the timing of the movement, breaking the traditional electrode fixed stimulation mode and making the training movements more precise and smooth. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings: Figure 1 A schematic diagram of the structure of a motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface provided by an embodiment of the present invention; Figure 2 Schematic diagram of the coherent electrical stimulation principle in an embodiment of the present invention; Figure 3 A circuit diagram including a programmable constant current source, a Wilson current mirror, and an H-bridge topology structure in an embodiment of the present invention; Figure 4 Schematic diagram of the rehabilitation training experimental process for patients in an embodiment of the present invention.

[0022] Reference numerals 1-Layered stimulation navigation module, 10-Tissue structure image acquisition unit, 11-Tissue segmentation and annotation unit, 12-Modeling unit, 13-Stimulation area setting unit, 14-Solution unit; 2-Brain-computer interaction module, 20-Virtual animation presentation unit, 21-EEG equipment, 22-Preprocessing unit, 23-Motion intention decoding unit; 3-Multi-channel functional electrical stimulation module, 4-Motion feedback module. DETAILED DESCRIPTION

[0023] To facilitate a clear description of the technical solutions of the embodiments of the present invention, the words "first" and "second" are used in the embodiments of the present invention to distinguish between identical or similar items with substantially the same functions and effects. For example, the first threshold and the second threshold are merely used to distinguish between different thresholds and do not limit their order. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0024] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] In the present invention, "at least one" means one or more, and "more" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The following at least one item (item) or similar expressions refers to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one item (item) of a, b or c can mean: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or plural.

[0026] The embodiments of the present invention aim to provide a motor rehabilitation system and method based on layered stimulation navigation and closed-loop brain-computer interface, and to construct a bidirectional constant current drive structure with "PWM+Wilson current mirror+H-bridge" as the core. Multiple constant current sources, Wilson current mirrors and H-bridge modules can be expanded as needed to form a flexible and scalable multi-channel electrical stimulation framework that supports high-precision, high-synchronization and multi-muscle group neuromuscular electrical stimulation applications. In addition, the present application also establishes a layered precision stimulation mechanism of shallow low-frequency stimulation and deep coherent electrical stimulation, which, with the synergistic effect of timing control, frequency control and current amplitude control, can improve the activation rate of deep muscles and achieve precise stimulation control.

[0027] In the first aspect, the embodiment of the present invention provides a motor rehabilitation system based on hierarchical stimulation navigation and closed-loop brain-computer interface, see Figure 1 , including: layered stimulation navigation module 1, brain-computer interaction module 2, multi-channel functional electrical stimulation module 3 and movement feedback module 4; The layered stimulation navigation module 1 receives image data, performs modeling, tissue segmentation and annotation, stimulation area setting, and solves the optimization algorithm to output a stimulation plan; the stimulation area includes a superficial muscle stimulation area and a deep muscle stimulation area; the stimulation plan includes the working electrodes to be activated in the corresponding stimulation area and the injection current corresponding to each working electrode; As a possible implementation, the layered stimulation navigation module 1 includes a tissue structure image acquisition unit 10, a tissue segmentation and annotation unit 11, a modeling unit 12, a stimulation region setting unit 13, and a solution unit 14; The tissue structure image acquisition unit 10 acquires image data of the recovered person, and the tissue structure image includes an MRI image or an ultrasound image; The tissue segmentation and annotation unit 11 combines the anatomical structure and uses a deep learning method to segment and annotate the skin, fat, and muscle in the image data to obtain a two-dimensional annotation set; The modeling unit 12 performs voxel interpolation using a three-dimensional voxel interpolation algorithm to establish a three-dimensional model; Based on the established three-dimensional model, the stimulation area setting unit 13 assigns electrical characteristic values ​​corresponding to the frequencies to the superficial muscle stimulation areas and the deep muscle stimulation areas. The electrical characteristic values ​​are equal to the electrical characteristic values ​​of human tissue at the corresponding frequencies. The finite element method is used to solve the unit electric field distribution when a single pair of electrodes is activated. According to rehabilitation needs or anatomical positioning, the superficial muscle stimulation areas and the deep muscle stimulation areas are divided in the three-dimensional model based on the unit electric field distribution, and the desired effective electric field threshold is set accordingly. For example, before using this rehabilitation system, a magnetic resonance imaging (MRI) or ultrasound image of the patient's arm is first acquired. Deep learning methods are used to automatically segment and annotate the skin, fat, and muscle tissue. A 3D voxel interpolation algorithm is then used to construct a personalized 3D electrical stimulation navigation model for the forearm. Based on neuroanatomy, muscle groups related to finger extension are annotated, with a focus on the extensor pollicis longus, extensor pollicis brevis, and abductor pollicis longus muscles, which are associated with thumb extension.

[0028] The solving unit 14 uses an optimization algorithm, takes the effective electric field threshold as input, solves the initial stimulation plan, and continuously optimizes the electrode layout and injection current through simulation iteration, and finally obtains the optimal stimulation plan.

[0029] As a possible implementation method, in the stimulation scheme, low-frequency electrical stimulation is used for the superficial muscle stimulation area, and coherent electrical stimulation is used for the deep muscle stimulation area.

[0030] As a possible implementation method, two pairs of working electrodes located at different positions around the target area constitute the stimulation conditions of coherent electrical stimulation; the stimulation mechanism of coherent electrical stimulation is: high-frequency bidirectional rectangular wave current is applied to the two pairs of working electrodes, and the high-frequency bidirectional rectangular wave current is driven by a set of high-frequency signals with a preset frequency difference; the high-frequency bidirectional rectangular wave current undergoes coherent interference in the deep muscle stimulation area, forming a low-frequency envelope electric field with a preset frequency difference.

[0031] For example, conventional low-frequency electrical stimulation requires only one channel, or two pairs of electrodes. However, generating coherent electrical stimulation requires two channels, or two pairs of electrodes. For example, if a 2000Hz current is applied to one channel and a 2050Hz current is applied to another, the two pairs of electrodes will interfere coherently in the target area, generating a 50Hz coherent electrical current.

[0032] Next, we will use deep muscle stimulation area coherent electrical stimulation as an example to illustrate. Figure 2 By applying two high-frequency currents to couple, AM modulation is formed, and a low-frequency envelope electric field is coupled in the deep muscle stimulation area. The calculation formula is shown in formula (1): (1) Where, and It is an array of coefficients related to the input current of the electrode pair, which is used to control the current applied to each pair of electrodes; and For the control array of different high-frequency currents injected into the relevant electrode pairs, the Laplace equation is used to make the control equation of finite element analysis, as shown in formula (2): (2) Where, represents the conductivity, Represents electric potential.

[0033] This application establishes a layered precision stimulation mechanism of shallow low-frequency stimulation and deep coherent electrical stimulation, which cooperates with the synergistic effect of timing control, frequency control and current amplitude control to help improve the activation rate of deep muscles and achieve precise stimulation control. Specifically, timing control refers to activating the surface supporting muscle groups first according to the differences in muscle tissue hierarchy and activation threshold, providing a stable platform for the movement of deep muscles. Delaying the start time of coherent electrical stimulation so that it intervenes after the activation of the superficial muscle groups is stable, thereby enhancing the deep electric field focusing effect, which can effectively reduce the misactivation of non-target muscles and enhance the movement pattern matching of the target action.

[0034] Because stimulation frequency directly affects nerve excitation and muscle contraction patterns, this application uses low-frequency stimulation (e.g., 20Hz-50Hz) to target superficial muscle groups, achieving rhythmic contraction and relaxation and reducing the risk of muscle fatigue. High-frequency coherent electrical stimulation (e.g., 2kHz-5kHz) targets deep tissue, forming a low-frequency envelope electric field in the targeted area, effectively breaking through the shielding effect of skin and fat, and enhancing the activation efficiency and targeted stimulation of deep muscle groups.

[0035] The current amplitude directly determines the activation intensity of the muscle. The shallow stimulation electrode is set with a small current (such as 5-15mA) to stabilize the activation of the superficial muscle groups; the deep stimulation electrode is set with a larger current (such as 20-40mA) to overcome the high impedance of the deep tissue and ensure the stimulation effect. This current amplitude control strategy can achieve fine activation of different muscle layers while ensuring the safety and comfort of stimulation, while reducing muscle fatigue caused by long-term stimulation.

[0036] The brain-computer interaction module 2 is used to guide the rehabilitator to generate motor imagery, collect, pre-process, and decode the EEG signals generated by motor imagery, and output the decoding results representing the motor intention; As a possible implementation, see Figure 1 , the brain-computer interaction module 2 includes a virtual animation presentation unit 20, an EEG device 21, a pre-processing unit 22, and a motion intention decoding unit 23; The virtual animation presentation unit 20 is used to present virtual animation to guide the rehabilitator to observe the virtual animation and generate movement imagination; the virtual animation is generated based on the movement imagination task paradigm guided by movement observation; The EEG device 21 collects at least EEG signals generated during motor imagery; A preprocessing unit 22, configured to preprocess the EEG signal; As an example, preprocessing includes at least downsampling, 50Hz notch filtering, 1~40Hz bandpass filtering, myoelectric and eye movement artifact removal, removal of bad leads and bad segments, common average reference, data segmentation, baseline correction and other operations.

[0037] The movement intention decoding unit 23 uses a common space pattern algorithm to extract the EEG features of the EEG signal, and uses a support vector machine to classify the EEG features to achieve decoding of the movement intention and obtain a decoding result.

[0038] As an example, in the specific implementation, the motor imagery task paradigm guided by motion observation is used to intuitively show the patient the task to be performed, guide the patient to complete specific movements, while maintaining the patient's attention and reducing fatigue. The EEG signals of the patient during the task are collected, and the spatial distribution characteristics of the EEG signals of each type of task are extracted based on the Common Spatial Pattern (CSP) algorithm. The Support Vector Machine (SVM) is used for classification to achieve the decoding of the movement intention. CSP finds an optimal spatial filter Maximize the energy difference between the two types of signals after filtering. The calculation is as follows: (3) Where, Indicates the class sample covariance matrix, , The estimation of determines whether the CSP algorithm has a good effect in classification.

[0039] SVM works by finding a classification hyperplane that maximizes the interval between two types of samples. The hyperplane can be described as a linear equation: (4) in, represents the normal vector of the hyperplane, Represents displacement, and the samples closest to the hyperplane are called support vectors, which enable the hyperplane to correctly classify the training samples.

[0040] The multi-channel functional electrical stimulation module 3 receives and calls the stimulation scheme according to the decoding result, injects current into the working electrodes determined by the stimulation scheme, implements electrical stimulation and generates a motor response; As a possible implementation, see Figure 1 and Figure 3 The multi-channel functional electrical stimulation module 3 includes a main control chip, which is equipped with peripheral ports equal to the number of working electrodes. Every two peripheral ports are connected to an independent stimulation channel, and the stimulation current applied to the corresponding two working electrodes is controlled by the independent stimulation channels. For example, Figure 3 As shown in , there are two working electrodes, so the peripheral ports are also configured as two, that is, Figure 3 DAC1 and DAC2 in the device are connected to an independent stimulation channel, which connects electrode 1 and electrode 2. The stimulation current applied to the corresponding two working electrodes is controlled separately through the independent stimulation channel.

[0041] As a possible implementation, a pair of electrodes corresponds to one stimulation channel; the stimulation circuit corresponding to one stimulation channel includes a symmetrically arranged programmable constant current source, a Wilson current mirror, and an H-bridge topology; The main control chip determines the analog voltage signal output by the peripheral port based on the injection current corresponding to each working electrode provided by the stimulation scheme; the analog voltage signal is input to the input port of the programmable constant current source, and after voltage division, a reference voltage signal is obtained to drive the operational amplifier and transistor. Driven by the reference voltage signal, the operational amplifier and transistor form a closed-loop feedback control; based on the virtual short and virtual open characteristics of the operational amplifier, the reference voltage signal is converted into the stimulation current output by the programmable constant current source through a resistor connected in series with the transistor emitter; As an example, see Figure 3 The main control chip determines the analog voltage signal output by the peripheral port DAC1 based on the injection current corresponding to electrode 1 provided by the stimulation scheme. The analog voltage signal enters the input port of the programmable constant current source and generates a reference voltage signal after being divided by resistors R6 and R7. Driven by the reference voltage signal, the operational amplifier U1 and the transistor Q15 form a closed-loop feedback control, forming a constant current source that combines the operational amplifier and the transistor, which is also a programmable constant current source. In specific implementation, the main control chip dynamically outputs the corresponding analog voltage signal based on the stimulation current amplitude set by the user. The analog voltage signal is divided by resistors R6 and R7 to generate a reference voltage signal. , the calculation of this process is shown in formula (5): (5) Still taking the peripheral port DAC1 as an example, based on the virtual short and virtual open characteristics of the operational amplifier, the reference voltage signal is connected to the transistor Q15 emitter through the resistor R10 in series. Converts the stimulus current into a programmable constant current source output , the calculation of this process is shown in formula (6): (6) The input port of the Wilson current mirror is connected to the output port of the programmable constant current source. After replicating and isolating the stimulation current output by the programmable constant current source, it is input into the H-bridge topology structure to achieve control of the stimulation current loaded on the working electrode.

[0042] As an example, see Figure 3 , the Wilson current mirror is composed of Figure 3 The input port of the Wilson current mirror composed of Q5, Q6, Q7, Q8 is connected to the output port of the programmable constant current source, and the stimulation current output by the programmable constant current source is High-precision replication and isolation are performed to significantly improve the consistency of output current and the ability to resist load disturbances. Figure 3 The MOS tubes Q9, Q10, Q11 and Q13 form an H-bridge topology structure to control the stimulation current loaded on the working electrode.

[0043] As a possible implementation, see Figure 3 The Wilson current mirror is integrated between the upper and lower arms of the H-bridge topology structure, and the pulse width modulation signals PWM1 and PWM2 output by the main control chip are used to realize time-sharing conduction control of the upper and lower arms, that is, the pulse width modulation signal PWM1 controls the H-bridge upper arm Q11 and the H-bridge lower arm Q10, and the pulse width modulation signal PWM2 controls the H-bridge upper arm Q13 and the H-bridge lower arm Q9. The main control chip accurately schedules the conduction timing of the pulse width modulation signals PWM1 and PWM2 to form a bipolar constant current output with alternating forward and reverse switching. When the H-bridge upper arm Q11 and the H-bridge lower arm Q10 are controlled to be turned on at the same time and the H-bridge upper arm Q13 and the H-bridge lower arm Q9 are turned off at the same time, a forward stimulation current is output; when the H-bridge upper arm Q13 and the H-bridge lower arm Q9 are controlled to be turned on at the same time and the H-bridge upper arm Q11 and the H-bridge lower arm Q10 are turned off at the same time, a reverse stimulation current is output, thereby forming a high-frequency bidirectional rectangular wave current loaded on the working electrode.

[0044] This application constructs a bidirectional constant-current drive structure centered around "PWM + Wilson current mirror + H-bridge." Multiple constant-current sources, Wilson current mirrors, and H-bridge modules can be expanded as needed, forming a flexible and scalable multi-channel electrical stimulation framework that supports high-precision, high-synchronization, and multi-muscle neuromuscular electrical stimulation applications. This application incorporates multiple safety protection mechanisms. By monitoring the sampling resistor in real time, it can monitor the current level. When the current exceeds a set threshold, it automatically triggers overcurrent protection, promptly shutting off the H-bridge circuit. Furthermore, based on the monitored current level, the system can promptly detect electrode detachment and trigger an early warning. Furthermore, the system features charge accumulation monitoring and control capabilities, capable of real-time recording the on-time of the upper and lower arms of the H-bridge during each stimulation cycle and automatically calculating the accumulated charge. When the accumulated charge offset exceeds a set safety threshold, the system dynamically adjusts the PWM duty cycle and implements a charge neutralization strategy, effectively preventing tissue polarization and electrochemical damage, and improving the biosafety and stability of long-term stimulation.

[0045] See also Figure 1 The motion feedback module 4 collects multimodal feedback signals under motion response and fuses the multimodal feedback signals to generate stimulation parameter adjustment signals and neural feedback signals; wherein, the stimulation parameter adjustment signals are sent to the multi-channel functional electrical stimulation module to adjust the stimulation scheme executed in the previous rehabilitation cycle; the neural feedback signals are sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

[0046] As a possible implementation method, the sports rehabilitation system is applied to the rehabilitation of fine movements of the upper limbs; the sports feedback module 4 includes: an inertial sensor, a bending sensor, a flexible film sensor, a multimodal data extraction unit, a multimodal data fusion unit and a signal decomposition unit; Among them, there are multiple inertial sensors, which are installed on the back of the hand and the fingertips of the five fingers to obtain acceleration and angular velocity. A flexible bending sensor is configured along the extension direction of each finger to obtain the finger bending angle. The posture angle data is determined based on the acceleration, angular velocity and bending angle. The flexible film sensor is attached to the inside of the finger to collect surface pressure data between the fingers or between the fingers and the rehabilitation device. The multimodal data extraction unit receives the posture angle data and the surface pressure data, performs normalization processing, extracts the joint bending angle value from the posture angle data to characterize the posture characteristics of the hand movement, and extracts the pressure value from the surface pressure data to characterize the finger pressure characteristics; The multimodal data fusion unit configures the weights of the joint bending angle value and the pressure value, and obtains the initial fusion data after weighted summation; As an example, the following formula (7) is used to calculate the initial fusion data: (7) Where, and Respectively represent the weights of the configured joint bending angle value and pressure value, A feature map representing the joint bending angle values, A feature map representing pressure values.

[0047] The multimodal data fusion unit is also equipped with a linear state transfer model, which uses the linear state transfer model to take the initial fusion data as input, recursively obtain the optimal estimate of the fusion result, and obtains the multimodal feedback signal based on the optimal estimate; As an example, the Kalman filter algorithm is used to establish a linear state transfer model, as shown in the following formula (8), and the optimal estimate of the fusion result is obtained recursively: (8) Where, represents the current state vector, Represents the state transfer matrix, describing the feedback system from Time has come The evolution of time, Represents process noise and compensates for model instability caused by non-ideal actions or other interference factors.

[0048] The signal decomposition unit generates a stimulation parameter adjustment signal and a neural feedback signal based on the multimodal feedback signal.

[0049] This application constructs an inner-loop feedback path of "real-time perception-feedback-closed-loop execution". When the fusion result shows that the actual action does not meet the expected action of the EEG decoding, the electrical stimulation parameters are adjusted dynamically in time. Based on the data fusion results, the system automatically generates a reward and punishment feedback mechanism and automatically displays it on the screen of the virtual animation presentation unit. The quality of the action can be judged by the color of the prompt on the screen. For example, the green light represents a high degree of action completion, encouraging the user to maintain the current state; the yellow light represents an average degree of action completion, prompting the user to stay focused; the red light represents a poor degree of action completion, requiring the user to adjust the motor imagination strategy; construct an outer-loop feedback mechanism of "long-term learning-optimization closed loop", and continuously improve the optimal rehabilitation strategy through progressive reinforcement learning.

[0050] In addition, this application automatically switches the electrodes and current parameters during exercise based on the joint bending angle value and the action timing decomposition, breaking the traditional electrode fixed stimulation mode and making the training movements more precise and smooth.

[0051] In a second aspect, an embodiment of the present invention provides a motor rehabilitation method based on layered stimulation navigation and closed-loop brain-computer interface, comprising the following steps: Develop a stimulation plan for each stimulation area, including superficial muscle stimulation areas and deep muscle stimulation areas. The stimulation plan includes the working electrodes that need to be activated in the corresponding stimulation area and the injection current corresponding to each working electrode; Guide the rehabilitator to generate motor imagery, collect, pre-process, and decode the EEG signals generated by motor imagery to output decoding results representing motor intention; Receive and call the stimulation plan according to the decoding result, load the working electrode determined by the stimulation plan with injection current, implement electrical stimulation and generate movement response; Multimodal feedback signals under movement response are collected and fused to generate stimulation parameter adjustment signals and neural feedback signals; the stimulation parameter adjustment signals are sent to the multi-channel functional electrical stimulation module to adjust the stimulation scheme executed in the previous rehabilitation cycle; the neural feedback signals are sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

[0052] Next, we will use patient Tang as an example to illustrate the implementation of this invention. Patient Tang, 51, was six months post-stroke and in the recovery phase. He had high muscle tone, and his hand naturally held a semi-clenched fist. Based on his condition, he underwent hand extension rehabilitation training at the doctor's recommendation. Pre-enrollment functional assessments and preliminary experiments revealed that conventional electrical stimulation was limited to surface stimulation. However, because the muscles involved in thumb extension are mostly deep, thumb extension is difficult, especially in the distal phalanx.

[0053] Before enrollment in the rehabilitation training program, MRI images of the patient's arm were collected. Deep learning methods were used to segment and annotate the skin, fat, and muscle in the MRI images, generating a 2D annotation set. A 3D voxel interpolation algorithm was then used to interpolate the 2D annotation set. Based on neuroanatomy, the muscles involved in finger extension were annotated, with a focus on the extensor pollicis longus, extensor pollicis brevis, and abductor pollicis longus muscles, which are associated with thumb extension. This led to the construction of a personalized 3D forearm electrical stimulation navigation model.

[0054] For the constructed personalized forearm 3D electrical stimulation navigation model, the superficial and deep muscle stimulation areas were assigned electrical characteristic values ​​corresponding to the frequency, and the tissue types were assigned conductivity. In the finite element calculation, the conductivity of the arm bones was defined as 0.0202 S / m, the conductivity of the arm muscles as 0.33 S / m, the conductivity of the arm fat as 0.0423 S / m, and the conductivity of the arm skin as 0.0002 S / m.

[0055] The STM32F429 is the main control chip, supporting multiple PWM outputs and synchronized DAC outputs. The DAC module of the main control chip outputs analog signals for linear regulation of the stimulation current. During stimulation, current, charge balance, and electrode detachment are monitored in real time. If a fault occurs, the stimulation module automatically shuts off power.

[0056] A layered electrical stimulation scheme was constructed for thumb extension. The superficial muscles focused on the extensor carpi radialis longus, with the main purpose of providing mechanical support for thumb extension. Electrode positioning was performed based on MRI image segmentation, using a bidirectional rectangular wave. To address the problem of unsatisfactory extension of the distal phalanx of the thumb, the focus was on the related deep muscles, including the extensor pollicis longus, extensor pollicis brevis, and abductor pollicis longus. Since the three muscles are adjacent and have a small area, two pairs of electrodes were used for high-frequency coherent electrical stimulation. The electrodes were placed through 3D model simulation so that their interference area covered the cluster. The two pairs of electrodes injected high-frequency bidirectional rectangular waves of 2000Hz and 2050Hz, respectively, to generate an electric field focus range covering the extensor pollicis longus, extensor pollicis brevis, and abductor pollicis longus muscles.

[0057] A dual-decoding experimental paradigm of "movement observation + motor imagery" was established, allowing patients to observe a virtual animation of hand extension while simultaneously imagining the corresponding movement. This guided them to focus on a specific motor task, thereby stimulating their motor intention. An EEG acquisition device recorded the patient's brain activity in real time. CSP extracted EEG features related to motor intention. Combined with the output of the SVM classification model, these classification results were converted into control commands, which then activated an external functional electrical stimulation device to assist the patient in completing the hand extension movement.

[0058] See also Figure 4 The system's workflow is demonstrated in the figure below. The timeline below, starting at -4 seconds, presents the complete treatment cycle: First, a virtual animation of an open hand guides the patient into a preparatory state. Next, the task phase begins, collecting and analyzing EEG signals in real time while the patient imagines opening their hand. Based on the decoded results, functional electrical stimulation is triggered to achieve precise neuromuscular stimulation. Finally, a rest phase ensures the quality of subsequent training. Ultimately, a closed rehabilitation loop of "EEG signal acquisition - intelligent decision-making - electrical stimulation output" is established, enabling efficient human-computer interaction and rehabilitation training.

[0059] Patients wear data gloves equipped with multiple flexible bending sensors and inertial sensors to record and analyze the bending angles of each thumb joint in real time. Flexible thin-film pressure sensors are used to measure the force applied to the thumb, and multimodal signal fusion is performed in combination with EEG to assess motor ability. Threshold grading is performed based on multimodal fusion. If the movement meets the standard, a green light is displayed to maintain the current stimulation parameters. If there is a slight deviation from the expected movement, a yellow light is displayed, and the functional electrical stimulation parameters are automatically fine-tuned at regular intervals to ensure the accuracy of muscle contraction. If the movement is seriously deviated from the expected movement, a red light is displayed, requiring immediate adjustment of the stimulation parameters, and the patient is reminded to concentrate on completing the motor imagery task to achieve short-term inner-loop feedback. If the red light is displayed three times in a row, the experiment needs to be suspended for a safety check.

[0060] In addition, based on long-term training, the global rehabilitation strategy is optimized through machine learning algorithms to achieve individualized adaptation. In the early stage of rehabilitation, the bending angle weight accounts for a large proportion in the feedback process, the EEG weight accounts for a small proportion, and the stimulation current amplitude is large; through long-term training, the EEG weight proportion gradually increases, and the stimulation current amplitude required to complete the movement decreases, reducing dependence on external stimulation, promoting autonomous control, and achieving long-term adaptive optimization of the outer loop feedback. The inner loop is based on multimodal data fusion, which significantly reduces the feedback error compared to single signal feedback. Compared with fixed rehabilitation plans, outer loop control significantly shortens the rehabilitation cycle. Ultimately, the dual-loop synergy of "short-term precise control" and "long-term adaptive optimization" is achieved to complete the functional reconstruction of fine movements.

[0061] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the accompanying drawings. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit can implement several functions listed in the specification. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0062] Although the present invention has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations thereof may be made without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the present invention and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the invention. It will be apparent that various modifications and variations of the present invention may be made by those skilled in the art without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such modifications and variations as fall within the scope of the invention and its equivalents.

Claims

1. A motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface, characterized in that: include: The layered stimulation navigation module receives image data and outputs a stimulation plan through modeling, tissue segmentation and annotation, stimulation area setting, and optimization algorithm solution. The stimulation areas include superficial muscle stimulation areas and deep muscle stimulation areas. The stimulation plan includes the working electrodes to be activated in the corresponding stimulation areas and the corresponding injection current for each working electrode. The brain-computer interaction module guides the patient to generate motor imagery, collects, pre-processes, and decodes the EEG signals generated by motor imagery, and outputs decoding results that represent motor intentions; The multi-channel functional electrical stimulation module receives and calls the stimulation scheme according to the decoding results, injects current into the working electrodes determined by the stimulation scheme, implements electrical stimulation and generates a motor response; The motion feedback module collects multimodal feedback signals under motion response and fuses the multimodal feedback signals to generate stimulation parameter adjustment signals and neural feedback signals; among them, the stimulation parameter adjustment signal is sent to the multi-channel functional electrical stimulation module to adjust the stimulation plan executed in the previous rehabilitation cycle; the neural feedback signal is sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

2. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 1 is characterized in that: The hierarchical stimulation navigation module includes a tissue structure image acquisition unit, a tissue segmentation and annotation unit, a modeling unit, a stimulation area setting unit, and a solution unit; Wherein, the tissue structure image acquisition unit obtains image data of the recovered person; The tissue segmentation and annotation unit combines anatomical structures and uses deep learning methods to segment and annotate the skin, fat, and muscle in the image data to obtain a two-dimensional annotation set; The modeling unit uses a three-dimensional voxel interpolation algorithm to perform voxel interpolation to establish a three-dimensional model; The stimulation area setting unit, based on the established three-dimensional model, assigns electrical characteristic values ​​corresponding to the frequency to the superficial muscle stimulation area and the deep muscle stimulation area. The electrical characteristic values ​​are equal to the electrical characteristic values ​​of human tissue at the corresponding frequency. The finite element method is used to solve the unit electric field distribution when a single pair of electrodes is activated. According to rehabilitation needs or anatomical positioning, the superficial muscle stimulation area and the deep muscle stimulation area are divided in the three-dimensional model based on the unit electric field distribution, and the desired effective electric field threshold is set accordingly. The solving unit uses an optimization algorithm and takes the effective electric field threshold as input to solve the initial stimulation plan, and continuously optimizes the electrode layout and injection current through simulation iteration to finally obtain the optimal stimulation plan.

3. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 2 is characterized in that: In the stimulation scheme, low-frequency electrical stimulation is used for the superficial muscle stimulation area, and coherent electrical stimulation is used for the deep muscle stimulation area.

4. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 3 is characterized in that: The stimulation conditions for coherent electrical stimulation are formed by two pairs of working electrodes located at different positions around the target area; the stimulation mechanism of coherent electrical stimulation is: high-frequency bidirectional rectangular wave current is applied to the two pairs of working electrodes, and the high-frequency bidirectional rectangular wave current is driven by a group of high-frequency signals with a preset frequency difference; the high-frequency bidirectional rectangular wave current undergoes coherent interference in the deep muscle stimulation area, forming a low-frequency envelope electric field with a preset frequency difference.

5. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 4 is characterized in that: The multi-channel functional electrical stimulation module includes a main control chip, which is equipped with peripheral ports equal to the number of working electrodes. Every two peripheral ports are connected to an independent stimulation channel, and the stimulation current applied to the corresponding two working electrodes is controlled separately through the independent stimulation channels.

6. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 5 is characterized in that: A pair of electrodes corresponds to one stimulation channel; the stimulation circuit corresponding to one stimulation channel includes a symmetrically arranged programmable constant current source, a Wilson current mirror, and an H-bridge topology; The main control chip determines the analog voltage signal output by the peripheral port based on the injection current corresponding to each working electrode provided by the stimulation scheme; the analog voltage signal is input to the input port of the programmable constant current source, and after voltage division, a reference voltage signal is obtained to drive the operational amplifier and transistor. Driven by the reference voltage signal, the operational amplifier and transistor form a closed-loop feedback control; based on the virtual short and virtual open characteristics of the operational amplifier, the reference voltage signal is converted into the stimulation current output by the programmable constant current source through a resistor connected in series with the transistor emitter; The input port of the Wilson current mirror is connected to the output port of the programmable constant current source. After replicating and isolating the stimulation current output by the programmable constant current source, it is input into the H-bridge topology structure to achieve control of the stimulation current loaded on the working electrode.

7. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 6 is characterized in that: The Wilson current mirror is integrated between the upper and lower arms of the H-bridge topology structure, and the pulse width modulation signal output by the main control chip is used to realize the time-sharing conduction control of the upper and lower arms, forming a bipolar constant current output that alternates between forward and reverse directions, thereby forming a high-frequency bidirectional rectangular wave current loaded on the working electrode.

8. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 1 is characterized in that: The brain-computer interaction module includes a virtual animation presentation unit, EEG equipment, a preprocessing unit, and a motion intention decoding unit; The virtual animation presentation unit is used to present virtual animation to guide the rehabilitator to observe the virtual animation and generate movement imagination; the virtual animation is generated based on the movement imagination task paradigm guided by movement observation; EEG equipment, at least for collecting EEG signals generated during motor imagery; A preprocessing unit, used for preprocessing EEG signals; The movement intention decoding unit uses the common space pattern algorithm to extract the EEG features of the EEG signal, and uses the support vector machine to classify the EEG features to realize the decoding of the movement intention and obtain the decoding result.

9. The motor rehabilitation system based on layered stimulation navigation and closed-loop brain-computer interface according to claim 1, characterized in that: The sports rehabilitation system is used for the rehabilitation of fine motor skills of the upper limbs; the sports feedback module includes: inertial sensor, bending sensor, flexible film sensor, multimodal data extraction unit, multimodal data fusion unit and signal decomposition unit; Among them, there are multiple inertial sensors, which are installed on the back of the hand and the fingertips of the five fingers to obtain acceleration and angular velocity. A flexible bending sensor is configured along the extension direction of each finger to obtain the finger bending angle. The posture angle data is determined based on the acceleration, angular velocity and bending angle. The flexible film sensor is attached to the inside of the finger to collect surface pressure data between the fingers or between the fingers and the rehabilitation device. The multimodal data extraction unit receives the posture angle data and the surface pressure data, performs normalization processing, extracts the joint bending angle value from the posture angle data to characterize the posture characteristics of the hand movement, and extracts the pressure value from the surface pressure data to characterize the finger pressure characteristics; The multimodal data fusion unit configures weights for the joint bending angle and pressure values, and obtains the initial fusion data through weighted summation. The multimodal data fusion unit is also configured with a linear state transfer model, which uses the linear state transfer model to take the initial fusion data as input, recursively obtains the optimal estimate of the fusion result, and obtains the multimodal feedback signal based on the optimal estimate. The signal decomposition unit generates a stimulation parameter adjustment signal and a neural feedback signal based on the multimodal feedback signal.

10. A motor rehabilitation method based on layered stimulation navigation and closed-loop brain-computer interface, characterized in that: The steps include: Develop a stimulation plan for each stimulation area, including superficial muscle stimulation areas and deep muscle stimulation areas. The stimulation plan includes the working electrodes that need to be activated in the corresponding stimulation area and the injection current corresponding to each working electrode; Guide the rehabilitator to generate motor imagery, collect, pre-process, and decode the EEG signals generated by motor imagery to output decoding results representing motor intention; Receive and call the stimulation plan according to the decoding result, load the working electrode determined by the stimulation plan with injection current, implement electrical stimulation and generate movement response; Multimodal feedback signals under movement response are collected and fused to generate stimulation parameter adjustment signals and neural feedback signals; the stimulation parameter adjustment signals are sent to the multi-channel functional electrical stimulation module to adjust the stimulation scheme executed in the previous rehabilitation cycle; the neural feedback signals are sent to the brain-computer interaction module to guide the rehabilitator to generate movement imagination that is more in line with the movement intention.

Images