A stroke patient upper limb function reconstruction system

Abstract

The application relates to a stroke patient upper limb function reconstruction system, which at least comprises: a signal acquisition unit for acquiring physiological electrical signals of a stroke patient; a signal processing unit for processing the physiological electrical signals; and an upper limb training unit for assisting upper limb movement based on the control of the signal processing unit. The signal acquisition unit can at least acquire electroencephalogram signals of the stroke patient and first electromyogram signals generated by the upper limb under the assistance of the upper limb training unit, wherein the signal processing unit extracts the attention level of the patient according to the electroencephalogram signals, and the upper limb training unit assists the upper limb to move when the attention level reaches a preset value; the signal processing unit can also derive second electromyogram signals according to the electroencephalogram signals, and the signal processing unit continuously corrects the mapping relationship between the electroencephalogram signals and the motor imagery task based on the difference between the first electromyogram signals and the second electromyogram signals, so as to gradually improve the accuracy of the mapping relationship between the electroencephalogram signals and the motor imagery task.

Description

Technical Field

[0001] This invention relates to the field of stroke rehabilitation training technology, and in particular to a system for reconstructing upper limb function in stroke patients. Background Technology

[0002] Stroke, also known as apoplexy or cerebrovascular accident, is defined by the World Health Organization as "an acute neurological disorder of vascular origin, with symptoms and signs consistent with the site of brain injury." In cerebrovascular patients, various triggering factors cause narrowing, occlusion, or rupture of cerebral arteries, resulting in acute cerebral circulatory disturbances. Clinically, this manifests as transient or permanent brain dysfunction symptoms and signs, primarily including cerebral hemorrhage, cerebral infarction, and subarachnoid hemorrhage. It is estimated that the incidence of stroke in my country is as high as 1.5 million per year, with 75% of stroke patients experiencing functional impairment, the vast majority being motor dysfunction, primarily hemiplegia. Motor dysfunction causes physical pain and inconvenience to patients. Compared to the lower limbs, upper limb motor dysfunction is more difficult to recover. The recovery of upper limb function in most stroke patients progresses from proximal to distal, from coarse to fine motor functions, and from simple to complex. Rehabilitation of hemiplegic upper limb function after stroke remains one of the most challenging clinical problems. Studies have shown that, in addition to surgery and drug treatment, scientific rehabilitation training plays an important role in the limb function recovery of stroke patients with hemiplegia.

[0003] In the prior art, such as the robot-assisted multimodal mirror rehabilitation training method and scoring system for upper limbs of stroke hemiplegic patients proposed in patent document CN113679568B, the system is reasonable and innovative. Based on the mirror rehabilitation training theory in rehabilitation medicine, it designs a robot-assisted upper limb mirror rehabilitation training method and uses a multimodal approach to randomly generate set training trajectories, which helps to improve training efficiency. In terms of safety, the robot is equipped with an emergency stop button to prevent secondary injury to the patient during training. In terms of data processing and storage, the system fits the actual training generated trajectory with the set trajectory to calculate the degree of overlap, providing a basis for doctors to quantify the scoring, and finally stores the training information in a cloud database, which helps to compare and analyze multiple rehabilitation training data.

[0004] Most existing rehabilitation training methods use mechanical structures such as robots and lifting devices to drive the patient's upper limbs to perform repetitive and monotonous movements according to a pre-set movement trajectory. The movements are mechanically driven, and the functional control connection between the movements and the brain is relatively limited. In other words, when the patient's upper limbs are undergoing the above-mentioned rehabilitation training, the connection between the movement state and the brain's consciousness is broken. Although this method can achieve a certain recovery effect, it is extremely difficult to gradually restore the patient's brain's control over the upper limbs through this method.

[0005] Furthermore, existing technologies for upper limb rehabilitation training systems for stroke patients require stimulation of specific body parts based on characteristic commands. Traditional stimulation methods involve non-invasive physical stimulation of the brain, central nervous system, and muscles using sound, light, electromagnetic, and other physical stimuli. However, during long-term upper limb functional rehabilitation training for stroke patients, this can lead to numbness, loss of sensation, fear, or damage to sensory organs, easily causing patients to lose interest and confidence in training, and significantly impacting both the patient and the rehabilitation process. Therefore, there is an urgent need for a system that can reconstruct upper limb function in stroke patients without requiring external stimulation, and that can gradually improve the accuracy and stability of predictions through machine learning, thus providing an effective system for upper limb functional reconstruction in stroke patients.

[0006] Furthermore, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the inventors studied a large number of documents and patents when making this invention, but due to space limitations, not all details and contents were listed in detail. However, this does not mean that the present invention does not possess the features of these prior art. On the contrary, the present invention already possesses all the features of the prior art, and the applicant reserves the right to add relevant prior art to the background art. Summary of the Invention

[0007] To address the shortcomings of existing technical solutions, this application proposes an upper limb function reconstruction system for stroke patients, comprising at least: a signal acquisition unit for acquiring physiological electrical signals from stroke patients; a signal processing unit for processing physiological electrical signals; and an upper limb training unit for assisting upper limb movement based on the control of the signal processing unit. The signal acquisition unit is capable of acquiring at least the electroencephalogram (EEG) signals of the stroke patient and the first electromyographic (EMG) signals generated by the upper limb under the assisted movement of the upper limb training unit. The signal processing unit establishes a mapping relationship between the EEG signals and the motor imagery task based on the EEG signals. The signal processing unit can substitute the EEG signals into the EEG-EMG mapping model to derive a second EMG signal. The signal processing unit continuously corrects the mapping relationship between the EEG signals and the motor imagery task based on the difference between the first and second EMG signals.

[0008] The signal processing unit continuously modifies the motor imagery task determined by the EEG signal based on the actual movement trajectory generated by the electromyographic signals of the upper limb movement. This allows for the acquisition of upper limb movements that are more in line with the brain's motor imagery task in the next assisted movement, thereby gradually restoring the brain's control connection to the upper limbs in stroke patients and enabling them to gradually regain the ability to voluntarily control their upper limb movements.

[0009] Motor imagery tasks involve participants imagining a movement or action without actually moving their bodies. In this task, participants activate neuronal populations and neural circuits by imagining, representing, planning, and executing an activity, in a manner similar to the neural activation that occurs when the movement or action is actually performed. This type of task is commonly used to study the brain's motor control and sensory systems. It offers many advantages, such as the ability to complete simple and repetitive motor imagery tasks, develop more effective treatment programs instead of actual movement, and avoid negative physical effects. Furthermore, motor imagery tasks can be used with brain-computer interfaces to convert imagined movements into action control signals, thereby enabling the control of machines or external devices.

[0010] Preferably, the signal processing unit acquires the mapping relationship between EEG signals and motor imagery tasks based on a continuously trained motor intention recognition model. Specifically, each time the upper limb training unit performs an upper limb assisted movement, it uses the generated EEG signal and the corrected motor imagery task as training data to train the motor intention recognition model. The motor imagery task is corrected based on differences in electromyographic (EMG) signals. Specifically, the original surface EMG signal, as the most direct representation of the occurrence and resting state of EMG activity, can be analyzed for its initial relationship without considering amplitude. That is, the density and height of the original EMG signal during muscle activity can, to a certain extent, reflect the amplitude and force of muscle contraction. The higher the density and height, the stronger the surface EMG signal, and thus the stronger the contraction. Therefore, the information processing unit can infer the difference between the desired motor imagery task and the actual motor imagery task displayed in limb movement through the difference between the second and first EMG signals, thereby obtaining the corrected motor imagery task.

[0011] Preferably, the process by which the signal processing unit constructs the motion intent recognition model is as follows:

[0012] Multiple sets of sample information were obtained from normal individuals performing upper limb movements. Each set of sample information included sample electroencephalogram (EEG) signals and sample electromyogram (EMG) signals.

[0013] A motion intent recognition model based on a dual-stream Transformer encoder and a multi-head attention mechanism is established based on the information of each set of samples.

[0014] The mapping relationship between EEG signals and motor imagery tasks is determined based on a motor intention recognition model.

[0015] Preferably, the process by which the signal processing unit constructs the electroencephalogram and electromyogram mapping model is as follows:

[0016] Several normal subjects were selected to perform voluntary upper limb movements, and electroencephalogram (EEG) and electromyogram (EMG) signals were collected. The EEG and EMG signals were collected simultaneously.

[0017] Variational mode decomposition was performed on the preprocessed EEG and EMG signals to decompose them into several different intrinsic mode functions;

[0018] The transfer entropy between each pair of EEG and EMG signals obtained by variational mode decomposition is calculated, and an EEG-EMG mapping model is constructed based on the calculation results.

[0019] Preferably, the signal processing unit continuously corrects the motor imagery task derived from the electroencephalogram (EEG) signal based on the difference between the first and second EMG signals. The signal processing unit derives the motor imagery task based on the EEG signal and calculates the corrected motor imagery task based on the difference between the first and second EMG signals.

[0020] Preferably, the signal acquisition unit includes at least a brain electronics unit for acquiring electroencephalogram (EEG) signals and a muscle electronics unit for acquiring electromyogram (EMG) signals, and both the brain electronics unit and the muscle electronics unit maintain a data connection with the signal processing unit.

[0021] Preferably, the signal processing unit is at least capable of analyzing the patient's attention level based on the EEG signals acquired by the brain electronics unit in a single-channel manner. The signal processing unit then controls the upper limb training unit to perform assisted movements based on the attention level. The single-channel EEG signal acquisition method refers to the brain electronics unit using a single electrode to acquire the amplitude of the EEG signal. While the spatial resolution of the EEG signal acquired by the single-channel brain electronics unit is lower, its temporal resolution is not significantly different from that of the multi-channel EEG signal acquisition method. Furthermore, the single-channel configuration makes the brain electronics unit lighter and more portable. The brain electronics unit of the acquisition module acquires EEG signals in a single-channel manner, and the signal processing unit extracts attention level features based on the single-channel EEG signals to obtain an attention level value. This value is then compared with a preset attention level value. If the current attention level value is greater than or equal to the preset attention level value, the signal processing unit sends the motion information parameters of the target movement posture to the upper limb training unit. The upper limb training unit then performs assisted movements on the user's upper limbs based on these motion information parameters.

[0022] Preferably, the signal processing unit is at least capable of substituting the EEG signals acquired by the brain electronics unit in a multi-channel manner into the EEG-EMG mapping model to obtain a second EMG signal. The so-called multi-channel EEG signal acquisition method refers to the configuration of multiple measuring electrodes used by the brain electronics unit to acquire the amplitude of the EEG signal, which are distributed on the scalp of the stroke patient according to certain rules. Specifically, the most commonly used method for multi-channel EEG signal acquisition is the international 10-20 system method. This method is a standard method recommended by the International Electroencephalography Society. It is named the 10-20 system because the distance from the midpoint of the forehead to the root of the nose and the distance from the occipital point to the external occipital protuberance each account for 10% of the total length of the line connecting these points, and the remaining points are separated by 20% of the total length of this line. It contains a total of 21 acquisition channels.

[0023] Brainwave vibrations originate from the electrical activity of neurons in the brain. When brain neurons are active, they generate weak electrical signals, which are transmitted through cell membranes. These electrical signals are brainwaves, and their frequency and amplitude can reflect the state and activity level of the brain. Since the generation of brainwaves is closely related to psychological activities such as emotions, thinking, cognition, and attention, it can be used to study the patterns of brain activity under different cognitive and psychological states.

[0024] Preferably, the electromyography unit is disposed on the skin surface of the upper limb of a stroke patient, and it acquires the first electromyographic signal through a multi-channel method, including at least an electromyographic signal sensor capable of acquiring electromyographic signals of the upper arm, forearm and / or palm of the upper limb respectively.

[0025] Preferably, several electromyographic (EMG) signal sensors of the myoelectronic unit are positioned between the skin surface of the stroke patient's upper limb and the upper limb training unit worn on the upper limb, and are held by the upper limb training unit. The myoelectronic unit configured on the surface of the stroke patient's upper limb can be combined with the upper limb training unit for auxiliary configuration. Specifically, the upper limb training unit is essentially a drive structure configured in the form of a mechanical exoskeleton. It covers the upper arm, forearm, palm, and fingers of the stroke patient to drive the movement of various parts of the stroke patient's upper limb. Thus, the myoelectronic unit can be configured between the skin of the stroke patient's upper limb and the exoskeleton of the upper limb training unit, with the exoskeleton of the upper limb training unit providing fixed support to the myoelectronic unit, preventing the myoelectronic unit from accidentally falling off the skin surface.

[0026] Preferably, the upper limb training unit is worn as an exoskeleton on the upper limb of a stroke patient, comprising at least shoulder bones, upper arm bones, forearm bones, and finger bones. The upper limb training unit also includes a wearing part for wearing. The signal processing unit converts the EEG signals acquired by the signal acquisition unit into a motor imagery task and outputs the corresponding motor control signal to the upper limb training unit. The upper limb training unit drives the upper limb exoskeleton configuration to perform movements based on the motor control signal, thereby assisting the stroke patient's upper limb in performing motor imagery tasks based on EEG signals.

[0027] Based on the above scheme, the upper limb training unit of the stroke patient upper limb function reconstruction system of this application can autonomously control the upper limb training unit to assist the paralyzed upper limb in movement based on electroencephalogram (EEG) signals. There is no need for additional medical staff or caregivers to manually pull the stroke patient to move. Moreover, the movement control signal of the upper limb training unit determined by EEG signals can increase the probability of restoring the control connection between the EEG signal and the paralyzed upper limb compared to movement control applied manually by other people outside the patient or simply mechanically applied according to a few preset postures. This avoids ineffective or inefficient repetitive and mechanical upper limb rehabilitation exercise training.

[0028] In addition, the signal processing unit performs at least two signal processing procedures after acquiring the EEG signal. First, it converts the EEG signal into a corresponding second electromyographic signal using a brain-myoelectric coupling research method based on variational mode decomposition-transfer entropy. Second, it uses a trained motor intention recognition model to derive the mapping relationship between the EEG signal and the motor imagery task. After each upper limb movement training session, the myoelectric unit of the signal acquisition unit acquires a first electromyographic signal representing the actual upper limb movement. A correction value is derived based on the difference between the first and second electromyographic signals, and the mapping relationship between the EEG signal and the motor imagery task is corrected based on this correction value. Specifically, the corrected motor imagery task is derived from the correction value and the electromyographic signal. The EEG signal from this movement and the corrected motor imagery task are used as a training set to train the motor intention recognition model again. After each subsequent movement, the signal processing unit trains the motor intention recognition model in the above manner, thereby continuously improving the accuracy of the motor intention recognition model and further enhancing the mapping relationship between the EEG signal and the motor imagery task.

[0029] According to a preferred embodiment, this application also proposes an upper limb function reconstruction system for stroke patients, which can be applied to upper limb rehabilitation training for stroke patients in individual families. Common upper limb training devices require various signal stimuli, and the stimulation methods need to be supervised by professionals to avoid excessive stimulation leading to adverse training effects. However, for most families, the cost of professional training guidance is extremely expensive, and regular or long-term visits to training facilities are required, increasing the additional burden of rehabilitation training. The upper limb function reconstruction system of this application drives the upper limb training unit by analyzing the patient's attention and EEG signals. When the patient's attention level reaches a certain level, the upper limb training unit is activated to drive upper limb rehabilitation movements. The specific movement form is determined by EEG signals or directly observed video images, without the need for additional stimulation. This makes it more portable for individual families, reduces the difficulty of use, and eliminates the problem of serious consequences caused by improper use. Attached Figure Description

[0030] Figure 1 This is a simplified diagram of the module interaction structure of the upper limb function reconstruction system for stroke patients of the present invention.

[0031] Figure 2 This is a simplified structural diagram of the multi-channel brain electronics unit of the signal acquisition unit of the upper limb function reconstruction system for stroke patients of the present invention.

[0032] Figure 3 This is a simplified structural diagram of a single-channel brain electronics unit of the signal acquisition unit of the upper limb function reconstruction system for stroke patients of the present invention.

[0033] Figure 4 This is a simplified structural diagram of the upper limb training unit of the upper limb function reconstruction system for stroke patients of the present invention.

[0034] List of reference numerals

[0035] 100: Signal acquisition unit; 200: Signal processing unit; 300: Upper limb training unit; 110: Brain electronics unit; 111: Measuring electrode; 112: Reference electrode; 120: Myoelectronic unit; 310: Shoulder skeleton; 320: Upper arm skeleton; 330: Forearm skeleton; 340: Finger skeleton; 350: Wearable part. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings.

[0037] Example 1

[0038] Figure 1This diagram illustrates a simplified relational structure of the module interactions of the upper limb function reconstruction system for stroke patients according to this application. The system includes a signal acquisition unit 100, a signal processing unit 200, and an upper limb training unit 300. The signal processing unit 200 determines the stroke patient's motor imagery task based on the electroencephalogram (EEG) signals acquired by the signal acquisition unit 100 and converts the motor imagery task into a motor control signal, which is then sent to the upper limb training unit 300. The upper limb training unit 300 assists the stroke patient's upper limbs in movement based on the motor control signals sent by the signal processing unit 200. During the upper limb movement, the signal acquisition unit 100 also acquires the electromyographic (EMG) signals generated by the upper limb movement. The signal processing unit 200 continuously modifies the motor imagery task determined by the EEG signals based on the actual movement trajectory generated by the EMG signals, so as to obtain an upper limb movement that is more in line with the brain's motor imagery task in the next assisted movement, thereby gradually restoring the brain's control connection to the upper limbs and enabling the stroke patient to gradually regain the ability to voluntarily control the upper limb movement.

[0039] Preferably, the signal acquisition unit 100 includes at least a brain electronics unit 110 for acquiring electroencephalogram (EEG) signals and a muscle electronics unit 120 for acquiring upper limb movement signals.

[0040] Preferably, the brain electronics unit 110 of the signal acquisition unit 100 can be designed as a headset structure, which is worn on the head of a stroke patient. The brain electronics unit 110 mainly acquires brain signals by collecting electroencephalogram (EEG) signals. EEG acquisition is a commonly used non-invasive technology for measuring brain activity. It monitors voltage fluctuations generated by current in human neurons by measuring the amplitude of brain wave signals through electrodes placed on the scalp.

[0041] Preferably, the brain electronics unit 110 can employ single-channel or multi-channel EEG signal acquisition. According to... Figure 2 As shown, the so-called multi-channel EEG signal acquisition method refers to the configuration of multiple measuring electrodes 111 used by the brain electronic unit to acquire the amplitude of brain wave signals, which are distributed on the scalp of the stroke patient according to certain rules. Specifically, the most commonly used method for multi-channel EEG signal acquisition is the international 10-20 system method. This method is a standard method recommended by the International Society for Electroencephalography (ESE). It is named the 10-20 system because the distance from the midpoint of the forehead to the root of the nose and the distance from the occipital point to the external occipital protuberance each account for 10% of the total length of this line, and all other points are separated by 20% of the total length of this line. It contains a total of 21 acquisition channels. Figure 3As shown, the so-called single-channel EEG signal acquisition method refers to the fact that the electrode configuration of the brain electronic unit 110 for acquiring the amplitude of the brain wave signal is a single one. The EEG signal acquired by the single-channel brain electronic unit 110 has a lower spatial resolution, but the temporal resolution is not much different from that of the multi-channel EEG signal acquisition method. Moreover, the single channel makes the brain electronic unit 110 lighter and more portable.

[0042] Preferably, the brain electronics unit 110 of this application acquires brain signals for constructing a second electromyographic signal through a multi-channel brain signal acquisition method, and acquires brain signals for judging the user's attention level through a single-channel brain signal acquisition method.

[0043] Preferably, after acquiring the EEG signal, the brain electronics unit 110 first performs preprocessing by its built-in preprocessing module. Preprocessing includes filtering, amplification, and A / D conversion of the acquired EEG signal. Specifically, the preprocessing module in the brain electronics unit 110 can utilize the TGAM module found in the MindWave mobile EEG acquisition headset from NeuroSky. The preprocessing module transmits the acquired EEG signal to the signal processing unit 200 for analysis, enabling the signal processing unit 200 to acquire the motor imagery task of the stroke patient.

[0044] Preferably, the signal acquisition unit 100 further includes a myoelectronic unit 120 capable of acquiring the actual movement trajectory of the upper limb. The electromyographic signal acquired by the myoelectronic unit 120 is essentially a surface electromyographic signal. Surface electromyography (sEMG) is the superposition of action potentials of motor units in muscle fibers in time and space. The nervous system controls the activity of muscles (contraction or relaxation). Different muscle fiber motor units on the surface skin generate different signals at the same time. Specifically, the surface electromyographic signal acquired by the myoelectronic unit 120 is recorded as the first electromyographic signal.

[0045] Preferably, the electromyography unit 120 is disposed on the skin surface of the upper limb of the stroke patient, and it acquires the first electromyographic signal through a multi-channel method. Specifically, it includes at least an electromyographic signal sensor for acquiring electromyographic signals at the upper arm, at least an electromyographic signal sensor for acquiring electromyographic signals at the forearm, and at least an electromyographic signal sensor for acquiring electromyographic signals at the palm.

[0046] Preferably, the myoelectronic unit 120 disposed on the surface of the upper limb of the stroke patient can be configured in conjunction with the upper limb training unit 300. Specifically, the upper limb training unit 300 is essentially a drive structure configured in the form of a mechanical exoskeleton. It covers the upper arm, forearm, palm, and fingers of the stroke patient to drive the movement of various parts of the stroke patient's upper limb. Thus, the myoelectronic unit 120 can be disposed between the skin of the stroke patient's upper limb and the exoskeleton of the upper limb training unit 300. The exoskeleton of the upper limb training unit 300 provides fixed support to the myoelectronic unit 120, preventing the myoelectronic unit 120 from accidentally falling off the skin surface.

[0047] Preferably, the electromyography unit 120 further includes a low-pass filter connected to the electromyography signal sensor for filtering the electromyography signals acquired by the initial plurality of electromyography signal sensors.

[0048] Preferably, the electromyography unit 120 further includes a signal amplifier connected to a low-pass filter. The initial electromyography signal acquired by the electromyography signal sensor is filtered by the low-pass filter and then amplified by the signal amplifier.

[0049] Preferably, the electromyography unit 120 further includes an analog-to-digital converter (ADC). The ADC is connected to a signal amplifier. The initial electromyography signal acquired by the electromyography signal sensor is filtered by a low-pass filter and amplified by a signal amplifier. The amplified electromyography signal is then converted into a digital signal by the ADC, thereby obtaining a surface electromyography signal that can be processed and analyzed by the signal processing unit 200, which is denoted as the final first electromyography signal.

[0050] Preferably, the signal flow of the electromyography unit 120 is mainly as follows: electromyographic signals are acquired from several electromyographic signal sensors attached to the skin surface; the electromyographic signal sensors transmit the electromyographic signals to a low-pass filter for filtering; the low-pass filter then transmits the processed electromyographic signals to a signal amplifier for amplification; after amplification, the signal amplifier transmits the electromyographic signals to an analog-to-digital converter to convert the electromyographic signals into digital signals that can be recognized by the signal processing unit 200. For ease of understanding, this application refers to this digital signal as a surface electromyographic signal and denotes it as the first electromyographic signal.

[0051] Preferably, the signal processing unit 200 can determine the accuracy of the motor control signals analyzed from the EEG signals of stroke patients based on a pre-set EEG-EMG mapping model, and continuously correct the relationship between the EEG signals and the motor control signals based on the EEG-EMG mapping model, so that the actual movement trajectory under the assisted movement of the upper limb training unit 300 is consistent with the motor imagination task in the brain of stroke patients, thereby enhancing the accuracy of the connection between EEG signals and upper limb movement. Furthermore, accurate and effective training helps to increase the possibility of stroke patients regaining the ability to independently drive upper limb movement.

[0052] Example 2

[0053] This embodiment is an improvement and supplement to embodiment 1, and repeated content will not be repeated.

[0054] like Figure 1 The illustrated upper limb function reconstruction system for stroke patients includes a signal acquisition unit 100, a signal processing unit 200, and an upper limb training unit 300. The signal processing unit 200 determines the stroke patient's motor imagery task based on the electroencephalogram (EEG) signals acquired by the signal acquisition unit 100 and converts the task into motor control signals, which are then sent to the upper limb training unit 300. The upper limb training unit 300 assists the stroke patient in upper limb movement based on the motor control signals sent by the signal processing unit 200. During upper limb movement, the signal acquisition unit 100 also acquires electromyographic (EMG) signals generated by the upper limb movement. The signal processing unit 200 continuously modifies the motor imagery task determined by the EEG signals based on the actual movement trajectory generated by the EMG signals, aiming to achieve upper limb movement that better matches the brain's motor imagery task in the next assisted movement. This gradually restores the brain's control connection to the upper limbs and enables the stroke patient to gradually regain the ability to autonomously control upper limb movement.

[0055] When the signal processing unit 200 maps the EEG signal into a motor imagery task and a motor control signal, there may be some error. Although the upper limb can still move based on the EEG signal even with this error, its accuracy needs to be further improved. Therefore, this application designs the following technical solution.

[0056] Preferably, the signal processing unit 200 can determine the accuracy of the motor control signals analyzed from the EEG signals of stroke patients based on a pre-set EEG-EMG mapping model, and continuously correct the relationship between the EEG signals and the motor control signals based on the EEG-EMG mapping model, so that the actual movement trajectory under the assisted movement of the upper limb training unit 300 is consistent with the motor imagination task in the brain of stroke patients, thereby enhancing the accuracy of the connection between EEG signals and upper limb movement. Furthermore, accurate and effective training helps to increase the possibility of stroke patients regaining the ability to independently drive upper limb movement.

[0057] Preferably, the construction of the EEG / EMG mapping model can employ a research method based on variational mode decomposition-transfer entropy-based EEG / EMG coupling, specifically involving the following steps:

[0058] Step 1: Select several normal subjects to complete the experimental operation according to the experimental instructions. During the experiment, a large number of EEG and EMG signals are collected. The EEG and EMG signals are collected synchronously.

[0059] Step 2: Preprocessing of the above EEG and EMG signal data;

[0060] Step 3: Perform variational mode decomposition on the preprocessed EEG and EMG signals to decompose them into several different intrinsic mode functions;

[0061] Step 4: Calculate the transfer entropy between each pair of EEG and EMG signals obtained by variational mode decomposition, and construct an EEG-EMG mapping model based on the calculation results;

[0062] Step 5: Select different time scales to calculate the transfer entropy of intrinsic mode functions with different properties, components, and directions again. This can reduce the interference caused by frequency band aliasing to the subsequent calculation of transfer entropy between waves of different frequencies, and improve the accuracy of the EEG / EMG mapping model.

[0063] Preferably, based on the EEG-EMG mapping model, the signal processing unit 200 can calculate the second EMG signal based on the EEG signal. That is, the second EMG signal mainly refers to the EMG signal obtained by the signal processing unit 200 through the EEG-EMG mapping model. The first EMG signal obtained by the myoelectronic unit 120 of the signal acquisition unit 100 refers to the actual motor EMG signal after upper limb movement based on the motor imagery task, while the second EMG signal is the target motor EMG signal after upper limb movement based on the expected motor imagery task mapped from the EEG signal. In practice, the second EMG signal is determined by a model that has been repeatedly trained with a large amount of data. Its accuracy is higher than that of the model that converts the EEG signal of the stroke patient into a motor imagery task by the signal processing unit 200 while training and learning. Therefore, it may lead to a difference between the first EMG signal and the second EMG signal. This application uses the difference between the first EMG signal and the second EMG signal as a correction value and determines the corrected motor imagery task based on the correction value. In the existing technology, the research methods for deriving motor tasks from electromyographic signals or vice versa are relatively mature, and will not be elaborated here.

[0064] Preferably, the signal processing unit 200 is at least able to determine the motor imagery task of the stroke patient based on the electroencephalogram (EEG) signal. Specifically, the signal processing unit 200 decodes the EEG signal and generates the corresponding patient attention level. The signal processing unit 200 obtains the corresponding motor imagery task based on the attention level and establishes a mapping relationship between the motor imagery task and the patient's EEG signal. Furthermore, the signal processing unit 200 constructs the motor control signal for the upper limb training unit 300 based on the motor imagery task.

[0065] Preferably, the step of the signal processing unit 200 establishing the mapping relationship between EEG signals and the motor imagery task includes:

[0066] Multiple sets of sample information were obtained from normal individuals performing upper limb movements. Each set of sample information included sample electroencephalogram (EEG) signals and sample electromyogram (EMG) signals.

[0067] A motion intent recognition model based on a dual-stream Transformer encoder and a multi-head attention mechanism is established based on the information of each set of samples.

[0068] The mapping relationship between EEG signals and motor imagery tasks is determined based on a motor intention recognition model.

[0069] Preferably, the process of establishing a motion intent recognition model based on a dual-stream Transformer encoder and a multi-head attention mechanism based on information from each set of samples mainly includes the following steps:

[0070] The collected sample EEG signals are preprocessed to obtain the first part of the dataset data required to build a motor intention recognition model;

[0071] The first part of the dataset is expanded to obtain the second part of the dataset. The first part of the dataset and the second part of the dataset are then merged to form the sample dataset.

[0072] Establish a motion intent recognition network based on a dual-stream Transformer encoder and a multi-head attention mechanism;

[0073] The motion intent recognition network includes a two-stream Transformer encoder, a long and short sequence feature cross-attention module, a multi-scale feature fusion module, and a motion intent classification module;

[0074] The sample training dataset is input into the motion intent recognition network for training and learning to obtain the motion intent recognition model.

[0075] Specifically, the upper limb function reconstruction system for stroke patients of this application is equipped with at least a playback device for playing various upper limb movements, such as common devices like mobile phones, televisions, and computers. This playback device is placed in a position that the stroke patient can observe, and corresponding upper limb movement movements, such as upper limb extension and fist clenching, are played continuously. When the upper limb movement video is played, the stroke patient watches the video content and tries to imagine completing the upper limb movement in their mind. At this time, the signal acquisition unit 100 can acquire the brain signals of the stroke patient, and the signal processing unit 200 decodes the brain signals to generate the corresponding attention level. Then, based on the attention level, the corresponding motor imagination task is obtained, and a mapping relationship between the motor imagination task and the patient's brain signals is established.

[0076] According to the above scheme, the signal processing unit 200 performs at least two signal processing procedures after acquiring the EEG signal. The first is to convert the EEG signal into a corresponding second electromyographic signal using a brain-myoelectric coupling research method based on variational mode decomposition-transfer entropy. The second is to use the trained motor intention recognition model to derive the mapping relationship between the EEG signal and the motor imagery task. After each upper limb movement training, the myoelectronic unit 120 of the signal acquisition unit 100 acquires a first electromyographic signal representing the actual upper limb movement. A correction value is derived based on the difference between the first and second electromyographic signals, and the mapping relationship between the EEG signal and the motor imagery task is corrected based on this correction value. Specifically, the corrected motor imagery task is derived through the correction value and the electromyographic signal, and the EEG signal of this movement and the corrected motor imagery task are used as a training set to train the motor intention recognition model again. After each subsequent movement, the signal processing unit 200 trains the motor intention recognition model in the above manner, thereby continuously improving the accuracy of the motor intention recognition model and further improving the mapping relationship between the EEG signal and the motor imagery task.

[0077] Preferably, the signal processing unit 200 determines the motion control signal for the motion imagery task based on the aforementioned trained motion intention recognition model, and transmits the motion control signal to the upper limb training unit 300. The upper limb training unit 300 performs motion based on the motion control signal, while simultaneously assisting the stroke patient's upper limb in motion.

[0078] Preferably, the signal processing unit 200 can at least acquire the attention level of the EEG signal to drive the corresponding upper limb training unit 300 to move. In other words, the state of the attention level of the EEG signal determines the state of the upper limb training unit 300. Specifically, it determines whether the attention level is in a concentrated state and whether the attention level is concentrated on the upper limb. Normally, when performing motor imagery tasks involving upper limb movement, stroke patients should mainly concentrate their attention on the upper limbs. In other cases, their concentration level is relatively low. Therefore, the start and stop state of the upper limb training unit 300 can be controlled according to the concentration location or concentration state of the attention level, avoiding the upper limb training unit 300 from operating incorrectly when the stroke patient does not want to perform upper limb movement.

[0079] Example 3

[0080] This embodiment is an improvement and supplement to Embodiments 1 and 2, and the repeated content will not be repeated.

[0081] like Figure 1The illustrated upper limb function reconstruction system for stroke patients includes a signal acquisition unit 100, a signal processing unit 200, and an upper limb training unit 300. The signal processing unit 200 determines the stroke patient's motor imagery task based on the electroencephalogram (EEG) signals acquired by the signal acquisition unit 100 and converts the task into motor control signals, which are then sent to the upper limb training unit 300. The upper limb training unit 300 assists the stroke patient in upper limb movement based on the motor control signals sent by the signal processing unit 200. During upper limb movement, the signal acquisition unit 100 also acquires electromyographic (EMG) signals generated by the upper limb movement. The signal processing unit 200 continuously modifies the motor imagery task determined by the EEG signals based on the actual movement trajectory generated by the EMG signals, aiming to achieve upper limb movement that better matches the brain's motor imagery task in the next assisted movement. This gradually restores the brain's control connection to the upper limbs and enables the stroke patient to gradually regain the ability to autonomously control upper limb movement.

[0082] Existing or common treatments or devices for assisting upper limb rehabilitation training in stroke patients typically involve externally applying control to complete the assisted movement of the upper limbs. For example, medical staff or caregivers manually pull the stroke patient's upper limbs to perform certain postures for rehabilitation training. Another example is using mechanical devices to raise and bend the stroke patient's upper limbs in fixed postures for rehabilitation training. However, the rehabilitation effects achieved by the above common methods are minimal. This is because the aforementioned movements are externally applied to the stroke patient's upper limbs. The main reason for the inability to move the upper limbs is that the connection between the upper limb muscle neurons and the brain neurons is blocked or blocked. Therefore, simply moving the upper limbs cannot significantly restore the control connection between the brain and the upper limbs. The technical solution of this application is based on simulating a motor imagery task using electroencephalogram (EEG) signals, and then outputting the motor control signal of the upper limb training unit 300 according to the motor imagery task. This makes the upper limb training movements applied based on brain EEG signals. This method helps to restore the connection control between the brain and the upper limbs, achieving significant training effects.

[0083] Preferably, the signal processing unit 200 converts the EEG signals acquired by the signal acquisition unit 100 into a motor imagery task, and outputs the motor control signal corresponding to the motor imagery task to the upper limb training unit 300. The upper limb training unit 300 drives the upper limb, configured as an upper limb exoskeleton, to move based on the motor control signal, thereby assisting the upper limb of the stroke patient to perform motor imagery tasks based on EEG signals.

[0084] Preferably, the upper limb training unit 300 is configured as an upper limb exoskeleton. The upper limb training unit 300 is at least divided into a left exoskeleton and a right exoskeleton. The configuration of the left and right exoskeletons is roughly the same, and the control method is also similar. In actual operation, the signal processing unit 200 of this application determines whether the stroke patient wants to move the left or right upper limb based on the patient's attention level, and then drives the corresponding left or right exoskeleton to perform assisted movement.

[0085] Preferably, Figure 4 A simplified structural diagram of the upper limb training unit 300 of the stroke patient upper limb function reconstruction system of this application is shown. The upper limb training unit 300 includes at least a shoulder bone 310, an upper arm bone 320, a forearm bone 330, and finger bones 340. The shoulder bone 310 and the upper arm bone 320 are connected in a manner that mimics the shoulder joint, the upper arm bone 320 and the forearm bone 330 are connected in a manner that mimics the elbow joint, and the forearm bone 330 and the finger bones 340 are connected in a manner that mimics the wrist joint. The upper limb training unit 300 thus constitutes an upper limb training unit 300 capable of multi-degree-of-freedom movement in three-dimensional space.

[0086] Preferably, the upper limb training unit 300 can first determine the coordinates of the starting point and the ending point in a Cartesian space constructed by referring to the space around the human body, wherein the origin of the Cartesian space can be set as the center of the line segment connecting the left and right shoulders. The starting point is selected as the initial position of the exoskeleton, that is, the upper limb training unit 300 is in a suspended state, and the initial position coordinates of the geometric center of each exoskeleton module are recorded. The ending point is the extreme position of each joint of the exoskeleton when flexed or internally rotated, and the extreme position coordinates of the geometric center of each exoskeleton module are recorded.

[0087] Preferably, for the movement state of the upper limb, the various movements of the entire upper limb can be broken down into the rotational movements of each joint. Specifically, no matter what state or form of normal movement the upper limb is in, the length and shape of its bones will not change. The flexibility of the upper limb mainly depends on the multi-angle flexible rotation of the joints connecting each bone. Therefore, when applied to the exoskeleton of the upper limb training unit 300, it is only necessary to control the rotation angle and rotation direction of the joint connection between each bone to achieve upper limb assisted movement.

[0088] Preferably, for the motor imagery task obtained by the signal processing unit 200 based on the EEG signal conversion and the aforementioned trained motor intention recognition model, the signal processing unit 200 can decompose the motor imagery task into rotational movements of each joint, specifically:

[0089] An exoskeleton kinematic model of the upper limb training unit 300 was constructed using an improved DH model.

[0090] The starting and ending coordinates are determined in Cartesian space. Then, the trajectory is smoothed using a cubic polynomial and the planned path is divided into multiple discrete points. The starting point is selected as the initial position of the exoskeleton, that is, the skeleton is in a suspended state. The ending point is the joints of the exoskeleton in the motion imagination task flexing or internally rotating to the target position.

[0091] After calculating the joint angles corresponding to each discrete point in the motion imagination task using robot inverse kinematics, differential operations are performed to obtain angular velocity and angular acceleration, thereby controlling the upper limb training unit 300 to move.

[0092] Specifically, the signal processing unit 200 is able to virtually model the upper limb training unit 300 using the Newton-Euler iterative method.

[0093] In summary, this application proposes an upper limb function reconstruction system for stroke patients, including a signal acquisition unit 100, a signal processing unit 200, and an upper limb training unit 300.

[0094] Preferably, the signal acquisition unit 100 includes EEG electrodes and an EEG amplifier. After the patient wears the EEG electrodes on their head, the signal acquisition unit 100 acquires the patient's EEG signals. The EEG amplifier is based on a differential amplifier circuit to acquire the patient's EEG signals by constructing a single-channel EEG acquisition method, and transmits the acquired EEG signals to the signal processing unit 200.

[0095] Preferably, the signal processing unit 200 processes, analyzes, identifies, transmits, and stores human electroencephalogram (EEG) signals, decodes the EEG signals, generates corresponding patient attention levels, and then transmits them to the training unit. The signal processing unit 200 obtains the patient's current attention level by calculating the power in the 8–12 Hz and 12–30 Hz frequency bands. The signal acquisition unit 100 and the signal processing unit 200 can also be used to monitor patient attention.

[0096] Preferably, the signal processing unit 200 can also obtain corresponding motor imagery tasks based on the attention level, such as arm swinging, fist clenching, etc., thereby establishing a mapping relationship between the EEG motor imagery tasks and converting them into motion trajectory instructions using an algorithm. The training and learning unit can predict the motor imagery tasks and transmit them to the upper limb training unit 300. The accuracy and stability of the prediction can be gradually improved through feedback from the upper limb training unit 300.

[0097] Preferably, the upper limb training unit 300 responds to the motion control signal of the signal processing unit 200 to complete the motion imagination task and feeds back the motion result of the motion imagination task to the signal processing unit 200.

[0098] Preferably, the upper limb training unit 300 mainly includes a shoulder skeleton 310, an upper arm skeleton 320, a forearm skeleton 330, and finger bones 340. In addition, the upper limb training unit 300 also includes a wearable part 350 for wearing on the body of a stroke patient. Specifically, the upper limb training unit 300 can be worn on the waist, back, or other parts of the stroke patient's body. Each exoskeleton module included in the upper limb training unit 300 internally houses the myoelectronic unit 120 from the aforementioned signal acquisition unit 100. This unit is attached to the skin surface of various muscle areas of the patient's upper limb in the form of patches to acquire dynamic data of the patient's limb coordinates and feed it back to the training and learning unit, thereby realizing and optimizing the reconstruction of upper limb function in stroke patients.

[0099] Based on the above scheme, the upper limb training unit 300 of the stroke patient upper limb function reconstruction system of this application can autonomously control the upper limb training unit 300 to assist the paralyzed upper limb in movement based on electroencephalogram (EEG) signals. There is no need for additional medical staff or caregivers to manually pull the stroke patient to move. Moreover, the movement control signal of the upper limb training unit 300 determined according to the EEG signals can increase the probability of restoring the control connection between the EEG signals and the paralyzed upper limb compared to the movement control applied manually by other people outside the patient or simply mechanically according to a few preset postures. This avoids ineffective or inefficient repetitive and mechanical upper limb rehabilitation exercise training.

[0100] Example 4

[0101] This embodiment is an improvement and supplement to any of the foregoing embodiments or combinations of embodiments, and repeated content will not be described again.

[0102] This embodiment provides a more convenient and simpler preferred implementation of the upper limb function reconstruction system for stroke patients based on the aforementioned technical solution. Specifically, the upper limb function reconstruction system for stroke patients of this application provides the user with a target movement posture through devices such as VR devices, televisions, and mobile phones that can display relevant images and video information. The user looks at the target movement posture and then imagines how their upper limbs move according to the displayed image information. During the imagination process, the patient's attention level will be greatly improved compared to the normal attention level. When the attention level reaches a certain threshold, the upper limb training unit 300 controls the exoskeleton to drive the upper limbs to train according to the target movement posture displayed in the image.

[0103] Preferably, in the above scheme, the acquisition module 100 only needs to acquire EEG signals in a single-channel manner through the brain electronics unit 110, and the signal processing unit 200 extracts attention level features based on the single-channel EEG signals and obtains the attention level value. Then, it compares the current attention level value with the preset attention level value. If the current attention level value is greater than or equal to the preset attention level value, the signal processing unit 200 sends the motion information parameters of the target movement posture to the upper limb training unit 300. Then, the upper limb training unit 300 performs assisted movements on the user's upper limbs based on the motion information parameters.

[0104] Specifically, the brain electronics unit 110 of this application preferably adopts a single-channel EEG signal acquisition method, according to Figure 3 As shown, the brain electronics unit 110 includes at least a measuring electrode 111 and a reference electrode 112. The measuring electrode 111 is used to acquire the potential caused by the current in the neurons of the human brain when the brain waves vibrate. The initial potential of the reference electrode 112 is set to zero. The difference between the potential of the measuring electrode 111 and the reference potential represents the potential value of the measuring electrode 111.

[0105] Preferably, after the measuring electrodes 111 in the brain electronics unit 110 acquire the EEG signal, it is first preprocessed by the preprocessing module installed in the brain electronics unit 110. The preprocessing includes filtering, amplification, and A / D conversion of the acquired EEG signal. Specifically, the preprocessing module in the brain electronics unit 110 can use the TGAM module installed in the MindWave mobile EEG acquisition headset launched by NeuroSky. The preprocessing module transmits the acquired EEG signal to the signal processing unit 200 for analysis, so that the signal processing unit 200 can acquire the motor imagery task of the stroke patient.

[0106] Preferably, the measuring electrodes 111 of the brain electronics unit 110 are mainly distributed in the forehead of the user's brain, and sufficient EEG signals for analyzing attention level can be obtained through a single measuring electrode 111.

[0107] The advantage of a preferred embodiment of the upper limb function reconstruction system in this example is that it employs an EEG-based attention detection method, enabling the reconstruction of upper limb motor function without external stimulation. Furthermore, this method, through machine learning, can progressively improve the accuracy and stability of predictions, thus providing an effective means for upper limb function reconstruction after stroke. First, a single-channel EEG acquisition system is built based on a differential amplifier circuit to acquire EEG signals. Then, the power in the 8-12Hz and 12-30Hz frequency bands is calculated, along with the patient's current attention level. When a video of hand movements is played, the patient watches it and makes an effort to complete the movement. The exoskeleton device is then controlled based on the perceived attention level to perform the corresponding action.

[0108] It should be noted that the specific embodiments described above are exemplary, and those skilled in the art can devise various solutions inspired by the disclosure of this invention. These solutions all fall within the scope of this invention and its protection. Those skilled in the art should understand that this specification and its accompanying drawings are illustrative and not intended to limit the scope of the claims. The scope of protection of this invention is defined by the claims and their equivalents.

Claims

1. A system for reconstructing upper limb function in stroke patients, comprising at least: The signal acquisition unit (100) is used to acquire the physiological electrical signals of stroke patients; Signal processing unit (200) is used to process the physiological electrical signals; The upper limb training unit (300) performs upper limb assisted movements based on the control of the signal processing unit (200); Its features are, The signal acquisition unit (100) is capable of acquiring at least the electroencephalogram (EEG) signal of the stroke patient and the first electromyographic (EMG) signal generated by the upper limb under the assisted movement of the upper limb training unit (300). The signal acquisition unit (100) includes at least a brain electronic unit (110) for acquiring EEG signals and a muscle electronic unit (120) for acquiring EMG signals. A plurality of EMG signal sensors of the muscle electronic unit (120) are disposed between the skin surface of the upper limb of the stroke patient and the upper limb training unit (300) and are held by the upper limb training unit (300). The signal processing unit (200) obtains the mapping relationship between EEG signals and motor imagery tasks based on a continuously trained motor intention recognition model constructed based on a dual-stream Transformer encoder and a multi-head attention mechanism. The signal processing unit (200) can substitute the EEG signals into an EEG-EMG mapping model constructed based on variational mode decomposition and transfer entropy calculation to obtain a second EMG signal. The first EMG signal refers to the actual motor EMG signal after upper limb movement based on the motor imagery task. The second EMG signal is the target motor EMG signal after upper limb movement based on the expected motor imagery task mapped from the EEG signals. The signal processing unit (200) continuously corrects the mapping relationship between EEG signals and motor imagery tasks based on the difference between the first EMG signal and the second EMG signal. The signal processing unit (200) is at least able to analyze the patient's attention level based on the EEG signals acquired by the brain electronics unit (110) in a single-channel form, and the signal processing unit (200) controls the upper limb training unit (300) to perform assisted movements based on the attention level; the signal processing unit (200) is at least able to substitute the EEG signals acquired by the brain electronics unit (110) in a multi-channel form into the EEG-EMG mapping model to obtain a second EMG signal; Each time the upper limb training unit (300) performs an upper limb assisted movement, the signal processing unit (200) uses the generated EEG signal and the corrected motor imagery task as training data to train the motor intention recognition model.

2. The upper limb function reconstruction system for stroke patients according to claim 1, characterized in that, The process by which the signal processing unit (200) constructs the motion intention recognition model is as follows: Multiple sets of sample information were obtained from normal individuals performing upper limb movements. Each set of sample information included sample electroencephalogram (EEG) signals and sample electromyogram (EMG) signals. A motion intent recognition model based on a dual-stream Transformer encoder and a multi-head attention mechanism is established based on the information of each set of samples. The mapping relationship between EEG signals and motor imagery tasks is determined based on a motor intention recognition model.

3. The upper limb function reconstruction system for stroke patients according to claim 2, characterized in that, The process by which the signal processing unit (200) constructs the electroencephalogram and electromyogram mapping model is as follows: Several normal subjects were selected to perform voluntary upper limb movements, and electroencephalogram (EEG) and electromyogram (EMG) signals were collected. The EEG and EMG signals were collected simultaneously. Variational mode decomposition was performed on the preprocessed EEG and EMG signals to decompose them into several different intrinsic mode functions; The transfer entropy between each pair of EEG and EMG signals obtained by variational mode decomposition is calculated, and an EEG-EMG mapping model is constructed based on the calculation results.

4. The upper limb function reconstruction system for stroke patients according to claim 1, characterized in that, The signal processing unit (200) derives a motor imagery task based on the electroencephalogram (EEG) signal and calculates a corrected motor imagery task based on the difference between the first electromyogram (EMG) signal and the second EMG signal.

5. The upper limb function reconstruction system for stroke patients according to claim 1, characterized in that, Both the brain electronics unit (110) and the muscle electronics unit (120) maintain a data connection with the signal processing unit (200).

6. The upper limb function reconstruction system for stroke patients according to claim 5, characterized in that, The electromyography unit (120) is disposed on the skin surface of the upper limb of a stroke patient. It acquires the first electromyographic signal through a multi-channel method and includes at least an electromyographic signal sensor capable of acquiring electromyographic signals of the upper arm, forearm and / or palm of the upper limb respectively.

7. The upper limb function reconstruction system for stroke patients according to any one of claims 1 to 6, characterized in that, The stroke patient upper limb function reconstruction system can be applied to upper limb rehabilitation training for stroke patients in individual families.

Images