A brain-computer interface-based mechanical exoskeleton rehabilitation training system and method

Abstract

The application relates to a brain-computer interface-based mechanical exoskeleton rehabilitation training system and method, which comprises an electroencephalogram acquisition device, a preprocessing unit, a feature extraction unit, a motion intention decoding unit, a mechanical exoskeleton control unit, a muscle tension state monitoring unit and an adaptive functional electrical stimulation feedback unit. The method is used for preprocessing electroencephalogram and electromyogram signals; then electroencephalogram and electromyogram time-frequency features are obtained through feature extraction; a patient motion intention is obtained through decoding by using an electroencephalogram decoding model, an exoskeleton control instruction is generated, a mechanical exoskeleton control unit is controlled to drive the limb movement of a rehabilitation patient to perform rehabilitation training; meanwhile, the muscle tension state of the patient is analyzed according to the time-frequency features of the electroencephalogram and electromyogram, different intensity functional electrical stimulation is adaptively applied, stimulation feedback is strengthened, and the muscle tension state of the patient is adjusted. The application can realize the deep fusion of brain-controlled exoskeleton training and low-frequency neuroelectric stimulation adjustment, and improve the rehabilitation effect of a stroke patient.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation training equipment technology, and in particular to a mechanical exoskeleton rehabilitation training system and method based on a brain-computer interface. Background Technology

[0002] Stroke and other brain diseases can cause limb motor dysfunction. Patients often experience varying degrees of loss of limb motor function after a stroke, impacting their daily activities and quality of life. Stroke rehabilitation primarily relies on physical therapy, rehabilitation training, and the use of assistive devices. For example, stroke patients often experience varying degrees of lower limb motor function loss. Lower limb rehabilitation after a stroke mainly depends on physical therapy, rehabilitation training, and the use of assistive devices, aiming to help patients restore normal gait and lower limb function.

[0003] In recent years, with the development of neuroscience and rehabilitation engineering, brain-computer interface (BCI) technology has gradually become an important tool in lower limb rehabilitation training. A BCI is a technology that collects and decodes electroencephalogram (EEG) signals, converting brain activity into control commands for external devices, enabling direct communication between the human brain and external devices. Motor imagery (MI), as an important application of BCI technology, utilizes the brain activity characteristics of patients when imagining specific limb movements to achieve real-time control of rehabilitation equipment, helping to enhance patients' active participation and promote neural remodeling. However, current lower limb rehabilitation training systems based on BCI technology still have the following shortcomings:

[0004] (1) Most existing lower limb rehabilitation equipment is mainly mechanical exoskeleton, which lacks a linkage mechanism with brain activity, cannot achieve real-time control based on the patient's active motor imagination, and is difficult to fully activate the brain's motor cortex, thus limiting the effect of nerve function recovery.

[0005] (2) Existing EEG data analysis methods are mostly focused on simple time-domain or frequency-domain feature analysis, ignoring the functional connections and dynamic information flow relationships between different brain regions. They cannot fully reflect the state and changes of the patient's brain activity, thus limiting the evaluation and optimization of rehabilitation training effects.

[0006] (3) In practical applications, traditional brain-computer interface systems typically require the collection of a large amount of motor imagery EEG data for each patient individually in order to complete the training of a personalized classifier. This process not only places high demands on the patient's attention span, which can easily lead to fatigue, but may also affect the continuity and overall effectiveness of rehabilitation training.

[0007] (4) Stroke patients often experience varying degrees of abnormal muscle tone, manifested as muscle stiffness, spasticity, or bradykinesia. Abnormal muscle tone affects the voluntary movement ability of the limbs, interferes with the patient's normal response to external stimuli during rehabilitation training, and reduces the quality and effectiveness of training.

[0008] Therefore, to address the aforementioned issues, there is an urgent need for a rehabilitation training system that integrates brain-computer interface technology, exoskeleton rehabilitation training, and muscle tone monitoring and relief. This system should extract motor imagery features through high-precision EEG signal acquisition and processing, drive the lower limb exoskeleton device in real time, dynamically monitor and provide feedback on the patient's muscle tone status, and utilize functional electrical stimulation (fEP) technology to alleviate abnormal muscle tone, thereby enhancing the stimulation feedback effect of motor rehabilitation and providing effective technical support and good rehabilitation outcomes for lower limb motor rehabilitation in stroke patients. Summary of the Invention

[0009] The purpose of this invention is to provide a brain-computer interface-based mechanical exoskeleton rehabilitation training system and method to assist patients in using mechanical exoskeletons to achieve rehabilitation training, realize real-time control of rehabilitation equipment, help enhance patients' active participation, and promote neural remodeling.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] According to one aspect of the present invention, a brain-computer interface-based mechanical exoskeleton rehabilitation training system is provided, comprising:

[0012] The electroencephalogram (EEG) acquisition device is configured to collect EEG signals of cerebral cortex activity in rehabilitation patients during motor imagery.

[0013] An electromyography (EMG) acquisition device is configured to acquire EMG signals from rehabilitation patients during motor imagery.

[0014] The preprocessing unit is configured to preprocess the acquired electroencephalogram (EEG) and electromyogram (EMG) signals.

[0015] The feature extraction unit is configured to extract features from the preprocessed EEG and EMG signals to obtain the time-frequency features of EEG and EMG;

[0016] The motion intention decoding unit is configured to decode the patient's motion intention based on the EEG time-frequency characteristics using an EEG decoding model, and generate mechanical exoskeleton control commands based on the motion intention.

[0017] The mechanical exoskeleton control unit is configured to drive the limb movements of the rehabilitation patient for rehabilitation training according to the mechanical exoskeleton control commands;

[0018] The muscle tone monitoring unit is configured to analyze the patient's muscle tone status based on the time-frequency characteristics of EEG and EMG.

[0019] The adaptive functional electrical stimulation feedback unit is configured to adaptively apply functional electrical stimulation of different intensities according to the patient's muscle tone state during rehabilitation training, thereby enhancing stimulation feedback and regulating the patient's muscle tone state.

[0020] According to one embodiment of the present invention, the system further includes an evaluation unit, which monitors the time-frequency characteristics of the electroencephalogram (EEG) of the rehabilitation patient in real time, generates a brain topography map, a functional brain network and a time-frequency characteristic map based on the EEG time-frequency characteristics, and evaluates the degree of improvement of the rehabilitation patient's motor ability by combining feedback data obtained from the mechanical exoskeleton, wherein the feedback data includes gait, joint angles and movement speed.

[0021] According to one embodiment of the present invention, the EEG time-frequency features and EMG time-frequency features are obtained by extracting EEG signals based on wavelet transform; the brain topography and functional brain network are obtained by phase-locked value analysis of the functional brain network, which are used to evaluate the information transmission between different regions of the brain.

[0022] According to one embodiment of the present invention, the system further includes an EEG display unit for real-time or offline display of brain topography maps, functional brain networks, and time-frequency feature maps generated by the evaluation unit.

[0023] According to one embodiment of the present invention, the preprocessing unit performs preprocessing on the acquired electroencephalogram (EEG) signals, including noise reduction, filtering, and artifact removal.

[0024] According to one embodiment of the present invention, the motor intention decoding unit is a personalized EEG signal decoding model adapted to the target subject, obtained through transfer learning fine-tuning; wherein: by preprocessing the resting data of the source domain, the motor imagery data of the source domain, and the resting data of the target domain, a preprocessed EEG signal is obtained; the preprocessed EEG signal is used to pre-train a GAN generator network including a generator and a discriminator to obtain a pre-trained generator; the preprocessed source domain motor imagery data is used to train a transfer learning network to obtain a pre-trained transfer learning model; using the target resting data, the resting data of the source domain, and the motor imagery data of the source domain after preprocessing and data fusion, the pre-trained generator generates target motor imagery features; the generated target motor imagery features are used to retrain the pre-trained transfer learning model through transfer learning fine-tuning to obtain a personalized EEG signal decoding model adapted to the target subject.

[0025] According to one embodiment of the present invention, the mechanical exoskeleton control unit processes the patient's real-time motor imagery EEG signals based on a personalized decoding model, decodes the patient's motor intentions, and converts them into exoskeleton drive signals, thereby controlling the mechanical exoskeleton to drive the patient's limb movements.

[0026] According to one embodiment of the present invention, the muscle tone state monitoring unit utilizes the time-frequency characteristics of electroencephalogram (EEG) signals and electromyogram (EMG) signals, and uses coherence analysis to analyze the patient's current muscle tone state.

[0027] According to one embodiment of the present invention, the adaptive functional electrical stimulation feedback unit applies functional electrical stimulation of different intensities during brain-controlled exoskeleton rehabilitation training based on muscle tone monitoring results, thereby alleviating the abnormal muscle tone state of the patient's limbs and enhancing the stimulation feedback effect of motor rehabilitation.

[0028] On the other hand, the present invention also provides a brain-computer interface-based mechanical exoskeleton rehabilitation training method, comprising the following steps:

[0029] Collect electroencephalogram (EEG) and electromyogram (EMG) signals of cerebral cortex activity in rehabilitation patients during motor imagery.

[0030] The collected electroencephalogram (EEG) and electromyogram (EMG) signals were preprocessed;

[0031] Feature extraction was performed on the preprocessed EEG and EMG signals to obtain their time-frequency features;

[0032] Based on the time-frequency characteristics of EEG, the patient's motor intention is obtained by decoding using an EEG decoding model, and control commands for the mechanical exoskeleton are generated based on the motor intention.

[0033] Using control commands to control a mechanical exoskeleton to drive limb movements in rehabilitation patients for rehabilitation training;

[0034] During rehabilitation training, the patient's muscle tone status is analyzed using coherence analysis based on the time-frequency characteristics of electroencephalography (EEG) and electromyography (EMG). Based on the patient's muscle tone status, functional electrical stimulation of different intensities is applied adaptively to alleviate the patient's abnormal muscle tone and enhance the stimulation feedback effect of rehabilitation training.

[0035] After the rehabilitation training process is completed, based on the EEG time-frequency characteristics of the rehabilitation training process, brain topography, brain network diagram and time-frequency characteristic diagram are used in combination with exoskeleton feedback data to assess the degree of improvement in the rehabilitation patient's motor ability.

[0036] Beneficial effects:

[0037] This invention proposes a brain-computer interface-based mechanical exoskeleton rehabilitation training system and method, which has the following advantages compared with the prior art:

[0038] 1. Combining brain-computer interface (BCI) and lower limb exoskeleton technology enhances patient participation in rehabilitation. BCI technology acquires real-time EEG signals during patient motor imagery and decodes the corresponding motor intentions, thereby driving the lower limb exoskeleton to perform the corresponding movements. This control method closely integrates the patient's subjective intentions with the rehabilitation training process, greatly enhancing the patient's sense of active participation and autonomy, which is conducive to the remodeling of neural function and the recovery of motor abilities. Compared with traditional passive rehabilitation training, the active participation mode of this invention can effectively improve rehabilitation efficiency, reduce dependence during training, and enhance the patient's confidence and willingness to rehabilitate.

[0039] 2. Using adversarial generative network (EGN) data generation technology and transfer learning technology, a personalized EEG signal decoding model is rapidly constructed. A large amount of motor imagery EEG data features of the target subjects are generated using EGN data generation technology based on resting data. Combined with transfer learning technology, a personalized EEG signal decoding model for the target subjects is rapidly constructed, reducing the time required for EEG data collection and alleviating the physiological and psychological stress on the subjects.

[0040] 3. Combining real-time muscle tone monitoring technology with adaptive functional electrical stimulation feedback technology alleviates abnormal muscle tone in patients and enhances motor rehabilitation stimulation feedback. Through coherence analysis of EEG and EMG characteristics, the patient's muscle tone is monitored in real time. Based on the patient's muscle tone status, functional electrical stimulation of varying intensities is applied to alleviate abnormal muscle tone, improve rehabilitation training effectiveness, and optimize neurofeedback regulation.

[0041] 4. Using EEG signal analysis techniques such as brain topography, brain network mapping, and time-frequency feature mapping, along with feedback data from the mechanical exoskeleton, the rehabilitation effect of patients can be evaluated from multiple perspectives. Using EEG data collected during rehabilitation training, after preprocessing and feature extraction, the features are plotted as brain topography, brain network mapping, and time-frequency feature maps. This allows for multi-faceted analysis of changes in EEG signals during the patient's rehabilitation process, thus assessing the patient's motor rehabilitation effect. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0043] Figure 1 This is a schematic diagram of a brain-computer interface-based mechanical exoskeleton rehabilitation training system according to an embodiment of the present invention.

[0044] Figure 2 This is a flowchart of a lower limb mechanical exoskeleton rehabilitation training method based on a brain-computer interface according to an embodiment of the present invention.

[0045] Figure 3This is a schematic diagram of the lower limb mechanical exoskeleton rehabilitation training and assessment process based on a brain-computer interface according to an embodiment of the present invention.

[0046] Figure 4 This is a schematic diagram of a pre-collection experiment of multiple motor imagery EEGs according to an embodiment of the present invention.

[0047] Figure 5 This is a schematic diagram of a generative adversarial network module according to an embodiment of the present invention.

[0048] Figure 6 This is a schematic diagram of the BAdaptNet transfer learning network according to an embodiment of the present invention.

[0049] Figure 7 This is a schematic diagram of the BAdaptNet network undergoing transfer learning training according to an embodiment of the present invention.

[0050] Figure 8 This is a schematic diagram of EEG and EMG characteristic analysis during rehabilitation training according to an embodiment of the present invention.

[0051] Figure 9 This is a flowchart illustrating the testing phase before formal training for patients, according to an embodiment of the present invention.

[0052] Figure 10 This is a schematic diagram of data acquisition for a lower limb rehabilitation training experiment according to an embodiment of the present invention. Detailed Implementation

[0053] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0054] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0055] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0056] This invention relates to a brain-computer interface-based mechanical exoskeleton rehabilitation training system and method. It acquires the patient's electroencephalogram (EEG) signals during motor imagery in real time via BCG technology and decodes the corresponding motor intentions, thereby driving the mechanical exoskeleton to complete the corresponding movements. This control method closely integrates the patient's subjective intentions with the rehabilitation training process, greatly enhancing the patient's sense of active participation and autonomy, and contributing to the remodeling of neural function and the recovery of motor abilities. Using resting-state data of the target subjects, adversarial generative network (EGN) data generation technology is employed to generate a large amount of EEG data features for the target subjects' motor imagery. Combined with transfer learning technology, a personalized EEG signal decoding model for the target subjects is rapidly constructed, reducing the time required for EEG data acquisition and lowering the physiological and psychological stress on the subjects. Through coherence analysis of EEG and electromyographic (EMG) features, the patient's muscle tone state is monitored in real time. Based on the patient's muscle tone state, functional electrical stimulation of varying intensities is applied to alleviate abnormal muscle tone, improve the rehabilitation training effect, and optimize neural feedback regulation. Using EEG data collected during rehabilitation training, after preprocessing and feature extraction, the features are plotted as brain topography, brain network diagram and time-frequency feature map, and then the changes of EEG signals during the patient's rehabilitation process are analyzed from multiple perspectives to evaluate the patient's motor rehabilitation effect.

[0057] Lower limb mechanical exoskeletons assist patients in standing and walking. Lower limb mechanical exoskeletons are classified according to different dynamic joints, namely single hip drive, single knee drive, single ankle drive, hip-knee drive, knee-ankle drive, and hip-knee-ankle drive. The dynamic joints are mainly the hip, knee, and ankle, enabling the hip joint to achieve flexion and extension, the knee joint to achieve flexion and extension, and the ankle joint to achieve dorsiflexion and plantarflexion.

[0058] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0059] like Figure 1 The diagram shows a schematic of a brain-computer interface-based mechanical exoskeleton rehabilitation training system. The brain-computer interface-based mechanical exoskeleton rehabilitation training system includes:

[0060] The electroencephalogram (EEG) acquisition device is configured to collect EEG signals of cerebral cortex activity in rehabilitation patients during motor imagery.

[0061] An electromyography (EMG) acquisition device is configured to acquire EMG signals from rehabilitation patients during motor imagery.

[0062] The preprocessing unit is configured to preprocess the acquired electroencephalogram (EEG) and electromyogram (EMG) signals.

[0063] The feature extraction unit is configured to extract features from the preprocessed EEG and EMG signals to obtain the time-frequency features of EEG and EMG;

[0064] The motion intention decoding unit is configured to decode the patient's motion intention based on the EEG time-frequency characteristics using an EEG decoding model, and generate mechanical exoskeleton control commands based on the motion intention.

[0065] The mechanical exoskeleton control unit is configured to drive the limb movements of the rehabilitation patient for rehabilitation training according to the mechanical exoskeleton control commands;

[0066] The muscle tone monitoring unit is configured to analyze the patient's muscle tone status based on the time-frequency characteristics of EEG and EMG.

[0067] The adaptive functional electrical stimulation feedback unit is configured to adaptively apply functional electrical stimulation of different intensities according to the patient's muscle tone state during rehabilitation training, thereby enhancing stimulation feedback and regulating the patient's muscle tone state.

[0068] After the rehabilitation training process is completed, based on the EEG time-frequency characteristics of the rehabilitation training process, brain topography, brain network diagram and time-frequency characteristic diagram are used in combination with exoskeleton feedback data to assess the degree of improvement in the rehabilitation patient's motor ability.

[0069] The brain topography and brain network data were obtained using phase-locked value analysis (PLA) to assess information transmission between different brain regions. The feedback data included gait, joint angles, and movement speed.

[0070] Furthermore, the system also includes an EEG display unit for real-time or offline display of the brain topography map, functional brain network, and time-frequency feature map generated by the evaluation unit.

[0071] The preprocessing unit preprocesses the acquired EEG signals, including noise reduction, filtering, and artifact removal.

[0072] The motor intention decoding unit is a personalized EEG signal decoding model adapted to the target subject, fine-tuned through transfer learning. Specifically: Preprocessing of resting-state data from the source domain, motor imagery data from the source domain, and resting-state data from the target domain yields preprocessed EEG signals; the preprocessed EEG signals are used to pre-train an RM-GAN network (including a generator and discriminator) to obtain a pre-trained generator; the preprocessed source domain motor imagery data is used to train a transfer learning network to obtain a pre-trained BAdaptNet network model; the pre-trained generator generates target motor imagery features using the feature data from the target resting-state data, source domain resting-state data, and source domain motor imagery data after preprocessing and data fusion; the generated target motor imagery features are used to retrain the pre-trained BAdaptNet network model through transfer learning, resulting in a personalized EEG signal decoding model adapted to the target subject through fine-tuning via transfer learning.

[0073] The resting-state data in the source domain refers to resting-state data from publicly available motor imagery EEG databases and previously collected motor imagery EEG datasets from the target subject. The resting-state data in the target domain refers to resting-state data collected when the target subject uses the method. The target subject typically refers to the current user or patient using the method.

[0074] The RM-GAN network, short for Rest and Motor Imagery Generative Adversarial Network, comprises a discriminator D and a generator G. The discriminator D uses a multi-layer convolutional network to extract complex distribution information of motor imagery features, thereby determining the authenticity of the input data. The generator G employs an encoder-decoder structure, combining convolutional and deconvolutional layers to achieve nonlinear mapping of the input feature matrix, progressively generating data that approximates the target motor imagery features. The generator's encoding module consists of four two-dimensional convolutional layers, effectively extracting the time-frequency features of EEG signals. A regularization layer is added after each convolutional layer for stable training, and LeakyReLU is used as the activation function to enhance nonlinear expressive power. The encoded features are progressively recovered through four deconvolutional layers in the decoding module, matching the target motor imagery features. Regularization layers and ReLU activation functions are added after the first three deconvolutional layers to ensure decoding stability and nonlinear mapping capability; the last deconvolutional layer uses a Tanh activation function to map the output features to the target feature space.

[0075] The BAdaptNet network, short for Brain Adapt Net, begins by extracting preliminary features through standard convolutional layers and adaptively adjusting channel importance using SE attention. Subsequently, pooling and Dropout are used for dimensionality reduction and regularization, reducing redundant information and preventing overfitting. Subsequent dynamic convolutional layers adaptively extract multi-scale features, improving the model's adaptability to different data patterns. In the deep feature extraction stage, the network progressively extracts high-dimensional features through multiple convolutional layers, using batch normalization and ReLU activation functions to ensure training stability. Finally, fully connected layers integrate the features, and Softmax is used to output class probabilities, achieving a binary classification task.

[0076] The mechanical exoskeleton control unit processes the patient's real-time motor imagery EEG signals based on a personalized decoding model, decodes the patient's motor intentions, and converts them into exoskeleton drive signals, thereby controlling the mechanical exoskeleton to drive the patient's limb movements.

[0077] The muscle tone monitoring unit analyzes the patient's current muscle tone state using the time-frequency characteristics of electroencephalogram (EEG) and electromyogram (EMG) signals. The adaptive functional electrical stimulation feedback unit applies functional electrical stimulation of varying intensities during brain-controlled exoskeleton rehabilitation training based on the muscle tone monitoring results, alleviating abnormal muscle tone in the patient's limbs and enhancing the stimulation feedback effect of motor rehabilitation.

[0078] like Figure 2 As shown, a brain-computer interface-based mechanical exoskeleton rehabilitation training method is presented, including the following steps:

[0079] Collect electroencephalogram (EEG) signals of cerebral cortical activity in rehabilitation patients during motor imagery.

[0080] The collected EEG signals are preprocessed;

[0081] Feature extraction was performed on the preprocessed EEG signals to obtain EEG time-frequency features;

[0082] Based on the time-frequency characteristics of EEG, the motor intention is obtained by decoding using an EEG decoding model, and control commands are generated based on the patient's motor intention.

[0083] Using control commands to control a mechanical exoskeleton to drive limb movements in rehabilitation patients for rehabilitation training;

[0084] Based on EEG and EMG characteristics, coherence analysis was used to analyze the patient's muscle tone status.

[0085] During rehabilitation training, functional electrical stimulation of different intensities is applied adaptively according to the patient's muscle tone state to alleviate the patient's abnormal muscle tone state and enhance the stimulation feedback effect of rehabilitation training.

[0086] After the rehabilitation training process is completed, based on the EEG time-frequency characteristics of the rehabilitation training process, brain topography, brain network diagram and time-frequency characteristic diagram are used in combination with exoskeleton feedback data to assess the degree of improvement in the rehabilitation patient's motor ability.

[0087] Furthermore, the method also includes setting a training paradigm, which comprises a combination of steps including resting data acquisition, reading motion prompts, and performing motor imagery according to the prompts. By setting a personalized motor imagery training paradigm for rehabilitation patients, it helps patients focus and interact. The training paradigm is a personalized lower limb rehabilitation training paradigm designed according to the individual needs of the patient. This paradigm includes resting state, motor imagery tasks, etc., to ensure that appropriate training content is customized for each patient's rehabilitation needs. Preferably, the paradigm design focuses on stepping with both legs, using upper-computer rehabilitation auxiliary software to design a personalized data acquisition and rehabilitation training paradigm that meets the patient's needs and objective conditions, thereby improving the patient's immersion and focus.

[0088] During training, EEG acquisition devices are used to record the patient's EEG signals in real time, especially during motor imagery tasks, to capture the patient's motor intentions. After processing the EEG data, a motor intention decoding algorithm using a decoding model is employed to analyze the EEG signals and decode the patient's motor imagery intentions. Based on these decoding results, the system generates control signals for controlling the lower limb mechanical exoskeleton, thereby assisting the patient in lower limb motor training. The mechanical exoskeleton dynamically adjusts its movement patterns according to the decoded signals to assist the patient in gait training and other lower limb motor training. During training, based on EEG and electromyography characteristics, coherence analysis is used to analyze the patient's muscle tone state. Depending on the patient's muscle tone state, functional electrical stimulation of varying intensities is adaptively applied to alleviate abnormal muscle tone and enhance the stimulus feedback effect of rehabilitation training. After the rehabilitation training process is completed, based on the time-frequency characteristics of the EEG during the rehabilitation training process, brain topography, brain network maps, and time-frequency feature maps, combined with exoskeleton feedback data, are used to assess the degree of improvement in the patient's motor abilities.

[0089] Example 1: Rehabilitation Training and Assessment Process for Lower Limb Mechanical Exoskeleton Based on Brain-Computer Interface

[0090] like Figure 3 As shown, a schematic diagram of the rehabilitation training and assessment process for a lower limb mechanical exoskeleton based on a brain-computer interface is presented.

[0091] Firstly, the paradigm design and data acquisition module designs personalized lower limb rehabilitation training paradigms based on the individual needs of each patient. This paradigm includes resting state exercises and motor imagery tasks to ensure that appropriate training content is tailored to the rehabilitation needs of each patient.

[0092] During the training process, EEG acquisition devices are used to record the patient's EEG signals in real time, especially the EEG data in motor imagery tasks, to capture the patient's motor intentions.

[0093] EEG data (electroencephalogram data) will be transmitted to the host computer system in real time for data storage and preliminary processing. After data processing, a motor intention decoding algorithm is used to analyze the EEG signals and decode the patient's motor imagery intentions. Based on these decoding results, the system generates control signals for controlling the lower limb mechanical exoskeleton, thereby assisting the patient in lower limb motor training.

[0094] The exoskeleton device dynamically adjusts its movement patterns based on control commands generated using decoded signals to assist patients in gait training and other lower limb movement training. During training, the system analyzes EEG and EMG characteristics to monitor the patient's muscle tone in real time. Based on the muscle tone status, it adaptively adjusts the intensity of functional electrical stimulation to alleviate abnormal muscle tone and enhance the stimulus feedback effect of rehabilitation training. After the rehabilitation training process, based on the EEG time-frequency characteristics of the rehabilitation training process, the degree of improvement in the patient's motor ability is assessed using power spectrum (PSD)-based brain topography, phase-locked value (PLV)-based brain network mapping, and wavelet transform (WT)-based time-frequency feature maps combined with exoskeleton feedback data.

[0095] Example 2: Pre-collection of EEG data and model training of patients before rehabilitation training

[0096] Before formal training begins, patients undergo multiple pre-collection experiments of motor imagery EEG. For example... Figure 4 As shown, the patient wears an EEG acquisition device to collect EEG signals during resting and motor imagery tasks. The patient performs resting (eyes closed, eyes open) and motor imagery tasks (e.g., imagining movement of the left or right leg). The system records the patient's EEG data in real time under different states and processes and stores this data offline for training a motor intention decoding model. The pre-collected data is used to extract features related to motor imagery and train the decoding algorithm, thereby accurately decoding the patient's motor intentions.

[0097] The decoding model includes a generative adversarial network module and a transfer learning network.

[0098] like Figure 5 As shown, the Generative Adversarial Network (GAN) module consists of two neural networks: a generator (G) and a discriminator (D), which are mutually optimized through adversarial training. The generator G's task is to optimize the input fused features W. fusion Generate fake features G(W) that approximate the motion imagination features of the real target domain. fusionThe discriminator D is used to distinguish whether the input features are real target domain motion imagination features. The forged feature G(W) is still generated by the generator. fusion Through adversarial training, the generator gradually learns to base its behavior on the fused feature W. fusion The generator generates high-quality fake features that closely resemble the real motion imagery features of the target domain, while the discriminator continuously improves its ability to distinguish between real and fake features. Ultimately, the trained generator can generate high-quality motion imagery features G(W) for the target domain. fusion This provides input support for the transfer learning stage. This process significantly alleviates the problem of insufficient target domain data and provides sufficient feature representation capabilities for subsequent decoding tasks.

[0099] The transfer learning network is the BAdaptNet network. For example... Figure 6 The diagram shows a schematic of the BAdaptNet transfer learning network. The BAdaptNet network is pre-trained using features from source domain motion imagery data. The BAdaptNet model consists of four core modules: Block 1, Block 2, Block 3, and the output layer.

[0100] Block 1 is a deep attention feature extraction layer that uses the SE attention mechanism to filter channels, enabling the network to perform dynamic channel feature recalibration to improve the network's representation ability and enhance the extraction of spatial representation information of EEG signals.

[0101] Block2 is a dynamic convolution feature extraction layer that uses dynamic convolution to adapt to the non-stationary characteristics of EEG signals and improve the flexibility of feature extraction.

[0102] Block3 is a multi-convolutional feature integration layer that enhances the model's information capture capability by fusing features through multiple convolutional layers.

[0103] The output layer uses a fully connected layer to predict features and outputs the prediction results.

[0104] During the pre-training process, firstly, the input layer processes the source domain motor imagery data into EEG features after preprocessing and feature extraction. The data is fed into a deep attention feature extraction layer (Block 1). This layer extracts local features through depthwise Conv2D operations and combines a Squeeze-and-Excitation Attention Layer (SE) to weight the importance of different channels, thereby achieving adaptive redistribution of global features. After global average pooling, the channel vector is obtained.

[0105]

[0106] Here, c represents the channel index, and H and W represent the height and width of the feature matrix, respectively. Then, the channel vector z is processed through two fully connected layers and a non-linear activation function to generate channel weights.

[0107] s=σ(W2·ReLU(W1·z))

[0108] Where W1 and W2 are trainable weight matrices, and σ is the sigmoid activation function. The final channel weights s are used to reweight the original feature matrix X, resulting in the enhanced feature matrix X. att .

[0109] X att,c (h,w)=s c ·X c (h,w)

[0110] Where c is the channel index, and h and w are the spatial position indices of the feature matrix, respectively.

[0111] Next, feature X att The module then moves to the dynamic convolution module (Block 2). This module uses two dynamically generated convolutional layers to perform adaptive convolution operations on different input features. The dynamic convolution module first generates K weight coefficients π through two fully connected layers and a non-linear activation function. k .

[0112]

[0113] Where W1 and W2 are trainable weight matrices, The Softmax activation function ensures that all π... k Satisfying nonnegativity and normalized. Define K convolution kernels. By using attention weight π k With convolution kernel W k Weighted summation generates a dynamic convolution kernel W. dynamic This dynamic convolution kernel can adaptively adjust the convolution operation based on the content of the input feature matrix.

[0114]

[0115] Using dynamically generated convolution kernel W dynamic For the input feature matrix X att Perform a two-dimensional convolution operation to obtain the output feature matrix X. dyn .

[0116] X dyn =Conv2D(Xatt W dynamic )

[0117] Conv2D(·) is a two-dimensional convolution function. Through the dynamic convolution module, the network can flexibly adjust the parameters of the convolution kernel to adapt to changes in input features, thereby improving its ability to adapt to data non-stationarity.

[0118] Then, the features are fed into a multi-convolutional feature fusion layer (Block 3), which processes the features through cascaded convolutional operations with multiple kernels of different sizes. The input features pass through two consecutive two-dimensional convolutional layers, each followed by a batch normalization layer and a ReLU activation function. The compressed features are then fed into a fully connected layer and classified using a Softmax activation function. Assuming the classification target is y, the network output is... The cross-entropy loss function is used for optimization during the training phase.

[0119]

[0120] Where N represents the total number of categories, y i For real labels, For the class probabilities predicted by the network, BAdaptNet achieves deep extraction of input features, adaptive enhancement, multi-feature fusion and classification prediction, providing an effective pre-trained model for transfer learning applications in the subsequent testing phase.

[0121] like Figure 7 As shown, a schematic diagram of BAdaptNet network transfer learning training is presented.

[0122] The process involves data preprocessing, BAdaptNet network pre-training, and generating EEG features of the target subjects' motor imagery using the RM-GAN network. The specific implementation flow for data generation is shown below.

[0123] Using the resting data of the collected target Source domain resting data Motion imagination data from the source domain Extract its time-frequency features, which are expressed as follows: and

[0124] To generate the target domain MI feature matrix, the source domain MI features are fused with the resting data using the following formula:

[0125]

[0126] Where Mean(·) is the mean function of the feature matrix along the time axis, and the feature matrix W fusionUsing a trained RM-GAN generator, motion visualization features of the target domain are generated.

[0127]

[0128] Wherein, G(·) is the data generation function of the generator after training. The generator adjusts the motion imagery features of the source domain to be consistent with the resting feature distribution of the target domain through the encoder-decoder structure, thereby generating motion imagery data with the characteristics of the target domain.

[0129] like Figure 7 As shown, in the BAdaptNet network transfer learning retraining, the transfer learning method is combined with freezing some pre-trained layers and inputting the personalized data of the target subjects and the target subject-specific data generated by RM-GAN into the model for fine-tuning training. By retaining the parameters of the general feature layers, the transfer learning stage reduces the computational cost and time of model training while enhancing the adaptability to the target subject data. The final model can quickly adapt to the personalized EEG signal features of the target subjects, improving the accuracy and efficiency of decoding.

[0130] The transfer learning retraining process is as follows:

[0131] The generated target domain motion imagination feature data This module is used to retrain the pre-trained BAdaptNet model for cross-domain adaptation. BAdaptNet consists of four modules: Block 1 and Block 2, which are deep attention feature extraction and dynamic feature enhancement modules for extracting key time-frequency features. Since these modules have already captured effective feature representations during training in the source domain, their parameters are frozen during the transfer learning phase to ensure knowledge transfer from the source domain to the target domain. Blocks 3 and 4 serve as heavy feature fusion and classification modules, and their parameters are adjusted during the transfer learning phase based on the generated motion visualization data from the target domain.

[0132]

[0133] in, Imagine a feature matrix for the generated target domain motion. The resting feature matrix of the target domain. For the parameters of the frozen Block 1 and Block 2, θ Block3,Block4 For the parameters of Block 3 and Block 4 that need to be trained, The optimized parameters for Block 3 and Block 4 are updated after retraining.

[0134] The BAdaptNet model, retrained through transfer learning, can be deployed in rehabilitation training systems for real-time EEG decoding, driving exoskeleton devices to achieve left and right leg stepping movements in target subjects. By decoding the target subject's motor intentions in real time, rehabilitation training tasks and plans can be dynamically adjusted, improving the application effectiveness of brain-computer interface systems in rehabilitation scenarios.

[0135] The BAdaptNet network transfer learning training and EEG decoding method includes the following steps:

[0136] The resting data of the source domain, the motor imagery data of the source domain, and the resting data of the target domain are preprocessed to obtain preprocessed EEG signals.

[0137] The pre-trained generator was obtained by pre-training the GAN generator network, which includes a generator and a discriminator, using pre-processed EEG signals.

[0138] The pre-trained transfer learning model is obtained by training a transfer learning network using pre-processed motion imagery data from the source domain.

[0139] Using the target resting data, source domain resting data, and source domain motion imagination data, after preprocessing and data fusion, a pre-trained generator is used to generate target motion imagination features.

[0140] By utilizing the generated target motion imagery features, the pre-trained transfer learning model is retrained through transfer learning, and a personalized EEG signal decoding model adapted to the target subjects is obtained through fine-tuning of transfer learning.

[0141] Using real-time collected target resting data, a personalized EEG signal decoding model is employed to obtain target motor imagery feature data.

[0142] The preprocessing includes: channel filtering of EEG signals including resting data from the source domain, motor imagery data from the source domain, and resting data from the target domain, to obtain channels related to the motor sensory cortex;

[0143] Bandpass filtering is used to extract rhythmic activities relevant to the motion imagery task.

[0144] In practical applications of brain-computer interface systems, the method of using the BAdaptNet network transfer learning pre-trained model described in this invention for EEG decoding can solve the problems of long model training time, difficulty in personalized adaptation, and low decoding accuracy caused by insufficient EEG data acquisition. In specific implementation, the GAN network is an RM-GAN network, and the transfer learning network is a BAdaptNet network. Based on resting data from the source domain, resting data from the target domain, and motor imagery data from the source domain, motor imagery data from the target domain is generated using RM-GAN. The BAdaptNet network is trained using the resting data and motor imagery data from the source domain, and then retrained using the generated data through transfer learning to achieve rapid adaptation and high-precision decoding.

[0145] The BAdaptNet model is adopted as the real-time motion intent decoding algorithm. The BAdaptNet model comprises three core modules: Block 1, Block 2, and Block 3. Block 1 uses an inverse bottleneck structure to deeply extract spatial information from EEG signals. The inverse bottleneck module employs extended-dimensional convolution operations, enhancing the ability to analyze low-dimensional features and reducing computational complexity. Block 2 introduces a dynamic convolution mechanism to adapt to the non-stationary nature of EEG signals, improving the flexibility of feature extraction and adding residual connections to enhance training stability. Block 3 employs multi-scale dilated convolution, with multiple dilated convolutions at different dilation rates extracting features in parallel, enhancing the network's robustness and generalization ability. It combines features using an inverse bottleneck structure and utilizes a padding strategy to optimize the receptive field of the convolution operation, further improving decoding performance.

[0146] Example 3: An Adaptive Functional Electrical Stimulation Feedback Method Based on Muscle Tone State

[0147] like Figure 8 As shown, coherence analysis of electroencephalography (EEG) and electromyography (EMG) is used to assess the patient's muscle tone status in real time, and low-frequency functional electrical stimulation (FES) is used for adaptive adjustment to alleviate abnormal muscle tone and improve the effectiveness of rehabilitation training. The core idea is to dynamically adjust electrical stimulation parameters based on coherence analysis to optimize neurofeedback regulation.

[0148] EEG-EMG coherence analysis was used to compare the changes in motor cortex (M1 area) activity under normal and abnormal muscle tone conditions. When abnormal muscle tone was detected, low-frequency functional electrical stimulation was applied to the target muscle group to relieve tension. The stimulation parameters were dynamically adjusted based on the coherence analysis results to gradually restore electromyographic activity to a normal state, achieving personalized adaptive intervention and improving the accuracy and targeting of rehabilitation effects.

[0149] During the data acquisition phase, two experimental conditions were set: one was a normal muscle tone state, serving as a baseline model for individual patients; the other was an abnormal muscle tone state, such as spasticity or rigidity, used to analyze the disordered patterns of neuromuscular regulation. The acquired EEG and EMG signals were used to calculate the coherence between the two, an indicator that quantifies the synchronicity between the two signals within a specific frequency band and is a key indicator for assessing the level of cortical-muscular neurocontrol. Coherence is defined as follows:

[0150]

[0151] in, The cross-power spectral density between EEG and EMG. and The self-power spectral densities are those of EEG and EMG, respectively. Coherence C xy (f), the higher the value, the tighter the neural regulation.

[0152] When the system detects that the coherence is below a set threshold (determined by the baseline model), it determines an abnormal muscle tone state and initiates low-frequency functional electrical stimulation (FES). The initial stimulation parameters are set to a frequency of 20-50 Hz and a pulse width of 200-400 μs. Low-frequency FES has the effect of modulating the excitability of motor neurons, which can alleviate excessive muscle contraction, promote the integration of sensory feedback, and enhance neural plasticity.

[0153] To further improve the individual adaptability of the intervention, the system introduces a closed-loop feedback control strategy based on coherence changes. Let C... baseline C represents the average coherence of the patient under normal conditions. current To determine the coherence of the current state, define an adjustment index:

[0154]

[0155] The stimulation parameters are dynamically adjusted based on the ΔC value. If |ΔC| < ∈ (the set tolerance range), the stimulation is considered effective, and the coherence tends to be normal; the current stimulation intensity is maintained. If ΔC < ∈, the coherence is low, indicating insufficient stimulation; the FES intensity should be appropriately increased or the stimulation duration extended. If ΔC > ∈, the coherence is high, which may cause excessive inhibition; the FES intensity should be appropriately reduced to prevent muscle weakness. This method enables adaptive electrical stimulation regulation driven by EEG-EMG coherence, ensuring the physiological effectiveness of the intervention while taking into account individual differences, ultimately promoting more precise and efficient muscle tone rehabilitation training.

[0156] Example 4: Rehabilitation Training Based on Decoding Models and Paradigms

[0157] like Figure 9As shown, after data pre-collection and model training are completed, patients will enter the testing phase before formal training. At this time, patients wear EEG acquisition devices and test the accuracy of the motor intention decoding model through a motor imagery task. The system provides real-time feedback based on the decoding results to ensure the model's stability and accuracy, preparing it for formal training.

[0158] A 15-day lower limb rehabilitation training experiment was conducted on the patient. Specifically, such as... Figure 10 As shown, before the experiment began, the patient wore a lower limb mechanical exoskeleton. After the exoskeleton was raised, the patient was in a standing position. The patient was kept in a stable and peaceful state. The patient wore an EEG acquisition cap (i.e., EEG acquisition device) and connected it to the host computer rehabilitation assistance software to open the rehabilitation training mode.

[0159] First, the system collects resting data for about 120 seconds. The patient observes the action prompts before the motor imagery, and then performs 6 seconds of motor imagery based on the prompts. The motor intention decoding algorithm decodes the data in real time and outputs action commands to control the exoskeleton to drive the patient's lower limbs for rehabilitation training. This cycle is repeated 30 times to complete one rehabilitation training session. Throughout the entire rehabilitation training process, the patient's EEG signal data is recorded during each motor imagery session.

[0160] In the above process, EEG signals of stroke patients during rehabilitation training are collected using an EEG acquisition device. In this embodiment, a 64-channel EEG signal acquisition system is used to collect the patient's EEG signal data at a sampling frequency of 1000Hz. In the EEG acquisition device, the position of each electrode represents a node in the brain network, and the number of electrodes is the number of nodes N. Considering that the prefrontal, parietal, temporal, and occipital lobes are the main areas of the brain responsible for cognitive tasks, 24 electrodes are selected for EEG acquisition to reduce interference from irrelevant information. Therefore, the corresponding electrodes for EEG acquisition in the 64-channel EEG signal acquisition system are: “Fp1”, “Fp2”, “F3”, “F4”, “Fz”, “AF3”, “AF4”, “P3”, “P4”, “P7”, “P8”, “Pz”, “CP1”, “CP2”, “CP3”, “CP4”, “T7”, “T8”, “Cz”, “O1”, “O2”, “Oz”, “C3”, “C4”.

[0161] Each acquired EEG data point is preprocessed, and then the power spectral density algorithm is used to extract energy features of different frequency bands. Wavelet transform is used to extract time-frequency features, and a functional brain network is drawn based on the phase-locked value (PLV). Based on these features, the patient's rehabilitation status is assessed, and objective evaluation indicators are provided to physicians to help them better judge the patient's rehabilitation status.

[0162] In this embodiment, the patient underwent a 15-day lower limb rehabilitation training experiment, which was repeated 30 times, twice a day, for a total of 15 days.

[0163] This brain-computer interface-based mechanical exoskeleton rehabilitation training system utilizes EEG acquisition equipment to collect and acquire patients' EEG signal data during each training experiment, particularly EEG data before and after the decision-making period. The lower limb mechanical exoskeleton device executes movement intentions based on EEG signal analysis and drives the lower limb exoskeleton to perform corresponding gait control. Combined with functional electrical stimulation feedback based on muscle tone status, it enhances the feedback effect of motor rehabilitation and records gait data (such as cadence and stride length) and training time for each experiment, as well as performing statistical analysis and visualization of the patient's motor performance. By performing steps such as EEG signal preprocessing, feature extraction, and movement intention recognition, and generating a visualized brain network map, the system can assess the patient's brain electrical activity and motor recovery progress.

[0164] This invention achieves a deep integration of brain-controlled exoskeleton training and low-frequency neuro-electrical stimulation modulation. While increasing the patient's active participation, it enhances the feedback input of the lower limb motor sensory pathway, promotes the plasticity reconstruction of the central nervous system, effectively alleviates motor disorders caused by excessive muscle tone, and improves the rehabilitation effect of stroke patients.

[0165] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and other materials. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several of the functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0166] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A brain-computer interface-based mechanical exoskeleton rehabilitation training system, characterized in that, include: The electroencephalogram (EEG) acquisition device is configured to collect EEG signals of cerebral cortex activity in rehabilitation patients during motor imagery. An electromyography (EMG) acquisition device is configured to acquire EMG signals from rehabilitation patients during motor imagery. The preprocessing unit is configured to preprocess the acquired electroencephalogram (EEG) and electromyogram (EMG) signals. The feature extraction unit is configured to extract features from the preprocessed EEG and EMG signals to obtain the time-frequency features of EEG and EMG; The motor intention decoding unit is configured to decode the patient's motor intention using an EEG decoding model based on the EEG time-frequency characteristics, and generate mechanical exoskeleton control commands based on the motor intention. The motor intention decoding unit is a personalized EEG signal decoding model adapted to the target patient, obtained through transfer learning fine-tuning. Specifically: Preprocessing of resting data from the source domain, source domain motor imagery data, and target domain resting data yields preprocessed EEG signals; pre-training a GAN generator network including a generator and discriminator using the preprocessed EEG signals yields a pre-trained generator; training a transfer learning network using the preprocessed source domain motor imagery data yields a pre-trained transfer learning model; generating target motor imagery features using the pre-trained generator with feature data from the target resting data, source domain resting data, and source domain motor imagery data after preprocessing and data fusion; and retraining the pre-trained transfer learning model using the generated target motor imagery features, and obtaining a personalized EEG signal decoding model adapted to the target patient through transfer learning fine-tuning. The mechanical exoskeleton control unit is configured to drive the limb movements of the rehabilitation patient for rehabilitation training according to the mechanical exoskeleton control commands; The muscle tone monitoring unit is configured to analyze the patient's muscle tone status based on the time-frequency characteristics of EEG and EMG. The adaptive functional electrical stimulation feedback unit is configured to adaptively apply functional electrical stimulation of different intensities according to the patient's muscle tone state during rehabilitation training, thereby enhancing stimulation feedback and regulating the patient's muscle tone state.

2. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 1, characterized in that, The system also includes an assessment unit that monitors the time-frequency characteristics of the EEG of the rehabilitation patient in real time, generates brain topography maps, functional brain networks and time-frequency characteristic maps based on the time-frequency characteristics of the EEG, and assesses the degree of improvement of the rehabilitation patient's motor ability by combining feedback data obtained from the mechanical exoskeleton. The feedback data includes gait, joint angles and movement speed.

3. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 2, characterized in that, The EEG time-frequency features and EMG time-frequency features are obtained by extracting EEG signals based on wavelet transform. The brain topography and functional brain network were obtained by phase-locked value analysis of the functional brain network, and were used to assess information transmission between different regions of the brain.

4. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 2, characterized in that, The system also includes an EEG display unit for real-time or offline display of brain topography maps, functional brain networks, and time-frequency feature maps generated by the evaluation unit.

5. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 1, characterized in that, The preprocessing unit preprocesses the acquired EEG signals, including noise reduction, filtering, and artifact removal.

6. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 1, characterized in that, The mechanical exoskeleton control unit processes the patient's real-time motor imagery EEG signals based on a personalized decoding model, decodes the patient's motor intentions, and converts them into exoskeleton drive signals, thereby controlling the mechanical exoskeleton to drive the patient's limb movements.

7. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 1, characterized in that, The muscle tone monitoring unit utilizes the time-frequency characteristics of electroencephalogram (EEG) and electromyogram (EMG) signals and employs coherence analysis to analyze the patient's current muscle tone status.

8. The brain-computer interface-based mechanical exoskeleton rehabilitation training system according to claim 1, characterized in that, The adaptive functional electrical stimulation feedback unit applies functional electrical stimulation of different intensities during brain-controlled exoskeleton rehabilitation training based on muscle tone monitoring results, thereby alleviating abnormal muscle tone in the patient's limbs and enhancing the stimulation feedback effect of motor rehabilitation.

Images