A personalized neuro-modulation physiotherapy method and system based on a brain-computer interface

Abstract

The application discloses a kind of individualized neuroregulation physiotherapy method and system based on brain-computer interface, comprising the following steps: S1, electroencephalogram signal acquisition and pretreatment;S2, multi-modal physiological signal fusion;S3, neural state assessment;S4, individualized physiotherapy scheme generation;S5, neuroregulation execution;S6, rehabilitation effect evaluation;S7, scheme dynamic optimization, the present application relates to the field of neuroregulation technology.The present application solves the problem of insufficient specificity, precision and safety of physiotherapy due to the fixed parameters of traditional neuroregulation physiotherapy scheme, dependence on artificial experience, single state assessment, poor anti-interference, lack of safety constraints and inability to adaptively and iteratively optimize according to rehabilitation effect.

Description

Technical Field

[0001] This invention relates to the field of neuromodulation technology, and in particular to a personalized neuromodulation therapy method and system based on brain-computer interface. Background Technology

[0002] Brain-computer interfaces (BCIs) establish direct communication between the human brain and external devices by acquiring and decoding electroencephalogram (EEG) signals. The combination of BCIs and neuromodulation technologies offers a new pathway for the rehabilitation of motor dysfunction, cognitive impairment, and mood disorders caused by stroke, traumatic brain injury, and neurodegenerative diseases. However, existing solutions have several unresolved issues: First, BCI and neuromodulation operate independently. BCI only performs signal acquisition and simple feedback, while neuromodulation is executed according to fixed parameters; there is no closed loop between the two. For example, CN122208943A only involves hierarchical regulation of single-modal EEG feedback, and CN122208977A focuses ultrasound modulation for device safety; neither establishes a closed-loop feedback between EEG signals and therapeutic effects.

[0003] Second, the generated treatment plans lack quantitative evidence of neurological status. Existing plans, such as the digital twin rehabilitation modeling in CN122224528A and the multi-physical factor therapy in CN122208435A, rely on subjective scales or simple rules for generating treatment plans, without using EEG signals to quantitatively assess neurological function, resulting in poor targeting.

[0004] Third, rehabilitation assessments rely on single-modal data. Existing protocols such as CN122220967A and CN122208168A depend on single-modal assessments of EEG or behavior, lacking multimodal fusion of EEG, behavior, and physiological parameters, resulting in incomplete and inaccurate assessments.

[0005] Fourth, there is no real-time EEG feedback to dynamically adjust the treatment plan. The existing plan does not establish a dynamic adjustment between real-time EEG signals and stimulation parameters. Once the physical therapy begins, it runs on fixed parameters without adjusting according to changes in the patient's condition.

[0006] In summary, there is a need for a method and system that integrates EEG acquisition, multimodal fusion assessment, personalized protocol generation, neuromodulation execution, and effect feedback optimization into a complete closed loop. Summary of the Invention

[0007] To address the problems of traditional neuromodulation therapy programs, such as fixed parameters, reliance on human experience, single state assessment with poor anti-interference capabilities, lack of safety constraints, and inability to adaptively and iteratively optimize based on rehabilitation effects, resulting in insufficient targeting, precision, and safety of the therapy, the present invention aims to provide a personalized neuromodulation therapy method and system based on brain-computer interface.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a personalized neuromodulation therapy method based on brain-computer interface, comprising the following steps: S1, EEG signal acquisition and preprocessing: real-time acquisition of the patient's EEG signal through a non-invasive EEG device, followed by filtering, artifact removal, and segmentation, and extraction of power spectral density, functional connectivity, and event-related potential features; S2, multimodal physiological signal fusion: simultaneous acquisition of behavioral data and physiological parameters, fusion with EEG signal features to construct a multimodal feature vector, with fusion weights dynamically allocated according to the signal-to-noise ratio of each modality; S3, neural state assessment: assessment of neural state using a deep learning model based on the fused features, and outputting a motor cortex excitability score. Cognitive function status score and mood status score Calculate the overall score , S4, Personalized Therapy Plan Generation: Based on the comprehensive score, a therapy plan is selected or generated from the plan knowledge base, and the stimulation intensity is determined. , S5, Neuromodulation Execution: The neuromodulation device is controlled to execute stimulation according to the protocol, with real-time monitoring of stimulation parameters and EEG response; It stops if abnormalities are detected. S6, Rehabilitation Effect Assessment: Comprehensive evaluation of changes in EEG characteristics. (Note: The text also mentions a safety upper limit for current, stimulation frequency, duration, and location, but this seems unrelated to the main point about stimulation.) , , Calculate the assessment value based on the degree of behavioral improvement. , S7, dynamic optimization of the scheme: based on E_Recovery, the evaluation model parameters and scheme generation parameters are updated using gradient descent, and fed back to S3 and S4 to form a closed loop.

[0009] Preferably, in S2: behavioral data includes motion trajectory, task completion rate, reaction time, and action accuracy; physiological parameters include heart rate variability, skin conductance, and electromyography signals; and the fusion weights w_EEG, w_Beh, and w_Phys satisfy the following... , , ,and Each feature was standardized by Z-score before fusion.

[0010] Preferably, in S1: a 16-64 channel non-invasive EEG device is used, with a sampling rate of 256-1024 Hz, 0.5-45 Hz bandpass filtering, independent component analysis to remove artifacts, a sliding window of 2 s / 50% overlap segmentation, to extract the power spectral density characteristics, phase lock values, and event-related potential characteristics of the α (8-13 Hz), β (13-30 Hz), θ (4-8 Hz), and δ (0.5-4 Hz) frequency bands.

[0011] Preferably, in S3, the deep learning model is CNN-LSTM: one-dimensional convolution extracts spatiotemporal local EEG patterns, bidirectional LSTM models time dependence, and multi-head fully connected branches output motor cortex excitability score S_MC, cognitive function status score S_CF, and emotional state score S_ES respectively; S_MC is calculated based on μ-wave and β-wave event-related desynchronization / synchronization features, S_CF is calculated based on the θ / β power ratio and P300 component amplitude, and S_ES is calculated based on prefrontal α asymmetry and skin conductance level; S_MC is used for stimulus intensity calculation, and S_CF and S_ES are used for rehabilitation effect evaluation and dynamic optimization of the program.

[0012] Preferably, the scheme selection rule in S4 is as follows: hour, Take the upper limit and not exceed , ; hour, , ; hour, , The above discretization rule is the engineering approximation of the continuous function I_Stim=min(I_Base·(1+β·(1-S_MC)),I_Max) in each interval of equation (3), and each level is affected by constraint.

[0013] Preferably, in S5: the neuromodulation device includes a transcranial direct current stimulation (tDCS) unit, a repetitive transcranial magnetic stimulation (rTMS) unit, and a functional electrical stimulation (FES) unit; the safety protection circuit monitors the stimulation current / voltage, electrode-skin impedance, and electrode temperature in real time, and the stimulation current exceeds... impedance exceeds Electrode temperature exceeds It may automatically shut down when epileptiform discharge is detected.

[0014] Preferably, in S6: , , , ; To significantly improve, This indicates a slight improvement. No change It is worsening.

[0015] Preferably, in step S7: when three consecutive assessments show significant improvement (E_Recovery ≥ 0.3), the stimulation intensity is decreased by 10% to 20% of the current stimulation intensity; when there is no change or deterioration after three consecutive assessments, the neural state assessment model parameters are updated; after every five accumulated treatment data points, gradient descent is used to update the model weights and protocol generation parameters, and the learning rate. loss function The treatment plan knowledge base is updated every 10 treatments.

[0016] A personalized neuromodulation therapy system based on a brain-computer interface includes: an EEG signal acquisition device for real-time acquisition of EEG signals via multi-channel non-invasive EEG electrodes, followed by filtering and artifact removal to extract power spectral density, functional connectivity, and event-related potential features; a multimodal signal fusion module for simultaneously acquiring behavioral data and physiological parameters, fusing them with EEG signal features, and outputting a multimodal feature vector; and a neural state assessment module for evaluating neural states based on the fused features using a CNN-LSTM model. , , and The personalized physiotherapy plan generation module is used to select or generate physiotherapy plans from a plan knowledge base based on neurological status scores, and to determine the specific treatment plan. Including frequency, duration, and location; neuromodulation actuator, including stimulation signal generator, stimulation electrodes / coils, and safety protection circuitry, for performing tDCS, rTMS, or FES stimulation; rehabilitation effect assessment module for calculating... The system categorizes the effectiveness levels; the dynamic optimization module updates the evaluation model and solution generation parameters using gradient descent based on E_Recovery, and feeds back to the neural state evaluation module and solution generation module to form a closed loop.

[0017] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: 1. This invention deeply integrates brain-computer interface (BCI) technology with neuromodulation therapy technology to construct a closed-loop system covering the entire process of signal acquisition, neural state assessment, therapy plan generation, modulation execution, effect evaluation, and parameter optimization. EEG data runs through all treatment stages from S1 to S7, enabling dynamic adaptive adjustment of therapy parameters based on the patient's real-time neural state. This completely abandons the traditional treatment mode of fixed preset parameters and significantly improves the pertinence and adaptability of neuromodulation therapy.

[0018] 2. This invention constructs a CNN-LSTM hybrid deep learning model, which automatically calculates motor cortex excitability scores, cognitive function scores, emotional state scores, and comprehensive neurological state scores based on the fusion features of EEG, behavior, and physiological multimodal data. It generates physiotherapy plans and calculates personalized stimulation intensity based on quantitative neurological state data, replacing the traditional extensive treatment method that relies on doctors' subjective scale scores and manual experience to formulate plans, and realizing the precise and data-driven generation of physiotherapy plans.

[0019] 3. This invention adopts a three-modal signal feature-level fusion strategy of EEG, behavior and physiology, which overcomes the limitations of single EEG assessment. At the same time, it dynamically allocates fusion weights according to the real-time signal-to-noise ratio of each modality, and assigns higher weights to modalities with high signal quality and large amount of effective information. This effectively suppresses the assessment bias caused by single signal noise and interference, and greatly improves the accuracy, stability and anti-interference ability of patient neurological state assessment.

[0020] 4. This invention, by setting a personalized adaptive stimulation intensity adjustment strategy with a safety threshold constraint, can dynamically adjust the stimulation intensity according to the patient's motor cortex excitability. At the same time, it sets a fixed safety upper limit for the maximum stimulation current, so that even if the patient's motor cortex excitability score is extremely low, the stimulation current will not exceed the safety threshold, thus avoiding the risk of excessive stimulation at the algorithm level. In addition, it is equipped with hardware-level safety protection circuits to monitor abnormal states such as overcurrent, high electrode impedance, electrode overheating, and epileptiform discharge in real time and automatically cut off power for protection, thus ensuring the safety of the physiotherapy process from both algorithm and hardware layers.

[0021] 5. This invention introduces a gradient descent optimization algorithm based on rehabilitation effect evaluation values ​​and a knowledge base iteration mechanism. It continuously updates the weights of the neurological state assessment model and the parameters for generating the physiotherapy plan based on the rehabilitation effect of each physiotherapy session. At the same time, it periodically iterates and updates the plan knowledge base, realizing the adaptive evolution of the physiotherapy model and treatment plan. This allows the system to continuously improve the evaluation accuracy and plan matching degree as treatment data accumulates, and has the iterative optimization characteristic of becoming more accurate with use. Attached Figure Description

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a schematic diagram of the system of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0023] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

[0024] Please see Figures 1 to 2It should be understood that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and to facilitate understanding and reading. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the technical content disclosed in this invention. Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.

[0025] This invention provides a technical solution: a personalized neuromodulation therapy method and system based on brain-computer interface, belonging to the field of brain-computer interface neurorehabilitation and intelligent physiotherapy technology. It mainly addresses the technical problems of traditional neuromodulation therapy programs, such as rigidity of protocols, standardized parameters, inability to adapt to differences in the neurological states of different patients, delayed feedback on rehabilitation effects, and lack of closed-loop optimization iteration. This invention achieves precise, personalized, and adaptive neuromodulation therapy for patients with motor, cognitive, and emotional neurological dysfunctions through multimodal physiological signal fusion, deep learning-based quantitative assessment of neurological states, personalized adaptive stimulation parameter matching, real-time safety monitoring, and closed-loop optimization of rehabilitation effects.

[0026] The overall operation process of this invention is as follows: First, the patient's EEG signals are collected using a non-invasive EEG device and preprocessed and feature extracted. Simultaneously, patient behavioral data and human physiological parameters are collected to complete multimodal feature fusion. Based on a CNN-LSTM deep learning model, the patient's motor cortex excitability, cognitive function, and emotional state are quantitatively evaluated to obtain a comprehensive neurological state score. Personalized neuromodulation physiotherapy parameters, including stimulation intensity, stimulation duration, stimulation frequency, and stimulation site, are adaptively matched according to the score. Safe and controllable neurostimulation treatment is completed using multiple types of neuromodulation hardware units. After treatment, the rehabilitation improvement level is quantitatively evaluated. Based on feedback data from multiple treatments, the model parameters and physiotherapy plan parameters are iteratively optimized using a gradient descent algorithm to form a fully closed-loop intelligent physiotherapy system of "collection—evaluation—treatment—feedback—optimization".

[0027] S1. Example of EEG signal acquisition and preprocessing: In this embodiment, a 32-channel non-invasive dry electrode EEG acquisition device was selected, with a sampling rate set to 512Hz, meeting the general sampling standard of 256–1024Hz and adapting to the signal acquisition accuracy requirements of clinical physiotherapy scenarios. The raw EEG signal was subjected to a 0.5–45Hz bandpass filter to remove power frequency interference, DC drift, and high-frequency electromyographic noise. Independent component analysis (ICA) algorithm was used to remove ocular artifacts, blinking artifacts, and limb movement artifacts, preserving the effective EEG components to the greatest extent possible.

[0028] The signal segmentation method employs a 2-second sliding time window with 50% overlap between adjacent windows to ensure the continuity and integrity of the temporal signal. After preprocessing, power spectral density features of four classic EEG frequency bands—δ (0.5–4Hz), θ (4–8Hz), α (8–13Hz), and β (13–30Hz)—are extracted. Simultaneously, brain region phase-locked value functional connectivity features and event-related potential temporal features are extracted to provide high-dimensional EEG feature support for subsequent neural state assessment.

[0029] S2. Example of multimodal physiological signal fusion: This invention abandons the single EEG assessment method and adopts a three-modal fusion assessment mechanism of "EEG + behavior + physiology". Among them, behavioral data is collected simultaneously from the patient's movement trajectory coordinates, task completion accuracy, response time and movement deviation accuracy during the physical therapy task; physiological parameters are collected simultaneously from heart rate variability, skin conductance level and surface electromyography signal, comprehensively covering the changes in the central nervous system, autonomic nervous system and somatic movement state during the patient's neuromodulation process.

[0030] Before feature fusion, all modality features are Z-score standardized to eliminate differences in feature dimensions. Fusion weights are dynamically assigned based on the real-time signal-to-noise ratio of each modality, with strictly defined weight values: EEG feature weights. Behavioral feature weights Physiological characteristic weights And satisfy By fusing and concatenating features at the feature level, a multimodal fusion feature vector with unified dimensions and complementary features is obtained, which greatly improves the accuracy and robustness of subsequent neural state assessment.

[0031] S3. Example of quantitative assessment of neural state: This invention employs a CNN-LSTM hybrid deep learning model to perform quantitative assessment of neural states. The model structure includes a one-dimensional convolutional layer, a bidirectional LSTM layer, and a multi-head fully connected output branch. The one-dimensional convolutional layer is responsible for extracting local spatiotemporal features and frequency band feature patterns of EEG signals, while the bidirectional LSTM layer accurately models the temporal dependencies of long-term EEG signals, thus solving the problem of insufficient temporal feature extraction in traditional models.

[0032] The model is configured with three independent output branches, each outputting one of three core neural state scores: motor cortex excitability score. Cognitive function status score Mood state score The specific calculation basis is as follows: Based on the quantitative calculation of event-related desynchronization / synchronization features of μ waves and β waves in the brain's motor cortex, the degree of activation of the motor cortex can be accurately reflected. Based on the θ / β power ratio and the amplitude of P300 event-related potential components, the patient's attention, memory, and cognitive regulation abilities are quantified. Based on the asymmetric characteristics of alpha waves in the prefrontal cortex and the calculation of skin conductance levels, the system provides objective feedback on patients' emotional states such as anxiety, relaxation, and excitement.

[0033] The final comprehensive neurological state score is obtained by weighted summation, and the calculation formula is as follows: The weighting coefficients satisfy The weights can be adaptively fine-tuned according to the type of patient's condition, increasing for patients with movement disorders. Weighting, increased in patients with cognitive impairment Weighting, increased in patients with mood disorders Weights.

[0034] S4. Personalized physiotherapy plan adaptively generates example: This invention achieves adaptive and personalized matching of physical therapy parameters based on the patient's real-time neurological status score. The core stimulation intensity calculation formula is as follows: In the formula Based on the basal stimulation current, For adaptive adjustment coefficients, The system's safe current limit is set so that all stimulation parameters do not exceed the safe threshold.

[0035] To adapt to engineering applications, this invention sets discretized parameter matching rules to simplify the engineering implementation of continuous functions: when This indicates that the patient's motor cortex is severely underexcitable. The maximum safe stimulation intensity and stimulation duration were used. ;when The patient's neurological state was moderately abnormal, and moderate stimulation intensity was used. Stimulation duration ;when The patient's neurological condition was good, and low-intensity conditioning stimulation was used. Stimulation duration It uses long-lasting, gentle regulation to consolidate the nerve state. At the same time, the system automatically matches the stimulation site and frequency according to the patient's lesion area, generating a complete physiotherapy plan.

[0036] S5. Multimodal Neural Modulation and Security Protection Implementation Example: This invention integrates three mainstream neuromodulation modes, including tDCS transcranial direct current stimulation, rTMS repetitive transcranial magnetic stimulation, and FES functional electrical stimulation. These modes can be freely switched or combined according to the patient's condition, and are suitable for various physiotherapy scenarios such as central nervous system injury, limb movement disorders, and cognitive and emotional abnormalities.

[0037] The hardware is equipped with a dedicated safety protection circuit, enabling real-time monitoring and abnormal self-locking protection throughout the entire process. During operation, it monitors the stimulation current, stimulation voltage, electrode-skin contact impedance, and electrode surface temperature in real time; if the stimulation current exceeds... Skin resistance exceeds Electrode temperature exceeds When epileptiform discharge waveforms are detected in real-time EEG signals, the system immediately and automatically cuts off the stimulation output, terminates the physiotherapy process, and records abnormal data, eliminating safety risks from a hardware perspective. During the physiotherapy process, the system simultaneously collects the patient's EEG response signals in real time, dynamically monitoring the neurofeedback state to ensure the safety and effectiveness of the treatment.

[0038] S6. Example of quantitative assessment of rehabilitation effect: After a single physiotherapy session, the system automatically collects the differences in core neurological scores before and after treatment. The comprehensive rehabilitation assessment value is obtained through weighted calculation. The calculation formula is as follows: The weights satisfy ,in , , This aligns with the core evaluation logic of neurorehabilitation.

[0039] The system classifies rehabilitation outcomes into four levels: Significant improvement was observed, and nerve function was markedly repaired. The improvement is slight, and the neurological state has been slightly optimized. No significant changes were observed, and the therapeutic effect remained unchanged. If the condition worsens, the physical therapy plan needs to be adjusted immediately. A quantitative grading method enables standardized and visualized evaluation of rehabilitation effectiveness.

[0040] S7. Example of Closed-Loop Dynamic Scheme Optimization: The core innovation of this invention is its closed-loop adaptive iterative optimization mechanism, which completely solves the shortcomings of traditional physiotherapy plans that are fixed and cannot adapt to the patient's recovery process. The system automatically iterates and optimizes based on the effects of multiple consecutive physiotherapy sessions: when three consecutive sessions show significant improvement, it indicates that the current stimulation intensity is too high, and the stimulation intensity is reduced by 10%–20% of the current stimulation parameters to consolidate the recovery effect with a gentle adjustment method and avoid overstimulation; when there is no change or deterioration occurs after three consecutive sessions, the system automatically updates the parameters of the neurological state assessment model and corrects the feature weights and scoring logic.

[0041] The data iteration and update rule is as follows: every 5 complete treatment data sessions, the gradient descent algorithm is used to update the weights of the CNN-LSTM model and the parameters for generating the physiotherapy plan, and the learning rate is set. The loss function uses mean squared error loss: By minimizing the loss function, the system continuously improves the model evaluation accuracy and the suitability of the treatment plan. After every 10 treatments, the system will input the optimized high-quality plan into the plan knowledge base, update the case database data, realize the global plan iterative upgrade, and continuously improve the overall physiotherapy accuracy of the system.

[0042] System Hardware Module Implementation Example: This invention also protects a personalized neuromodulation physiotherapy system adapted to the above-mentioned physiotherapy methods. The system consists of seven core modules, each corresponding to a step in the method: an EEG signal acquisition device responsible for multi-channel EEG signal acquisition and preprocessing feature extraction; a multimodal signal fusion module realizing feature-level fusion of EEG, behavioral, and physiological signals; a neural state assessment module outputting quantitative neural scores based on a CNN-LSTM model; a personalized physiotherapy plan generation module adaptively matching optimal stimulation parameters; a neuromodulation execution device realizing multi-modal safe neural stimulation; a rehabilitation effect assessment module quantifying the physiotherapy level; and a plan dynamic optimization module completing a closed-loop iteration throughout the entire process. All modules work together to achieve an intelligent, automated, and personalized neuromodulation physiotherapy process.

[0043] Table 1 lists all the symbols used in the entire text and their meanings: Table 2 Key Technical Parameters: Note: The range is It allows for flexible adjustments in different rehabilitation scenarios. The range is Similarly, adjustments can be made based on the main rehabilitation goals.

[0044] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A personalized neuromodulation therapy method based on brain-computer interface, characterized in that, Includes the following steps: S1, EEG signal acquisition and preprocessing: The patient's EEG signal is acquired in real time through a non-invasive EEG device, and after filtering, artifact removal, and segmentation, power spectral density, functional connectivity and event-related potential features are extracted; S2, Multimodal physiological signal fusion: Simultaneously collect behavioral data and physiological parameters, fuse them with EEG signal features, construct multimodal feature vectors, and dynamically allocate fusion weights according to the signal-to-noise ratio of each modality; S3, Neural State Assessment: Based on fused features, a deep learning model is used to assess neural state and output a motor cortex excitability score. Cognitive function status score and mood status score Calculate the overall score , Among them, the motor cortex excitability score is calculated based on the event-related desynchronization / synchronization features of EEG μ waves and β waves, the cognitive function status score is calculated based on the θ / β power ratio and P300 component amplitude, and the emotional state score is calculated based on prefrontal α wave asymmetry and skin conductance level. S4, Personalized Therapy Plan Generation: Based on the comprehensive score, a therapy plan is selected or generated from the plan knowledge base, and the stimulation intensity is determined. Where I_Base is the basic stimulation current, β is the adaptive adjustment coefficient, I_Max is the safe upper limit current, and the stimulation frequency, duration and site are also specified. S5, Neuromodulation Execution: Control the neuromodulation device to execute stimulation according to the plan, monitor stimulation parameters and EEG response in real time, and stop when abnormality occurs; S6, Rehabilitation Effectiveness Assessment: Comprehensive EEG Characteristic Changes , , Calculate the assessment value based on the degree of behavioral improvement. , To classify into levels; S7, Dynamic optimization of the scheme: The evaluation model parameters and scheme generation parameters are updated using gradient descent based on E_Recovery, and fed back to S3 and S4 to form a closed loop.

2. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S2: behavioral data includes movement trajectory, task completion rate, reaction time, and action accuracy; physiological parameters include heart rate variability, skin conductance, and electromyography signals; the fusion weights w_EEG, w_Beh, and w_Phys satisfy the following conditions: , , ,and Each feature was standardized by Z-score before fusion.

3. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S1: a 16-64 channel non-invasive EEG device is used, with a sampling rate of 256-1024 Hz, 0.5-45 Hz bandpass filtering, independent component analysis to remove artifacts, a sliding window of 2 s / 50% overlap segmentation, to extract power spectral density features, phase lock values ​​and event-related potential features in the α (8-13 Hz), β (13-30 Hz), θ (4-8 Hz), and δ (0.5-4 Hz) frequency bands.

4. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S3, the deep learning model is CNN-LSTM: one-dimensional convolution extracts spatiotemporal local EEG patterns, bidirectional LSTM models time dependence, and multi-head fully connected branches output motor cortex excitability score S_MC, cognitive function status score S_CF, and emotional state score S_ES, respectively; S_MC is calculated based on μ-wave and β-wave event-related desynchronization / synchronization features, S_CF is calculated based on the θ / β power ratio and P300 component amplitude, and S_ES is calculated based on prefrontal α asymmetry and skin conductance level; S_MC is used for stimulus intensity calculation, and S_CF and S_ES are used for rehabilitation effect evaluation and dynamic optimization of the program.

5. A personalized neuromodulation therapy method based on a brain-computer interface according to claim 1, characterized in that, The scheme selection rule in S4 is as follows: hour, Take the upper limit and not exceed , ; hour, , ; hour, , The above discretization rule is the engineering approximation of the continuous function I_Stim=min(I_Base·(1+β·(1-S_MC)),I_Max) in each interval of equation (3), and each level is affected by Constraints, where T_Stim is the duration of a single stimulus.

6. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S5: the neuromodulation device includes a transcranial direct current stimulation (tDCS) unit, a repetitive transcranial magnetic stimulation (rTMS) unit, and a functional electrical stimulation (FES) unit; the safety protection circuit monitors the stimulation current / voltage, electrode-skin impedance, and electrode temperature in real time, and the stimulation current exceeds... impedance exceeds Electrode temperature exceeds It may automatically shut down when epileptiform discharge is detected.

7. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S6: , , , ; To significantly improve, This indicates a slight improvement. No change It is worsening.

8. The personalized neuromodulation therapy method based on brain-computer interface according to claim 1, characterized in that, In S7: when three consecutive assessments show significant improvement (E_Recovery ≥ 0.3), the stimulation intensity is decreased by 10% to 20% of the current intensity; when there is no change or worsening after three consecutive assessments, the neural state assessment model parameters are updated; after every five accumulated treatment data points, gradient descent is used to update the model weights and protocol generation parameters, and the learning rate. loss function Where E_target is the target rehabilitation assessment value and η is the learning rate; the protocol knowledge base is updated after every 10 treatments.

9. A personalized neuromodulation therapy system based on a brain-computer interface, characterized in that, include: The EEG signal acquisition device is used to acquire EEG signals in real time through multi-channel non-invasive EEG electrodes, and extract power spectral density, functional connectivity and event-related potential features after filtering and artifact removal; The multimodal signal fusion module is used to simultaneously collect behavioral data and physiological parameters, fuse them with EEG signal features, and output multimodal feature vectors. The neural state assessment module is used to evaluate neural states based on fused features using a CNN-LSTM model. , , and The personalized physiotherapy plan generation module is used to select or generate physiotherapy plans from a plan knowledge base based on neurological status scores, and to determine the specific treatment plan. Including frequency, duration, and location; neuromodulation actuator, including stimulation signal generator, stimulation electrodes / coils, and safety protection circuitry, for performing tDCS, rTMS, or FES stimulation; rehabilitation effect assessment module for calculating... The system categorizes the effectiveness levels; the dynamic optimization module updates the evaluation model and solution generation parameters using gradient descent based on E_Recovery, and feeds back to the neural state evaluation module and solution generation module to form a closed loop.

Images