Central-peripheral combined electrical stimulation upper limb hand function regulation system and method

Abstract

This invention discloses a central-peripheral combined electrical stimulation system and method for upper limb hand function regulation, belonging to the field of neurorehabilitation engineering technology. The system includes a peripheral functional electrical stimulation regulation module, a central transcranial electrical stimulation regulation module, and a central-peripheral combined regulation module, forming a three-level closed-loop regulation architecture of hierarchical decoupling and recoupling. Methodically, it collects and processes multi-channel EEG signals, surface electromyography signals, and motor state signals, and performs independent closed-loop regulation; it calculates cortical-muscle coherence and functional cortical-muscle coherence to quantify the central-peripheral coupling strength, and combines this with conduction time difference, employing a deep reinforcement learning algorithm to perform online joint optimization of the relative triggering sequence and stimulation time difference between central and peripheral stimuli, achieving synergistic regulation through dual closed loops and parameter decoupling followed by recoupling. This invention effectively improves the activation efficiency of the central-peripheral circuit and the accuracy of rehabilitation regulation.

Description

Technical Field

[0001] This invention relates to the field of neurorehabilitation engineering technology, and in particular to a central-peripheral combined electrical stimulation system and method for regulating upper limb hand function. Background Technology

[0002] Upper limb motor dysfunction caused by diseases such as stroke, spinal cord injury, and peripheral nerve injury severely affects patients' independence in daily life and quality of life. Studies have shown that after central nervous system injury, the recovery of motor function depends on the coordinated efforts of central nervous system plasticity remodeling and peripheral effector functional reconstruction; neither simple central nor peripheral stimulation can achieve ideal rehabilitation results.

[0003] Currently, electrical stimulation technology has become an important means of motor function rehabilitation. Peripheral functional electrical stimulation (fEP) stimulates muscle contraction by applying electrical pulses to target muscle groups, providing early replacement or assistance for voluntary movement after central nervous system injury; central transcranial electrical stimulation (TCS) promotes neural plasticity changes by modulating the excitability of specific cortical areas. However, existing electrical stimulation intervention systems have the following technical limitations: (1) Insufficient precision in regulation. Existing systems rely heavily on physicians' subjective experience in setting stimulation parameters, lacking quantitative evaluation and adaptive adjustment capabilities based on real-time physiological signals, making it difficult to achieve precise regulation for individual differences among different patients.

[0004] (2) Lack of synergy in activation of central-peripheral circuits. Central and peripheral stimulation are usually used as independent modules. There is a lack of a joint optimization mechanism for temporal coordination and intensity coupling between the two, which cannot effectively activate the central-peripheral neural circuits and limits the role of synergistic rehabilitation effect.

[0005] (3) Lack of multidimensional biofeedback. In the process of regulating stimulation parameters, the existing system lacks real-time quantitative assessment and closed-loop feedback of central activation effect, peripheral action execution quality and central-peripheral coupling strength, which makes it impossible to dynamically adjust stimulation parameters according to the patient's real-time status.

[0006] To address the aforementioned issues, this invention proposes an integrated central-peripheral combined electrical stimulation modulation system and method. By constructing a three-level hierarchical, dual-closed-loop, parameter decoupling and re-coupling central-peripheral synergistic modulation stimulation mechanism, it achieves a comprehensive improvement in peripheral output stability, central modulation effectiveness, and central-peripheral synergistic efficiency. Summary of the Invention

[0007] The purpose of this invention is to provide a central-peripheral combined electrical stimulation system and method for regulating upper limb hand function, which solves the problems of insufficient regulation precision, strong subjectivity of parameter regulation, insufficient activation of central-peripheral circuits, and lack of biofeedback in the existing electrical stimulation intervention for motor function rehabilitation.

[0008] In a first aspect, the present invention provides a central-peripheral combined electrical stimulation upper limb hand function modulation system, comprising: The peripheral functional electrical stimulation modulation module is used to independently control the intensity and multi-channel asynchronous timing of peripheral stimulation. The central transcranial electrical stimulation modulation module is used to independently control the stimulation target and intensity of central transcranial electrical stimulation. It also includes a central-peripheral joint regulation module, which is connected to the peripheral functional electrical stimulation regulation module and the central transcranial electrical stimulation regulation module, respectively, and is used to jointly optimize and coordinate the temporal relationship and coupling strength between peripheral stimulation and central stimulation.

[0009] Preferably, the peripheral functional electrical stimulation modulation module includes a functional electrical stimulation triggering module, a peripheral signal decoding module, and a functional electrical stimulation control module; the functional electrical stimulation triggering module is used to generate peripheral electrical stimulation pulse signals, the peripheral signal decoding module is used to collect and process peripheral physiological signals, and the functional electrical stimulation control module is used to dynamically adjust the stimulation intensity and stimulation timing in a closed loop.

[0010] Preferably, the central transcranial electrical stimulation modulation module includes a central electrical stimulation triggering module, a central nervous system intention decoding module, and a central electrical stimulation control module; the central electrical stimulation triggering module is used to generate central electrical stimulation waveforms, the central nervous system intention decoding module is used to collect and process central nervous system signals, and the central electrical stimulation control module is used to dynamically adjust the stimulation target and stimulation intensity in a closed loop.

[0011] Preferably, the functional electrical stimulation control module includes a stimulation intensity control unit and a stimulation timing control unit; the stimulation intensity control unit dynamically adjusts the stimulation amplitude and stimulation frequency based on the comprehensive evaluation results of the action effect; the stimulation timing control unit extracts the muscle synergistic activation law based on multi-channel surface electromyography signals, and constructs the triggering sequence of each stimulation channel accordingly.

[0012] Preferably, the central electrical stimulation control module includes a stimulation target control unit and a stimulation intensity regulation unit; the stimulation target control unit determines individualized stimulation targets based on task-state EEG functional network analysis; and the stimulation intensity regulation unit dynamically adjusts the stimulation intensity based on the evaluation results of the central activation effect of multi-domain fusion.

[0013] Preferably, the central-peripheral joint regulation module includes a central-peripheral coupling assessment unit and a joint optimization regulation unit; the central-peripheral coupling assessment unit is used to calculate the functional coupling strength between central signals and peripheral signals; the joint optimization regulation unit is used to jointly optimize the temporal parameters and phase parameters between central stimuli and peripheral stimuli based on the functional coupling strength.

[0014] Preferably, the central-peripheral coupling assessment unit uses cortical-muscle coherence and functional cortical-muscle coherence as quantitative indicators of the central-peripheral neuromuscular coupling strength; cortical-muscle coherence is used to characterize the degree of functional synchronization between EEG signals and surface electromyography signals in the frequency domain, and functional cortical-muscle coherence is used to characterize the descending driving strength and directionality of the motor cortex to the peripheral muscles.

[0015] Preferably, the joint optimization control unit uses a deep reinforcement learning framework to perform online joint optimization of the adjustment amount of peripheral stimulus trigger delay and the adjustment amount of central stimulus phase; the deep reinforcement learning framework generates action instructions based on the real-time state of the system, and constructs a reward function based on the central-peripheral coupling strength and alignment accuracy to drive the policy network update.

[0016] Secondly, the present invention also provides a method for regulating upper limb hand function through central-peripheral combined electrical stimulation, used in the aforementioned system, characterized by comprising the following steps: S1: Collect multi-channel EEG signals, multi-channel surface electromyography signals, and motor state signals from the subjects, and preprocess the collected signals to remove noise and artifacts; S2: Extract central activation effect assessment features based on processed EEG signals, and assess peripheral motor effects based on processed surface electromyography signals and motor state signals; S3: Based on the evaluation results of peripheral motor effects, the stimulation intensity of peripheral electrical stimulation is adjusted in a closed loop, and the stimulation sequence of each stimulation channel is configured based on the muscle synergy pattern extraction results. S4: Based on the assessment results of the central activation effect, determine the individualized stimulation target and perform closed-loop regulation of the stimulation intensity of central electrical stimulation; S5: Calculate the conduction time difference between central and peripheral signals, and calculate cortical-muscle coherence and functional cortical-muscle coherence to quantify the central-peripheral coupling strength; S6: Based on the central-peripheral coupling strength and conduction time difference, a reinforcement learning algorithm is used to jointly optimize the temporal and phase differences between central and peripheral stimuli online, so that the central and peripheral stimuli are output and applied to the subject according to the optimized temporal and phase parameters.

[0017] Preferably, in step S6, a deep reinforcement learning framework is used to jointly optimize the peripheral stimulus trigger delay adjustment and the central stimulus phase adjustment online. The state vector of the deep reinforcement learning framework includes real-time cortical-muscle coherence, real-time functional cortical-muscle coherence, motor cortex band energy, target muscle group electromyographic amplitude, fatigue index, conduction time difference, current stimulus timing difference, and current stimulus phase difference. The action vector of the deep reinforcement learning framework includes the peripheral stimulus trigger delay adjustment and the central stimulus phase shift adjustment. The reward function of the deep reinforcement learning framework is based on a synergistic gain term, which represents the excess gain of the joint stimulus relative to the individual stimulus.

[0018] Therefore, the central-peripheral combined electrical stimulation upper limb hand function regulation system and method of the present invention, which adopts the above structure, has the following beneficial effects: (1) This invention constructs a three-level hierarchical control architecture, which organically integrates the peripheral functional electrical stimulation control module, the central transcranial electrical stimulation control module and the central-peripheral joint control module, and realizes the synergistic control of central stimulation and peripheral stimulation. It breaks through the technical limitation of the traditional electrical stimulation system where the central and peripheral are independent of each other, effectively activates the central-peripheral neural circuit, and improves the overall effectiveness of rehabilitation training.

[0019] (2) This invention achieves dual closed-loop regulation of peripheral stimulation intensity through cross-round iterative optimization and real-time error correction within rounds by using a composite control structure of iterative learning control and PID controller cascade. At the same time, it realizes the physiological mapping of multi-channel FES asynchronous timing based on muscle synergy mode decomposition, which significantly improves the accuracy of peripheral functional electrical stimulation and the naturalness of movement output.

[0020] (3) This invention achieves individualized stimulus target localization by combining task-state functional network analysis with electric field simulation optimization, and through The multi-domain fusion assessment of oscillatory burst statistical characteristics, motion-related cortical potential characteristics, energy-related characteristics, and fuzzy entropy characteristics, combined with a closed-loop control mechanism of hierarchical role division, significantly improves the spatial targeting accuracy and intensity control effectiveness of central transcranial electrical stimulation.

[0021] (4) This invention constructs a two-layer quantitative system of central-peripheral coupling strength through cortical-muscle coherence and functional cortical-muscle coherence, and uses a deep reinforcement learning framework to perform online joint optimization of peripheral stimulation trigger delay adjustment and central stimulation phase adjustment. For the first time, the quantitative assessment and real-time closed-loop optimization of central-peripheral synergistic coupling have been realized in an electrical stimulation rehabilitation system, filling the gap in the existing technology at the level of synergistic regulation.

[0022] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0023] Figure 1 This is a diagram illustrating the overall architecture of the closed-loop control method of the present invention. Figure 2 This is a schematic diagram of the operation of the peripheral functional electrical stimulation modulation module in the system flowchart of the present invention; Figure 3 This is a schematic diagram of the operation of the central electrical stimulation control module in the system flowchart of the present invention; Figure 4 This is a schematic diagram of the operation of the central-peripheral joint control module in the system flowchart of the present invention; Figure 5 This is a schematic diagram of the stimulation target setting of the electrical stimulation system of the present invention; Figure 6 This is a schematic diagram of the software interface for the electrical stimulation control part of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0026] Example like Figure 1As shown, this invention provides a central-peripheral combined electrical stimulation upper limb hand function regulation system, including a peripheral functional electrical stimulation regulation module, a central transcranial electrical stimulation regulation module, and a central-peripheral combined regulation module. The peripheral functional electrical stimulation regulation module is responsible for the independent regulation of the intensity and multi-channel asynchronous timing of peripheral stimulation, including a functional electrical stimulation trigger module, a peripheral signal decoding module, and a functional electrical stimulation control module. The central electrical stimulation regulation module independently regulates the stimulation target and intensity for central transcranial direct current stimulation, including a central electrical stimulation trigger module, a central nervous system intention decoding module, and a central electrical stimulation control module. The central-peripheral combined regulation module is connected to both the peripheral functional electrical stimulation regulation module and the central electrical stimulation regulation module, and is responsible for the joint optimization and coordinated regulation of the temporal relationship and coupling strength of peripheral and central stimulation. By constructing a three-level hierarchical, double-closed-loop, parameter decoupling and re-coupling central-peripheral coordinated regulation stimulation mechanism, the system comprehensively improves peripheral output stability, central modulation effectiveness, and central-peripheral synergistic efficiency.

[0027] For the peripheral functional electrical stimulation (FES) control module, the FES trigger module consists of a pulse generation circuit unit, a communication unit, and peripheral circuitry. The pulse generation circuit unit uses a BOOST boost circuit to provide voltage support. The BOOST circuit boosts the lithium battery voltage to a stable high voltage, and its output voltage can be adjusted via upper computer software commands to adapt to different stimulation intensity requirements. The required bidirectional pulse current is generated via an H-bridge. All key FES parameters—stimulation pulse width 20-1000μs, stimulation frequency 0-500Hz, stimulation waveform (pulse, triangular wave, sine wave, trapezoidal wave), and stimulation voltage 0-55V—can be precisely set and remotely configured through the upper computer software interface.

[0028] The communication unit integrates the HC05 wireless Bluetooth communication module and transmits commands to the host computer and main control unit through Bluetooth serial port and GPIO serial port respectively. Combined with the communication packet and command design, it enables flexible control of various basic parameters of electrical stimulation from the host computer interface.

[0029] The peripheral signal decoding module consists of a peripheral signal acquisition unit, a peripheral signal processing unit, and a motion effect evaluation unit. The peripheral signal acquisition unit takes existing or real-time data as input as the peripheral signal provider for the overall system. This data includes muscle activation signals and motion state signals. Muscle activation signals can be surface electromyography signals or muscle needle potential signals, etc., while motion state signals are generally motion inertial signals of key joints or limb positions.

[0030] The peripheral signal processing unit is responsible for removing interference components, artifacts, and signal noise from muscle activation signals and motion state signals. In this system, the main noise and interference components in the input peripheral signals are: baseline drift from the acquisition device and location, power frequency interference from the device and environment, high-frequency environmental noise, electrical stimulation pulse artifact interference, and the influence of gravitational acceleration. These noise signals can mask the user's true muscle activation signals and motion inertial signals, further affecting subsequent system implementation. Most of this noise and interference can be removed using a Butterworth filter bank. More importantly, an optimized decorrelation algorithm is used to remove the interference of stimulation pulse artifacts on the muscle activation signal, ultimately yielding a cleaner muscle activation signal. The core formula of the GS algorithm is as follows: ; ; ; In the above formula, For the first A vector of input signals. Let be the dimension of the signal. The initial residual vector, For the first In the decomposition of the first The component and the first The projection coefficients of the basis vectors For the first In the decomposition of the first Each residual vector For the first basis vectors This is the updated residual vector.

[0031] Based on the relatively pure muscle activation signals and motion state signals after processing, the motion effect evaluation unit extracts the corresponding signal features and comprehensively evaluates the motion effect from three dimensions: peripheral motion completion, motion stability, and muscle fatigue.

[0032] Specifically, peripheral movement completion is quantified using the Wilson amplitude of muscle activation signals and the angular change value of joint motion state signals. The Wilson amplitude is calculated as follows: ; ; in For the first The signal amplitude at each sampling point The total number of sampling points. The set amplitude threshold.

[0033] Motion stability is quantified using the sample entropy features of the angular velocity and acceleration signals in the motion state signal. The sample entropy is calculated as follows: ; ; in For power spectral density, Given the input signal sequence, For sequence length, The imaginary unit, For frequency index, For frequency resolution, This is the sample entropy value.

[0034] Muscle fatigue is quantified using the mean power frequency (MPF) and root mean square (RMS) characteristics of muscle activation signals.

[0035] The multi-attribute decision method is used to couple the three-dimensional quantitative evaluation features. The action effect evaluation result value is output by optimizing the superior and inferior solution distance method. This value represents the multi-dimensional similarity between the evaluated action and the optimal action, that is, the degree of approximation with the effect of autonomous action.

[0036] The functional electrical stimulation control module consists of a stimulation intensity control unit, a stimulation timing control unit, and a safety constraint unit. Before stimulation intensity control, the initial use of this system requires testing the basic configuration of peripheral electrical stimulation and performing threshold calibration on each target muscle group to obtain the minimum visible contraction threshold, the comfortable stimulation threshold, and the safety upper limit, thereby establishing an individualized stimulation parameter mapping.

[0037] The stimulus intensity regulation unit outputs a comprehensive evaluation score using the aforementioned multi-attribute decision-making method. Quantitative scores for each dimension and the weighting of each dimension of the evaluation As input to the model, a composite control structure combining iterative learning control and a cascaded PID controller is adopted to achieve dynamic closed-loop adjustment of stimulus amplitude and stimulus frequency.

[0038] In terms of control structure, ILC is responsible for iterative parameter optimization across training rounds, while PID is responsible for real-time error correction within a single round; the two work together. For the... The training cycle is repeated several times, and ILC is based on the previous... Tracking error sequence (in For the target trajectory, (For the actual limb movement trajectory) The feedforward stimulus parameters are updated according to the following Type D ILC law: ; in: ; in For the first Feedforward stimulus parameters of the wheel, For the first Feedforward stimulus parameters of the wheel, To learn the gain matrix, For the first-order difference of the error, For the first In the round The tracking error is mitigated by a monotonically decreasing error constraint across iterations, ensuring iterative convergence. The PID controller then compensates for the residual of the ILC feedforward output in real time within a single iteration. ; in , , These are the proportional, integral, and differential gain coefficients, respectively. This represents the instantaneous trajectory error within the current round. This is the output compensation amount of the PID controller. The final stimulation parameters are determined by the superposition of the feedforward and feedback compensation amounts. After amplitude limiting and rate of change limitation, the output drives the stimulation trigger module, ensuring that after repeated training on a fixed trajectory, the stimulation parameters of each channel are continuously updated iteratively, and the limb movement pattern gradually approaches the target autonomous movement trajectory, achieving dynamic closed-loop regulation with convergence across rounds.

[0039] The stimulation timing regulation unit, based on multi-channel surface electromyography (FES) signals, obtains the temporal coordinated activation pattern between muscle groups during forearm movements through hierarchical signal processing and muscle synergy pattern extraction, and uses it as a physiological reference for multi-channel FES temporal control.

[0040] In the signal preprocessing stage, the raw sEMG signal is sequentially processed by a 20-450Hz fourth-order Butterworth filter bandpass filter, a 50Hz power frequency notch filter, full-wave rectification, root mean square smoothing, and maximum spontaneous contraction normalization to obtain the standardized muscle activation envelope signal for each channel.

[0041] In the time-frequency feature extraction stage, continuous wavelet transform is applied to the activation envelope signal of each channel. Using Morlet wavelet as the basis function, multi-scale decomposition is performed in the frequency range of 1-200Hz to obtain the time-frequency coefficient matrix of each channel. Furthermore, the following dynamic muscle group activation features are extracted from the wavelet coefficients of each segmented signal: energy spectrum, instantaneous power, and band energy ratio. These features are then superimposed with the original normalized activation envelope signal to construct a multi-channel observation matrix. (in For the number of channels, (The number of time sampling points) serves as the input for subsequent collaborative decomposition.

[0042] In the muscle coordination pattern extraction stage, the observation matrix... Apply nonnegative matrix factorization to decompose it into a cooperative pattern matrix. With the temporal activation coefficient matrix The product of: ; in The number of collaborative modes is adaptively determined based on a reconstruction variance contribution rate threshold of 85%. The List of first The relative contribution weights of each muscle channel under each synergistic mode. The OK The temporal activation coefficient curve characterizes this cooperative mode. A multiplicative update rule is used to... and Alternating iterative optimization until convergence, with the iteration terminating when the change in error between two adjacent reconstructions is less than 1. .

[0043] During the timing control mapping phase, the timing activation coefficients of each cooperative mode are used. and their corresponding muscle channel weights Based on this, a trigger timing template for each FES channel is constructed: the muscle channel with the highest weight is based on the co-activation coefficient. When the value exceeds 20% of the peak value, the corresponding FES channel output is triggered. The triggering time for each channel is based on... The weights are sequentially delayed, and the delay step is adaptively determined by the slope of the rising edge of the synergy coefficient. This accurately maps the extracted muscle synergy activation pattern to the asynchronous triggering sequence of the multi-channel FES, driving the target muscle groups to activate sequentially according to the physiological synergy pattern, thereby enhancing the quality of synergistic contraction of the movement-related muscle groups and the naturalness of the movement output.

[0044] In the central electrical stimulation control module, the central electrical stimulation trigger module consists of a waveform generation circuit unit, a communication unit, and surrounding circuits. Its architecture is similar to that of the peripheral functional electrical stimulation trigger module, but it has been specifically adapted to the characteristics of transcranial direct current stimulation in the output parameter domain.

[0045] The waveform generation circuit unit can output a frequency range covering (1-4Hz) (4-8Hz) (8-13Hz) (13-30Hz) and The system displays DC stimulation waveforms in the 30-80Hz frequency band and supports real-time configuration of phase offset parameters. The communication unit uses a low-latency bus protocol to maintain synchronous communication with the host controller, ensuring the coordination and consistency of the stimulation phase and peripheral stimulation timing. Peripheral circuitry includes a current source drive circuit, a real-time electrode impedance monitoring circuit, and an overcurrent protection circuit, providing electrical safety assurance for the stimulation output.

[0046] The central nervous system intention decoding module consists of a multi-channel EEG signal acquisition unit, a central signal processing unit, and a central activation effect assessment unit. The multi-channel EEG signal acquisition unit employs an active dry / wet electrode array with no fewer than 8 channels, deployed according to the international 10-20 standard electrode system, focusing on covering regions related to upper limb motor control such as C3, C4, Cz, FC3, and FC4. The signal acquisition sampling rate is no less than 1000Hz, the hardware bandpass filter range is set to 0.1-200Hz, and the common-mode rejection ratio is no less than 100dB to ensure effective suppression of stimulus artifacts in scenarios with simultaneous electrical stimulation acquisition.

[0047] The central signal processing unit performs two-layer processing on the acquired multi-channel EEG signals. The preprocessing layer completes re-reference (average reference or bipolar reference), 0.1-100Hz bandpass filtering, and online removal of electromyography, electrooculography, and stimulation artifacts using independent component analysis or adaptive filtering methods. The feature extraction layer uses short-time Fourier transform or continuous wavelet transform to achieve time-frequency decomposition and obtain time-varying power spectral density estimates for each frequency band.

[0048] The central activation effect assessment unit comprehensively and quantitatively evaluates the central activation effect of transcranial electrical stimulation from the dimensions of time domain, frequency domain, and nonlinear dynamics, specifically including the following characteristic indicators: (1) Statistical characteristics of oscillatory bursts. Motor cortex. The oscillations are not continuous steady-state oscillations, but rather occur in discrete bursts. Their burst characteristics are closely related to motor control states and the cortical inhibition-excitation balance. The assessment unit is based on Hilbert transform extraction. Instantaneous amplitude envelope in frequency band (15-30Hz) The burst detection threshold is set at 75% of the envelope value exceeding its sliding baseline (window length 2s), enabling real-time identification. Oscillation outbreak events, and statistics The frequency of outbreaks, average duration, and peak intensity of outbreaks are used as characteristics.

[0049] (2) Characteristics of motor-related cortical potentials. Taking the motor triggering event or the onset of tDCS stimulation as the time zero point, relevant time-domain characteristic parameters of motor-related cortical potentials were extracted within the analysis window of [-2000ms, +500ms], including: BP slope (calculating the linear fitting slope of negative slow wave potentials within the window of [-1500ms, -500ms], reflecting the accumulation rate of cortical excitability during the motor preparation period); negative peak amplitude (extracting the maximum negative peak amplitude of MRCP waveform within the window of [-200ms, +100ms], reflecting the peak intensity of cortical activation during the motor execution phase); onset latency (the first moment when the signal amplitude is continuously lower than twice the baseline mean as the MRCP onset latency, quantifying the time sensitivity of the cortex to stimulus response).

[0050] (3) Energy-related characteristics. Bandwidth energy Calculate the target lead in a specific frequency band ( , , The absolute power within the range reflects the frequency-selective enhancement effect of cortical excitability after stimulation, and is calculated using the following formula: ; in Let be the power spectral density function. For the target frequency band boundary, This represents the total power within the target frequency band. Additionally, the band energy ratio characteristics were calculated, specifically the energy ratio of different frequency bands, such as... This is used to evaluate the fine-grained frequency domain contrast of cortical activation and inhibition under intervention.

[0051] (4) Fuzzy entropy. Fuzzy membership functions are introduced to replace the hard threshold judgment in traditional sample entropy, and the nonlinear complexity of EEG signals is robustly measured.

[0052] The central electrical stimulation control module consists of a stimulation target control unit, a stimulation intensity regulation unit, and a safety constraint unit.

[0053] The stimulation target control unit uses a combination of task-state EEG functional network construction and activation hotspot analysis to determine individualized stimulation targets, and completes electrode localization using a universal head model.

[0054] First, task-oriented functional networks were constructed based on task-oriented EEG. Multi-channel EEG signals were simultaneously acquired during the subject's upper limb hand grasping task. The functional connectivity strength between electrodes was calculated based on the preprocessed task-oriented EEG data. Phase-locked values ​​or coherence analysis were used in the motion-related frequency band (…). Constructing a whole-brain functional connectivity matrix (frequency band, 13-30Hz) ,in Number of electrode channels, matrix elements Reflecting the With the Functional connectivity strength between channels. Based on the functional network, the weighted degree centrality of each node is calculated as an evaluation index for cortical activation hotspots: ; in Indicates the first The weighted degree of each electrode node in the functional network during the motion task execution process is considered. A larger value indicates a more significant activation contribution of the node in the motion-related functional network. By combining the weighted degree distribution of multiple repeated tasks, the activation hotspots of the motion cortex are determined by the region where the weighted degree peak node is located. These hotspots are then mapped to the corresponding cortical locations using the electrode coordinates of the international 10-20 system, serving as the target area basis for subsequent electric field simulation and electrode configuration.

[0055] Subsequently, the system calculates the position of the anode on the scalp at different locations based on individualized head models (normalized MNI space or subject MRI-derived models). The predicted distribution of the induced electric field intensity at the location is shown in the figure. Maximizing the normal electric field component is the optimization objective. ; in The candidate scalp area is defined as centered at C3 / C4 with a radius of 3cm. The normal electric field intensity is predicted by the forward model. The target location is in the cortical layer. This is the anode position. This is the optimal anode position. The system will determine the optimal anode position. The output is sent to the operation interface, where the coordinate offset (the offset direction and distance relative to C3 / C4, with an accuracy of 0.5cm) guides the operator to adjust the electrode placement, thereby achieving individualized targeted positioning.

[0056] The stimulus intensity regulation unit establishes a multi-domain fusion closed-loop regulation mechanism with hierarchical role division based on the differences in response characteristics of the aforementioned four types of central activation effects. First, during the system initialization phase, stimuli are applied incrementally from the lowest safe intensity in steps of 0.25 mA, and the normalized sensitivity index of the relative intensity changes of each feature is calculated. (in, For the first Normalization sensitivity index of each feature The total number of intensity increment steps. This represents the change in the normalized value of this feature under two adjacent stimulus intensities. (This refers to the current increment between two adjacent stimulus intensities). Energy-related characteristics and... Due to their high temporal resolution and monotonic intensity response, the statistical characteristics of oscillation bursts are used as the main control features. They are weighted and fused with their respective sensitivity indices as initial weights to construct a composite control index. ; in For composite control indicators, For the first The fusion weights of the main control features at the current moment. This is the real-time normalized value of the feature.

[0057] The PI controller adjusts the current amplitude output in real time, with an adjustment step size not exceeding 0.1mA. While MRCP and fuzzy entropy features have low time resolution and are difficult to use as a basis for real-time control, they are highly sensitive to cortical slow potential plasticity. These are used as calibration features, and the intervention effectiveness score is calculated by comparing the score with the baseline response curve every 2 minutes. When the score is below 0.7, the CCI target value is automatically increased by 5%; when it is above 1.2, it is decreased by 5%, achieving periodic compensation for control target drift caused by fatigue and adaptation effects. The weights of the master control features are updated online every 30 seconds, automatically tilting the weights towards the most sensitive feature at the moment.

[0058] The central-peripheral joint control module is the core of the invention to realize the three-level hierarchical, dual-closed-loop, parameter decoupling and re-coupling control mechanism. It consists of a central-peripheral information acquisition unit, a central activation phase analysis unit, a central-peripheral coupling evaluation unit, a joint optimization control unit, and a safety constraint unit.

[0059] The information acquisition unit is responsible for synchronously acquiring and integrating multimodal physiological signals from the central electrical stimulation modulation module and the peripheral functional electrical stimulation modulation module. Specifically, this includes: multi-channel EEG signals (no less than 8 channels, sampling rate ≥1000Hz), multi-channel surface electromyography signals (no less than 4 channels, covering the target upper limb muscle groups, sampling rate ≥1000Hz), joint angle or hand movement trajectory signals, and the current stimulation parameter status (stimulation frequency, phase, amplitude, and timing) of both modules. The acquisition unit achieves precise time synchronization between EEG and sEMG through hardware trigger signals, with a synchronization error not exceeding 1ms, to ensure the timing consistency of subsequent phase analysis and coupling assessment.

[0060] The central activation phase analysis unit provides a detailed characterization of central activation phase features from a temporal perspective. The unit employs an event-related desynchronization / synchronization analysis method, using motion triggering (or stimulus triggering) as the zero point. Within an analysis window of [-500ms, +1000ms], it detects the conduction time difference between central activation events (the moment when ERD occurs in a specific frequency band) and peripheral electromyographic activation events (the moment when the sEMG envelope exceeds a threshold). : ; in This represents the onset time of peripheral electromyographic activation events. This represents the start time of the central activation event. (Conduction time window) This reflects the functional integrity and efficiency of the corticospinal conduction pathway and serves as a key input parameter for the joint optimization and regulation unit. It optimizes the triggering timing of peripheral FES to match the natural conduction rhythm of central descending commands. During the initialization phase, the system estimates individualized parameters through at least 20 repeated tests. The mean and range of variation are continuously updated during closed-loop operation.

[0061] The central-peripheral coupling assessment unit uses cortical-muscle coherence and functional cortical-muscle coherence as the core quantitative indicators of the strength of central-peripheral neuromuscular coupling. (1) Cortical-muscle coherence (CMC), CMC is defined as the normalized cross power spectrum of EEG signal and sEMG signal in the frequency domain, reflecting the degree of functional synchronization between the cortical motor area and the target muscle at a specific frequency: ; in EEG signal With sEMG signal Cross power spectral density, and Each of them represents its own power spectral density. For frequency The cortex-muscle coherence value at the location. The assessment unit focuses on calculating... The peak value of CMC in the frequency band (15-30Hz) and its corresponding frequency serve as physiological markers reflecting the functional status of the projection pathways of cortical and spinal cord neurons.

[0062] (2) Functional cortical-muscle coherence (FCMC): To overcome the limitations of CMC in linear correlation, the assessment unit further introduces FCMC. Through partial directional coherence or directional coherence analysis based on Granger causality, the descending driving strength and directionality of the cortical motor area to peripheral muscles are quantified: ; in This is the frequency domain representation of the parameter matrix of a multivariate autoregressive (MVAR) model. For from the first The signal to the first The transfer function of a signal. The intensity of the directional information flow of EEG signals to sEMG signals was quantified, which can effectively distinguish between the active drive of the cortex to the muscle and the passive coherent components caused by sensory feedback.

[0063] The CMC and FCMC together form a two-layer quantitative system for central-peripheral coupled evaluation, with their real-time output values ​​serving as the core component of the reward function of the joint optimization control unit.

[0064] The joint optimization and control unit adopts a deep reinforcement learning framework, focusing on online joint closed-loop optimization of the temporal and phase coupling relationships between central transcranial electrical stimulation and peripheral functional electrical stimulation. Based on the independent intensity control of each of the two sub-modules, it achieves precise parameter recoupling for cross-modal stimulation synergy.

[0065] The reinforcement learning framework consists of a state space, an action space, and a reward function. State space is the system state vector. It consists of the following multidimensional features: ; in This represents the real-time cortex-muscle coherence value. This represents the real-time functional cortex-muscle coherence value. This represents the normalized energy value of the β band in region M1. The root mean square normalized value of sEMG for the target muscle group. and These are the brain and muscle fatigue indices, respectively. The phase analysis unit estimates the EEG to electromyographic activation conduction time window in real time. and The timing and phase differences between the central and peripheral stimuli being executed at the moment constitute the agent's self-perceived input to its own control state.

[0066] The agent's action vectors are strictly limited to ,in The adjustment amount for the FES trigger delay relative to the central activation event (step size 5ms, range limited to the individualized conduction time window interval). Inside), This is the offset adjustment of the tDCS phase relative to the intrinsic cortical oscillation (step size 5°, range [-180°, +180°]).

[0067] The reward function uses a cooperative gain term. The core optimization objective is defined as the excess increment of the measured values ​​of CMC and FCMC under the current combined intervention state relative to the predicted values ​​of the linear superposition of the two stimuli acting individually: ; in and To establish a baseline for summation and prediction, an online linear regression model continuously estimates and updates the data based on historical single-modal stimulus data. The weighting coefficients for the FCMC term. This indicates that the current parameter configuration has generated a genuine co-emergent effect. This is supplemented by an instant alignment reward. The matching accuracy between FES trigger delay and individualized conduction time window is quantified. Simultaneously, penalties for exceeding brain and muscle fatigue limits are introduced. ; ; in and These are the safety thresholds for brain and muscle fatigue, respectively. and These represent the penalty values ​​for excessive brain and muscle fatigue, respectively, applying a linear penalty to fatigue states exceeding the safety threshold. The total reward function is: ; in For the total reward, To align the reward discount coefficients, and As fatigue penalty weights, the above parameters are calibrated based on the individual state of the subjects during the system initialization phase and can be dynamically adjusted during the rehabilitation process to adapt to the regulatory goals of different rehabilitation stages.

[0068] The unit employs a proximal policy optimization algorithm to parameterize the action selection strategy, simplifying the action space to a two-dimensional continuous control quantity. The policy network utilizes a lightweight multilayer perceptron structure to reduce online computational overhead. In the offline phase, the system uses simulation data for policy pre-training to establish reasonable prior ranges for temporal and phase differences. In the online phase, individualized fine-tuning is performed, achieving a two-stage learning framework of "general temporal prior + individual phase adaptation." The agent's decision cycle is set to 500ms-1000ms, matching the physiological response timescale of neural modulation. Simultaneously, the unit pre-tests the user's stimulus parameter tolerance threshold, which serves as the system's hardware constraint threshold. No algorithm output control parameters or strategies can exceed this threshold, ensuring safety during use. Furthermore, hardware shielding and other operations minimize system risks.

[0069] like Figure 2-4 As shown, the system flowchart illustrates the operation of the three modules: the peripheral functional electrical stimulation modulation module, the central electrical stimulation modulation module, and the central-peripheral combined modulation module. It demonstrates the individual operation processes of these modules and their data interaction relationships. For example... Figure 5 As shown in the diagram, the target placement of the electrical stimulation system illustrates the target placement method of the central transcranial electrical stimulation electrode in conjunction with the peripheral functional electrical stimulation electrode. For example... Figure 6 As shown, the software interface for the electrical stimulation control section demonstrates the centralized control interface of the host computer software over the parameters of each module in the system.

[0070] This invention also provides a method for modulating upper limb hand function through combined central-peripheral electrical stimulation, comprising the following steps: S1: Start the system, initialize the parameters of each module, set the basic stimulation parameters of the central electrical stimulation control module and the peripheral functional electrical stimulation control module, including stimulation pulse width 20-1000μs, stimulation frequency 0-500Hz, stimulation waveform type and stimulation voltage 0-55V, test the basic configuration of peripheral electrical stimulation, perform threshold calibration on each target muscle group, obtain the minimum visible contraction threshold, the comfortable stimulation threshold and the safety upper limit, and establish individualized stimulation parameter mapping.

[0071] S2: Acquire no less than 8 channels of EEG signals through a multi-channel EEG signal acquisition unit, with a sampling rate of no less than 1000Hz, a hardware bandpass filter range set to 0.1-200Hz, and a common-mode rejection ratio of no less than 100dB. Simultaneously, acquire muscle activation signals and motion state signals through a peripheral signal acquisition unit. The muscle activation signals are surface electromyography signals or muscle needle potential signals, and the motion state signals are motion inertial signals of key joints or limb positions. Achieve precise time synchronization between EEG and sEMG through a hardware trigger signal, with a synchronization error of no more than 1ms.

[0072] S3: The central signal processing unit performs two-layer processing on the acquired multi-channel EEG signals. The preprocessing layer completes re-reference, 0.1-100Hz bandpass filtering, and online removal of electromyography, electrooculography, and stimulation artifacts using independent component analysis or adaptive filtering methods. The feature extraction layer uses short-time Fourier transform or continuous wavelet transform to achieve time-frequency decomposition and obtain time-varying power spectral density estimates for each frequency band. The peripheral signal processing unit removes interference components, artifacts, and signal noise from the muscle activation signal and motion state signal. It removes baseline drift, power frequency interference, high-frequency environmental noise, and gravitational acceleration effects using a Butterworth filter bank, and uses an optimized decorrelation algorithm to remove the interference of stimulation pulse artifacts on the muscle activation signal, obtaining a pure muscle activation signal.

[0073] S4: The central activation effect assessment unit comprehensively and quantitatively assesses the central activation effect of transcranial electrical stimulation from the dimensions of time domain, frequency domain, and nonlinear dynamics, extracting statistical features of β-oscillation bursts, motor-related cortical potential features, frequency band energy and frequency band energy ratio, and fuzzy entropy features; The peripheral movement effect assessment unit extracts corresponding signal features based on the processed pure muscle activation signal and movement state signal, and comprehensively assesses the movement effect from three dimensions: peripheral movement completion, movement stability, and muscle fatigue. It uses a multi-attribute decision method to couple the three-dimensional quantitative evaluation features and outputs the movement effect assessment result value through the optimization of the superior and inferior solution distance method.

[0074] S5: The comprehensive evaluation score output by the stimulus intensity regulation unit using the aforementioned multi-attribute decision-making method. Quantitative scores for each dimension and the weighting of each dimension of the evaluation As the model input, a composite control structure combining iterative learning control and a cascaded PID controller is used to dynamically adjust the stimulus amplitude and frequency in a closed loop; for the th... The training cycle is repeated several times, and ILC is based on the previous... The tracking error sequence updates the feedforward stimulation parameters according to the D-type ILC law. The PID controller compensates for the residual of the ILC feedforward output in real time within a single round. The final stimulation parameters are determined by the superposition of the feedforward quantity and the feedback compensation quantity. The output is limited by amplitude and rate of change before driving the stimulation trigger module.

[0075] S6: The stimulation timing modulation unit, based on multi-channel surface electromyography (sEMG) signals, obtains the temporal coordinated activation patterns among muscle groups during forearm movements through hierarchical signal processing and muscle synergy pattern extraction. The raw sEMG signals are sequentially processed through a 20-450Hz fourth-order Butterworth bandpass filter, a 50Hz power frequency notch filter, full-wave rectification, root mean square smoothing, and maximal voluntary contraction normalization to obtain standardized muscle activation envelope signals for each channel. Continuous wavelet transform is applied to the activation envelope signals of each channel, and multi-scale decomposition is performed using Morlet wavelets as the basis function to obtain the time-frequency coefficient matrix of each channel. Energy spectrum, instantaneous power, and frequency band energy ratio features are extracted to construct a multi-channel observation matrix. ; for the observation matrix Apply nonnegative matrix factorization to decompose it into a cooperative pattern matrix. With the temporal activation coefficient matrix The product is updated using the multiplication rule. and Alternate iterations are performed until convergence. Based on the temporal activation coefficients of each synergistic mode and their corresponding muscle channel weights, a triggering temporal template for each FES channel is constructed to drive the target muscle group to activate sequentially according to the physiological synergistic mode.

[0076] S7: The central stimulation target control unit uses a combination of task-based EEG functional network construction and activation hotspot analysis to determine individualized stimulation targets: the functional connectivity strength between electrodes is calculated based on task-based EEG data. Frequency bands construct a whole-brain functional connectivity matrix Calculate the weighted degree centrality of each node. The activation hotspots of the motor cortex are determined by the region where the weighted peak node is located; the predicted distribution of induced electric field intensity at different positions on the scalp is calculated based on an individualized head model, with the maximization of the normal electric field component as the optimization objective. The optimal anode position is output to the operation interface to guide electrode placement.

[0077] S8: The central stimulation intensity regulation unit establishes a multi-domain integrated closed-loop regulation mechanism with hierarchical role division: Stimulus is applied incrementally from the lowest safe intensity with a step size of 0.25mA, and the normalized sensitivity index of the relative intensity change of each feature is calculated, based on energy-related features and... The statistical characteristics of oscillation bursts are weighted and fused with the main controlling characteristics to construct a composite control index. The PI controller is driven to adjust the current amplitude output in real time. Using MRCP features and fuzzy entropy features as calibration features, the intervention effectiveness score is calculated by comparing with the benchmark response curve every 2 minutes. The control target is periodically compensated, and the weight of the main control features is updated online every 30 seconds.

[0078] S9: The central-peripheral information acquisition unit of the central-peripheral joint regulation module synchronously collects and integrates multimodal physiological signals from the central electrical stimulation regulation module and the peripheral functional electrical stimulation regulation module; the central activation phase analysis unit uses the event-related desynchronization / synchronization analysis method, with motion triggering or stimulus triggering as the time zero point, and detects the conduction time difference between central activation events and peripheral electromyographic activation events within the analysis window of [-500ms, +1000ms]. The central-peripheral coupling assessment unit calculates cortical-muscle coherence and functional cortical-muscle coherence as core quantitative indicators of central-peripheral neuromuscular coupling strength. The joint optimization and control unit employs a deep reinforcement learning framework to perform online joint closed-loop optimization of the temporal and phase coupling relationships between central transcranial electrical stimulation and peripheral functional electrical stimulation. The state vector is: ; The action vector is: ; The reward function is: ; Among them, the cooperative gain term The policy is updated using a near-end policy optimization algorithm, and the agent's decision cycle is set to 500ms-1000ms.

[0079] S10: The safety constraint unit monitors all stimulation parameters to ensure that the control parameters and strategies output by all algorithms do not exceed the user's tolerance threshold for stimulation parameters in the pre-test. At the same time, it shields the operation and control system risks through hardware, outputs stimulation parameters and applies them to the subject to complete a single combined electrical stimulation modulation cycle.

[0080] Working Principle: After system startup, the central electrical stimulation modulation module and the peripheral functional electrical stimulation modulation module acquire EEG signals and sEMG / motor state signals, respectively. After signal processing and feature extraction, their respective modulation units independently complete the closed-loop regulation of stimulation intensity and target points. Based on this, the central-peripheral joint modulation module, through central activation phase analysis, central-peripheral coupling assessment, and deep reinforcement learning-driven joint optimization, adjusts the temporal and phase differences between central and peripheral stimuli in real time, achieving re-coupling of the synergistic parameters of the two stimuli. After verification by the safety constraint unit, the optimized stimulation parameters are output to the stimulation triggering module and applied to the subject. Simultaneously, each module continuously acquires physiological signals to enter the next modulation cycle, forming a three-level, hierarchical, dual-closed-loop synergistic modulation mechanism.

[0081] Therefore, the present invention provides a central-peripheral combined electrical stimulation upper limb hand function regulation system and method with the above-mentioned structure. Through a three-level hierarchical regulation architecture, a dual closed-loop feedback mechanism, and a synergistic strategy of parameter decoupling and re-coupling, it achieves precise synergistic activation of the central-peripheral neural circuit, effectively solving the technical problems of insufficient regulation precision, lack of synergy in central-peripheral circuit activation, and lack of biofeedback in traditional electrical stimulation systems.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A central-peripheral combined electrical stimulation system for upper limb hand function regulation, characterized in that, The system employs a hierarchical decoupling and recoupling control architecture consisting of a peripheral stimulation closed loop, a central stimulation closed loop, and a central-peripheral joint optimization closed loop located above both. The peripheral functional electrical stimulation modulation module includes a functional electrical stimulation triggering module, a peripheral signal decoding module, and a functional electrical stimulation control module. The functional electrical stimulation control module includes a stimulation intensity modulation unit and a stimulation timing modulation unit, which are used to evaluate the peripheral action effect based on multi-channel surface electromyography signals and motion state signals. It adopts a composite control structure with cascaded controllers to perform closed-loop regulation of stimulation amplitude and stimulation frequency through cross-round iterative optimization and real-time error correction within rounds. It also determines the initial triggering sequence of each functional electrical stimulation channel according to the muscle co-activation sequence to form the peripheral stimulation closed loop. The central transcranial electrical stimulation modulation module includes a central electrical stimulation triggering module, a central nervous system intention decoding module, and a central electrical stimulation control module. The central electrical stimulation control module includes a stimulation target control unit and a stimulation intensity regulation unit. The stimulation target control unit is used to determine individualized stimulation targets based on task-state EEG functional network analysis. The stimulation intensity regulation unit is used to dynamically adjust the stimulation intensity in a closed loop based on the evaluation results of the central activation effect of multi-domain fusion. The system also includes a central-peripheral joint regulation module, which is connected to both the peripheral functional electrical stimulation regulation module and the central transcranial electrical stimulation regulation module. The central-peripheral joint regulation module comprises a central-peripheral coupling assessment unit and a joint optimization regulation unit. The central-peripheral coupling assessment unit calculates cortical-muscle coherence and functional cortical-muscle coherence to quantify the central-peripheral coupling strength. The joint optimization regulation unit uses a deep reinforcement learning framework to perform online joint optimization of the peripheral stimulation trigger delay adjustment and the central stimulation phase adjustment based on the central-peripheral coupling strength and conduction time difference, thereby achieving coordinated central-peripheral regulation. The peripheral stimulation closed loop and the central stimulation closed loop respectively complete the independent closed-loop regulation of stimulation intensity. The central-peripheral joint optimization closed loop completes the joint regulation of the timing and phase of the two stimulations under the constraint of the stimulation intensity output by the two, thereby forming a three-level closed-loop regulation mechanism in which the stimulation intensity parameter is first decoupled and regulated, and the stimulation coordination parameter is then coupled and optimized.

2. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the stimulus intensity regulation unit, for the first The training cycle is repeated several times, and the iterative learning control is based on the previous... Tracking error sequence according to Iterative learning control law updates feedforward stimulus parameters: ； in , To learn the gain matrix.

3. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the stimulus intensity control unit, the PID controller performs real-time compensation on the residual of the iterative learning control feedforward output within a single round, with the compensation amount being... ； wherein , , are proportional, integral, derivative gain coefficients, respectively, is the instantaneous trajectory error within the current round; the final stimulation parameter is determined by superimposing the feedforward amount and the feedback compensation amount.

4. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the stimulation timing control unit, a multi-channel observation matrix is ​​constructed after preprocessing the multi-channel surface electromyography signals. , For the number of channels, This represents the number of time sampling points; Apply nonnegative matrix decomposition to the observation matrix to decompose it into a cooperative mode matrix. With the temporal activation coefficient matrix The product of: ； in This represents the number of collaborative modes; The temporal activation coefficients of each collaborative mode and their corresponding muscle channel weights Based on this, the muscle channel with the highest weight has the highest temporal activation coefficient. When the threshold percentage is exceeded, the corresponding functional electrical stimulation channel output is triggered. The triggering time of each channel is delayed sequentially according to its weight, and the delay step is adaptively determined by the rising slope of the synergy coefficient.

5. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the stimulation target control unit, task-state EEG data is used in... Frequency bands construct a whole-brain functional connectivity matrix Calculate the weighted degree centrality of each node: ； The activation hotspots of the motor cortex are determined by the region where the weighted peak node is located; The distribution of induced electric field intensity at different locations on the scalp was calculated based on an individualized head model, with the target dermal point as the primary focus. Maximizing the normal electric field component is the optimization objective. ； in Candidate areas for scalp The normal electric field intensity is predicted by the forward model. This is the anode position. The optimal anode position is determined and output to the operating interface to guide electrode placement.

6. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the central-peripheral coupling assessment unit, cortical-muscle coherence Defined as: ； in EEG signals With surface electromyography signals Cross power spectral density, and Each has its own power spectral density, and the calculations are focused on this. Frequency band cortical-muscle coherence peaks and their corresponding frequencies; functional cortical-muscle coherence Defined as: ； in This is the frequency domain representation of the parameter matrix of a multivariate autoregressive model. For from the first The signal to the first The transfer function of the signal is used to quantify the downward driving strength and directionality of the cortical motor area to the peripheral muscles.

7. The central-peripheral combined electrical stimulation upper limb hand function regulation system according to claim 1, characterized in that: In the joint optimization control unit, the state vector of the deep reinforcement learning framework Defined as: ； in This represents the real-time cortex-muscle coherence value. This represents the real-time functional cortex-muscle coherence value. M1 area Normalized value of frequency band energy The root mean square normalized value of surface electromyography of the target muscle group. and These are the brain and muscle fatigue indices, respectively. For the conduction time difference, and The timing difference and phase difference of the current stimulus; Action vectors ,in This is the adjustment amount for the trigger delay of peripheral stimulation. Central stimulus phase shift adjustment; reward function for: ； Among them, the cooperative gain term ； and The predicted value is the linear superposition of the two stimuli acting individually. These are the weighting coefficients. To align the reward discount coefficients, and As fatigue penalty weight, and These are penalties for exceeding limits in terms of brain and muscle fatigue.

8. A method for modulating upper limb hand function through central-peripheral combined electrical stimulation, applied to the central-peripheral combined electrical stimulation upper limb hand function modulation system according to any one of claims 1-7, characterized in that, Includes the following steps: S1: Simultaneously acquire multi-channel EEG signals from the motor cortex, multi-channel surface electromyography signals from the target muscle group, and upper limb motor state signals, and obtain the current parameters of functional electrical stimulation and transcranial direct current stimulation, and preprocess the central physiological signals and peripheral physiological and motor signals; S2: Determine the evaluation results of peripheral movement effect based on the preprocessed electromyographic signals and motor state signals, and independently adjust the amplitude and frequency of functional electrical stimulation accordingly to form a closed loop of peripheral stimulation intensity; at the same time, determine the muscle synergistic activation sequence based on the decomposition of multi-channel electromyographic signals, and configure the initial triggering sequence of functional electrical stimulation channels. S3: Based on the preprocessed EEG signals, determine the individualized stimulation target and the evaluation results of the central activation effect, and adjust the intensity of transcranial direct current stimulation independently to form a closed loop of central stimulation intensity. S4: Calculate cortical-muscle coherence and functional cortical-muscle coherence based on EEG and EMG signals, normalize and weight them to obtain the central-peripheral joint coupling evaluation quantity; S5: Determine the central activation initiation time based on EEG signals, determine the peripheral electromyographic activation initiation time based on EMG signals, calculate the time difference between the two to obtain the conduction time difference, and determine the current stimulation sequence difference and phase difference between functional electrical stimulation and transcranial direct current stimulation; Wherein, the stimulation phase difference is the relative phase difference between the transcranial direct current stimulation output enable moment or time window and the functional electrical stimulation trigger moment in the same training cycle; S6: Based on the closed loop of peripheral and central stimulus intensity, the central-peripheral joint coupling evaluation quantity, conduction time difference, current stimulus timing difference and phase difference are used as the state input of deep reinforcement learning, which generate the peripheral stimulus triggering delay adjustment quantity and the central stimulus phase adjustment quantity. S7: The peripheral stimulation trigger delay adjustment is superimposed on the initial trigger timing to obtain the updated functional electrical stimulation trigger timing, and the updated transcranial direct current stimulation output enable time or time window is determined according to the central stimulation phase adjustment. S8: The updated functional electrical stimulation trigger timing is sent to the peripheral control module, and the updated transcranial direct current stimulation output timing or time window is sent to the central control module. The closed loop of the two types of stimulation intensity is kept unchanged, forming a central-peripheral timing and time-series joint optimization closed loop. The adjustment amounts of both types are limited to the preset safety range.

9. A method for regulating upper limb hand function through central-peripheral combined electrical stimulation according to claim 8, characterized in that: In step S5, the formula for calculating the conduction time difference is: ,in This represents the onset time of peripheral electromyographic activation events. The starting point of the central activation event; cortical-muscle coherence and functional cortical-muscle coherence are calculated using formulas.

10. A method for regulating upper limb hand function through central-peripheral combined electrical stimulation according to claim 8, characterized in that: In step S6, the near-end policy optimization algorithm is used to update the policy, and the agent's decision cycle is set to 500ms to 1000ms.

Images