Closed-loop electrical stimulation control system

Abstract

The application discloses a closed-loop electrical stimulation control system, and relates to the field of neural signals.The system comprises a signal processing module, a prediction control module, a safe output control module and an electrical stimulation output module.The signal processing module determines a movement intention trigger signal and a movement intention confidence level based on an electroencephalogram signal, and determines an induced contraction degree, an agonist force index, a co-activation index and an artifact identification based on a mechanomyographic signal of a target muscle group.The prediction control module determines an induced contraction degree prediction sequence and an agonist force index prediction sequence based on a historical pulse width, a historical induced contraction degree and a historical agonist force index, and determines a candidate pulse width in combination with a target contraction degree, a movement intention state, a co-activation state and an agonist force safety boundary.The safe output control module discriminates the candidate pulse width, and the electrical stimulation output module outputs an electrical stimulation signal according to a final pulse width obtained through the discrimination.The application can improve the adaptability and safety of electrical stimulation output adjustment.

Description

Technical Field

[0001] This application relates to the field of neural signal technology, and in particular to a closed-loop electrical stimulation control system. Background Technology

[0002] Brain-computer interface-functional electrical stimulation (BCI-FES) systems typically acquire the subject's brain signals through electroencephalography (EEG), identify the brain signals, and obtain recognition results related to motor intentions; when trigger conditions are met, the functional electrical stimulation device outputs electrical stimulation to the target muscle group.

[0003] In practical applications, different subjects respond differently to electrical stimulation at different training stages, and the response of the same subject may also change during continuous training. Existing closed-loop electrical stimulation systems mostly use preset parameters and fixed thresholds to control the electrical stimulation output, which has limited adaptability to changes in state during training. The adjustment effect and safety control performance of electrical stimulation parameters still need to be improved.

[0004] Therefore, there is an urgent need for a closed-loop electrical stimulation control system that can improve the adaptability and safety of electrical stimulation output regulation. Summary of the Invention

[0005] This application provides a closed-loop electrical stimulation control system to address the technical problems that the electrical stimulation output is difficult to adapt to the continuous changes in the state of motor intention and muscle response during closed-loop electrical stimulation, and that the forward-looking adjustment capability and safety control performance of the electrical stimulation output need to be improved.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] A closed-loop electrical stimulation control system includes a signal processing module, a predictive control module, a safety output control module, and an electrical stimulation output module. The signal processing module is configured to: acquire electroencephalogram (EEG) signals and mechanomotor signals corresponding to a target muscle group; determine a motor intention trigger signal and a motor intention confidence level based on the EEG signals; and determine the induced contraction degree, agonist muscle production index, coactivation index, and artifact markers based on the mechanomotor signals. The predictive control module is configured to: acquire historical control data, including historical pulse width, historical induced contraction degree, and historical agonist muscle production index; determine an induced contraction degree prediction sequence based on the historical pulse width and the historical induced contraction degree; determine an agonist muscle production index prediction sequence based on the historical pulse width and the historical agonist muscle production index; and determine the induced contraction degree prediction sequence based on the motor intention confidence level and the coactivation index. The system determines the induced contraction tracking adjustment parameters; based on the motor intention confidence level, it determines the agonist muscle production safety boundary; based on the induced contraction degree prediction sequence, the agonist muscle production index prediction sequence, the pre-determined target contraction degree, the induced contraction tracking adjustment parameters, and the agonist muscle production safety boundary, it performs constrained optimization processing to determine candidate pulse widths; the safety output control module is configured to: determine the execution trigger signal and the final pulse width based on the motor intention trigger signal, the motor intention confidence level, the agonist muscle production index, the co-activation index, the artifact identifier, and the candidate pulse widths; the electrical stimulation output module is configured to: in response to the execution trigger signal, drive the electrical stimulation electrodes to output an electrical stimulation signal according to the final pulse width; wherein, the historical control data is determined based on the output electrical stimulation signal and the corresponding mechanograph signal.

[0008] In one of the above implementations, the signal processing module includes an EEG acquisition unit, an EEG decoding unit, and a mechanical response acquisition unit; wherein, the EEG acquisition unit is configured to: acquire the EEG signal; the EEG decoding unit is configured to: decode the EEG signal to determine the motor intention trigger signal and the motor intention confidence level; the mechanical response acquisition unit includes mechanical response acquisition units respectively corresponding to the agonist and antagonist muscles of the target muscle group, and the mechanical response acquisition units are configured to: acquire the mechanical muscle animation signal including muscle deformation response data and vibration response data.

[0009] In one of the above implementations, the signal processing module further includes a response state construction unit, which is configured to: perform multi-source response state estimation and abnormal data processing on the mechanograph signals corresponding to the agonist muscle and the antagonist muscle respectively to obtain the agonist muscle response state and the antagonist muscle response state; determine the induced contraction degree based on the agonist muscle response state; determine the stimulus-contraction switching capability based on the historical control data; determine the agonist muscle force index based on the stimulus-contraction switching capability, the spectral features and contraction timing features contained in the agonist muscle response state; and determine the co-activation index and the artifact identifier based on the agonist muscle response state and the antagonist muscle response state.

[0010] In one of the above implementations, a calibration control module is further included. The calibration control module is configured to: control the electrical stimulation output module to output a calibration stimulation sequence within a preset output limit; determine the induced maximum contraction and fresh induced response baseline based on the agonist muscle mechanical response corresponding to the calibration stimulation sequence; and determine the target contraction degree based on the induced maximum contraction and a preset target intensity coefficient.

[0011] In one of the above implementations, the predictive control module includes an induced contraction state prediction unit, which is configured to: obtain historical pulse width and historical induced contraction degree from the historical control data; update the induced contraction response model online based on the historical pulse width and the historical induced contraction degree; and determine the induced contraction degree prediction sequence based on the updated induced contraction response model.

[0012] In one of the above implementations, the predictive control module further includes an active muscle force state prediction unit, which is configured to: obtain historical active muscle force index and historical pulse width from the historical control data; update the active muscle force response model online based on the historical active muscle force index and the historical pulse width; and determine the active muscle force index prediction sequence based on the updated active muscle force response model.

[0013] In one of the above implementations, the predictive control module includes an optimization control unit, which is configured to: determine the induced contraction tracking adjustment parameters based on the motion intention confidence, the co-activation index, and the time weight of the corresponding prediction time in the current control cycle; determine the agonist muscle production safety boundary based on the motion intention confidence in the current control cycle; construct an optimization target, which is used to reduce the tracking deviation, pulse width variation, and safety state relaxation of the induced contraction degree prediction sequence relative to the target contraction degree; and perform the constrained optimization process to determine the candidate pulse width under the constraints that the candidate pulse width meets a preset pulse width value range, the pulse width variation meets a preset variation range, and the agonist muscle production index prediction sequence is not lower than the boundary of the agonist muscle production safety boundary after adjustment by the safety state relaxation.

[0014] In one of the above implementations, the safety output control module includes a timing coupling error determination unit and an output permission discrimination unit. The timing coupling error determination unit is configured to: perform timing alignment of the motion intention trigger signal and the electrical stimulation signal based on a common time reference, and determine the timing coupling error. The output permission discrimination unit is configured to: set the execution trigger signal to the execution state and determine the candidate pulse width as the final pulse width when all admission conditions are met. The admission conditions include: the motion intention trigger signal indicates triggering; the motion intention confidence level is not lower than a preset confidence threshold; the agonist muscle production index is not lower than the agonist muscle production safety boundary; the artifact identifier indicates no artifacts; the co-activation index is lower than a preset co-activation threshold; and the timing coupling error is not greater than a preset timing threshold.

[0015] In one of the above implementations, the safety output control module further includes an output termination control unit, which is configured to: set the execution trigger signal to a non-execution state and set the final pulse width to zero when any of the termination conditions are met; the termination conditions include: the agonist muscle force index is continuously lower than a preset force termination threshold; the stimulation contraction switching ability determined based on the historical control data is continuously lower than a preset switching ability termination threshold; within a continuous control cycle, the final pulse width continuously reaches a preset pulse width limit, and the induced contraction degree continuously fails to reach the target contraction degree; the cumulative number of electrical stimulation outputs reaches a preset number threshold; and the cumulative actual output duration reaches a preset duration threshold.

[0016] In one of the above implementations, the safety output control module further includes a hard interrupt unit, which is configured to: set the execution trigger signal to a non-execution state and set the final pulse width to zero when any of the hard interrupt conditions are met; the hard interrupt conditions include: the artifact identifier continuously indicates the presence of artifacts; the co-activation index is not lower than a preset co-activation interrupt threshold; abnormal acquisition of mechanical muscle animation signal is detected; abnormal data transmission link is detected.

[0017] Compared with the prior art, this application has the following beneficial effects:

[0018] 1. This application uses electroencephalogram (EEG) signals to generate motor intention states, and uses mechanomotor signals of target muscle groups to generate induced contraction states, agonist muscle production states, and bi-muscle synergistic states. These states are then used in candidate pulse width calculation and output admission discrimination. Therefore, the electrical stimulation output can simultaneously reflect central motor intention and peripheral muscle response.

[0019] 2. This application utilizes historical pulse width and historical induced contraction degree, and historical pulse width and historical agonist muscle production index to form two prediction sequences, respectively, and makes the induced contraction degree track the target contraction degree in the same constrained optimization process. This allows for prospective assessment of the contraction response and agonist muscle production state after the application of candidate pulse widths before the actual output of electrical stimulation, improving the adaptability of electrical stimulation parameters to response changes during training. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, wherein:

[0021] Figure 1 This is a schematic diagram of the system architecture of a closed-loop electrical stimulation control system provided in one embodiment of this application;

[0022] Figure 2 This is a schematic diagram of the structure of a signal processing module provided in one embodiment of this application;

[0023] Figure 3 This is a schematic diagram of the predictive control module provided in one embodiment of this application;

[0024] Figure 4 This is a schematic diagram of the structure of a safety output control module provided in one embodiment of this application;

[0025] Figure 5 This is a schematic diagram of the operation flow of a closed-loop electrical stimulation control system provided in one embodiment of this application.

[0026] In the picture:

[0027] 100 - Signal processing module, 101 - EEG acquisition unit, 102 - EEG decoding unit, 103 - Mechanical response acquisition unit, 104 - Response state construction unit;

[0028] 200-Predictive control module, 201-Active muscle force state prediction unit, 202-Optimization control unit, 203-Induced contraction state prediction unit;

[0029] 300-Safety output control module, 301-Timing coupling error determination unit, 302-Output enable judgment unit, 303-Output termination control unit, 304-Hard interrupt unit;

[0030] 400-Electrical Stimulation Output Module;

[0031] 500-calibration control module;

[0032] 600 - Timing Synchronization and Storage Module;

[0033] 700 - Training Effective Quality Scoring Module. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0035] The technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0036] This solution does not aim to obtain disease diagnosis results or health status, but only to process the neural signals of the target user and thereby achieve adaptive closed-loop training. All steps are information processing methods implemented by computers and other devices.

[0037] It should be fully understood that the target user information involved in this application (including but not limited to the target user's EEG and EMG signals) is information and data authorized by the target user or fully authorized by all parties. The use of the target user information shall comply with the privacy policies and practices of the industry that are generally considered to meet or exceed the privacy of the target user. The collection, use and processing of related data shall comply with relevant laws, regulations and standards, and provide corresponding operation access points for the target user to choose to authorize or refuse.

[0038] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0039] This application is based on the Hebbian plasticity principle: when central motor intention and peripheral muscle contraction are temporally synchronized, the synaptic connections between the motor cortex and peripheral pathways are strengthened and remodeled. The effect of synchronous coupling in the Hebbian plasticity principle is called the effective Hebbian gain. Since the number of stimuli and the duration of output in a single training session are limited, this application modulates induced contraction tracking through motor intention confidence and co-activation index, and determines candidate pulse widths under safety constraints to ensure that the allowed output electrical stimulation forms effective coupling as much as possible, thereby maximizing the effective Hebbian gain.

[0040] Example 1:

[0041] This embodiment provides a closed-loop electrical stimulation control system.

[0042] like Figure 1 As shown, the system includes a signal processing module 100, a predictive control module 200, a safety output control module 300, and an electrical stimulation output module 400. These modules can be deployed in the same computing device or deployed separately and exchange data via wired or wireless data transmission links.

[0043] In some embodiments, the signal processing module 100 is configured to: acquire electroencephalogram (EEG) signals and mechanomotor signals corresponding to the target muscle group; determine a motor intention trigger signal and a motor intention confidence level based on the EEG signals; and determine the induced contraction degree, voluntary muscle force index, coactivation index, and artifact identifier based on the mechanomotor signals. The prediction control module 200 is configured to: acquire historical control data, including historical pulse width, historical induced contraction degree, and historical voluntary muscle force index; determine an induced contraction degree prediction sequence based on the historical pulse width and historical induced contraction degree; determine a voluntary muscle force index prediction sequence based on the historical pulse width and historical voluntary muscle force index; determine induced contraction tracking adjustment parameters based on the motor intention confidence level and the coactivation index; determine a voluntary muscle force safety boundary based on the motor intention confidence level; and perform constrained optimization processing to determine candidate pulse widths based on the induced contraction degree prediction sequence, the voluntary muscle force index prediction sequence, the pre-determined target contraction degree, the induced contraction tracking adjustment parameters, and the voluntary muscle force safety boundary. The safety output control module 300 is configured to determine the execution trigger signal and the final pulse width based on the motor intention trigger signal, motor intention confidence, active muscle production index, co-activation index, artifact identifier, and candidate pulse width. The electrical stimulation output module 400 is configured to drive the electrical stimulation electrodes to output an electrical stimulation signal according to the final pulse width in response to the execution trigger signal. Historical control data is determined based on the output electrical stimulation signals and the corresponding mechanomotor signals.

[0044] In some embodiments, the signal processing module 100 is located on the subject's side, the prediction control module 200 and the safety output control module 300 are deployed on a host computer, and the electrical stimulation output module 400 is connected to the electrical stimulation electrodes located above the target muscle group's movement point. The host computer receives EEG signals and mechanomotor signals respectively, forms candidate pulse widths and performs safety discrimination, and then sends the final pulse width that is allowed to be output to the electrical stimulation output module 400.

[0045] In some embodiments, the electrical stimulation output module 400 can be connected to a host computer via a universal serial bus or a serial communication interface, and the electrical stimulation electrodes can be attached above the motor points corresponding to the agonist muscles. The host computer sends an execution trigger signal and a final pulse width to the electrical stimulation output module 400. When the execution trigger signal is in the execution state, the electrical stimulation output module combines the pre-configured electrical stimulation frequency and electrical stimulation amplitude to form an electrical stimulation signal.

[0046] In another embodiment, the system can also be configured with a parameter configuration interface and a manual emergency stop interface. The parameter configuration interface is used to write control parameters such as the target muscle group, the confidence threshold of the movement intention, the duration of a single stimulation, the maximum number of outputs, and the predicted time domain length before the closed-loop control begins. The manual emergency stop interface is connected to the safety output control module to disable electrical stimulation output when an emergency stop command is received.

[0047] The following uses discrete control steps Explain the data transfer relationships in the system. This indicates the sequence number of the discrete control step. The discrete control step is defined as a single brain-computer interface-functional electrical stimulation closed-loop trigger. The confidence level of the motor intention is denoted as... The degree of induced contraction is denoted as ; in the Historical control data generated before each discrete control step is executed includes historical pulse widths. Historically induced contraction degree and historical active muscle production index The predictive control module reads the corresponding data from historical control data according to the required data combination.

[0048] In some embodiments, such as Figure 1 and Figure 2 As shown, the signal processing module 100 includes an EEG acquisition unit 101, an EEG decoding unit 102, a mechanical response acquisition unit 103, and a response state construction unit 104. The EEG acquisition unit 101 acquires EEG signals through EEG acquisition electrodes placed on the subject's head. As a specific implementation, the EEG acquisition unit 101 can employ a sixteen-channel dry electrode EEG acquisition device, with each channel written with a timestamp generated by the same timing source. The EEG decoding unit 102 sequentially performs preprocessing, spatial feature extraction, and intent classification on the EEG signals to obtain the motor intent trigger signal. and confidence level of movement intention .in, Used to indicate whether a motion intent corresponding to the target motion has been detected; the value range can be... ; The value range used to characterize the reliability of the motion intent recognition result can be: .

[0049] In some embodiments, the EEG acquisition unit 101 acquires EEG signals at a sampling rate of 250Hz and transmits the EEG signals to the EEG decoding unit via a universal serial bus or wireless communication link. Upon initial use, twenty sets of target motor imagery data can be acquired, each lasting 4 seconds with a 6-second interval between adjacent acquisitions. The acquired data is used to determine individualized spatial filtering parameters and intent classification parameters. Before formal training begins, three sets of verification motor imagery data can be acquired, each lasting 10 seconds. If the recognition accuracy corresponding to the verification motor imagery data is lower than a preset accuracy threshold, the target motor imagery data is reacquired, and the EEG decoding parameters are updated. Thus, the motor intent trigger signal and the motor intent confidence level are determined by the EEG decoding parameters that have been verified during this training.

[0050] In some embodiments, the mechanical response acquisition unit 103 includes mechanical response acquisition sections respectively corresponding to the agonist and antagonist muscles. Each mechanical response acquisition section acquires muscle deformation response data and vibration response data. Each mechanical response acquisition section includes a muscle deformation response acquisition path, a low-frequency vibration response acquisition path, and a high-frequency vibration response acquisition path, thereby forming six channels of raw mechanical response data on the agonist and antagonist muscle sides. It should be noted that the six channels of raw mechanical response data are only one implementation of multi-source acquisition; different numbers of response acquisition paths can be used when muscle deformation response data and vibration response data can be formed.

[0051] In some embodiments, the mechanical response acquisition parts corresponding to the agonist and antagonist muscles are respectively attached to the muscle belly of the two muscles, and are kept in contact with the skin by elastic fixation members. The contact pressure can be adjusted to a preset pressure range during calibration, for example, to 0.5. Each mechanical response acquisition unit is connected to the data acquisition interface of the host computer via shielded cables to reduce external interference during transmission. The muscle deformation response acquisition circuit is used to obtain the induced contraction degree and contraction timing characteristics, while the low-frequency vibration response acquisition circuit and the high-frequency vibration response acquisition circuit are used to obtain the power characteristics of different vibration frequency bands.

[0052] The EEG acquisition unit 101, the mechanical response acquisition unit 103, and the electrical stimulation output module 400 use a common time reference. This common time reference can be provided by the same hardware clock or by a unified timestamp rule. The EEG decoding unit 102 outputs a motor intention trigger signal. and confidence level of movement intention The corresponding motion intention time information is retained. When the electrical stimulation output module 400 generates an electrical stimulation signal, it records the stimulation output time information so that the safety output control module 300 can calculate the timing coupling error between the two.

[0053] The response state construction unit 104 first performs bandpass filtering and outlier processing on the mechanomotor signals corresponding to the agonist and antagonist muscles, and then performs state estimation on the multi-source response data of the same muscle. Outlier processing can use statistical outlier detection to identify observations that significantly deviate from the same data window; for example, isolated outlier observations can be removed based on the Grubbs criterion. For the outlier-processed observation data, the response state construction unit can perform multi-source response state estimation using the following state equation and observation equation:

[0054]

[0055] In the formula, This represents the estimated mechanical response state. This represents an observation vector composed of multiple response data corresponding to the same muscle. Represents the state transition matrix. Represents the observation matrix. Indicates process noise. Represents observation noise. Process noise covariance. and observation noise covariance It can be determined from resting data within a preset time period before training begins, for example, by initializing the sample covariance of ten seconds of resting data.

[0056] After performing state estimation based on the above state equation and observation equation on the agonist and antagonist muscles, the agonist muscle response state and the antagonist muscle response state are obtained. The two then proceed to the determination process of induced contraction degree, agonist muscle force index, co-activation index and artifact marker.

[0057] To identify overall anomalies in the acquisition channels during the fusion process, the response state construction unit 104 constructs a signal consistency index based on the innovation of the state estimate and the innovation covariance:

[0058]

[0059] in,

[0060]

[0061]

[0062] In the formula, This indicates the information between current observations and observational predictions; Indicates the new information covariance; This represents the predicted response state value prior to the arrival of the current observation; Indicates the corresponding state prediction covariance; superscript The superscript represents the matrix transpose operation. This represents the inverse operation of a matrix; This indicates the signal consistency index.

[0063] when When the consistency threshold corresponding to the observed degrees of freedom is exceeded, the response state construction unit determines that the mechanical muscle motion signal acquisition is abnormal, and can then... Larger information components are used to locate abnormal acquisition paths, among which, Indicates the observation channel number. Indicates the first The new information components of each observation channel Indicates the relationship between the new information covariance matrix and the first... The diagonal elements corresponding to each observation channel. The positioning results are used to trigger baseline recalibration.

[0064] When the same muscle corresponds to three observation components The consistency threshold can be determined according to a chi-square distribution with 3 degrees of freedom. For example, at the significance level When the value is 0.05, the consistency threshold can be 7.81. If... Exceed The response state construction unit locates the abnormal acquisition path based on the contribution of each innovation component to the consistency index and triggers the recalibration of the corresponding acquisition path. When the abnormality manifests as a systematic drift of multiple induced responses, the baseline of the induced maximum contraction and the fresh induced response can also be reacquired.

[0065] In some embodiments, the response state construction unit 104 determines the first feature from the active muscle deformation response using the same feature extraction method as the calibration stage. Primitive induced characteristics of each discrete control step And based on the induced maximum contraction Original induced contraction characteristics Normalization was performed to obtain the degree of induced contraction. :

[0066]

[0067] As one implementation method, the original induced contraction feature This can be the peak value, integral value, or lateral shift amplitude relative to the resting baseline of the active muscle deformation response within the stimulation interval. It should be noted that regardless of the original evoked contraction feature used... Inducing maximum contraction All were determined using the same feature extraction method. (Normalized) It has the same comparison scale as the target degree of contraction and serves as the observed output of the induced contraction response model and the tracked quantity in the constrained optimization process.

[0068] In some embodiments, such as Figure 1 As shown, the system also includes a calibration control module 500. Before the closed-loop control begins, the calibration control module 500 controls the electrical stimulation output module 400 to output a calibration stimulation sequence. The calibration stimulation sequence may include multiple electrical stimulations with progressively increasing pulse widths, where the pulse width of each stimulation does not exceed a preset pulse width limit. The calibration control module 500 records the agonist muscle mechanical response corresponding to each calibration pulse width, determines the induced contraction features corresponding to each calibration pulse width using the same feature extraction method as the original induced contraction features, and determines the maximum value among these features as the induced maximum contraction. The stimulus-contraction transition characteristics, spectral characteristics, and contraction timing characteristics at the beginning of the calibration phase were recorded to form a fresh-state evoked response baseline. and the corresponding component baseline. As a normalized benchmark for inducing the degree of contraction, the calibration control module 500 uses the following target determination relationship to obtain the target degree of contraction:

[0069]

[0070] In the formula, This indicates the degree of target shrinkage during normalization. This represents the preset target intensity coefficient. It can be pre-configured based on the target muscle group and training intensity, for example, setting it to 70%. This induces maximum contraction. Used for the original induced contraction characteristics Normalization was performed to reduce the degree of induced contraction. The degree of normalization of target shrinkage They have the same comparison scale. The resulting... It can remain unchanged during a single training process and serve as a common target reference for all prediction times in the subsequent prediction time domain.

[0071] In some embodiments, the calibration control module 500 first sets the electrical stimulation frequency to 35Hz, and then gradually increases the electrical stimulation amplitude from 5mA to a predetermined tolerance amplitude; subsequently, while maintaining the electrical stimulation frequency and tolerance amplitude, the pulse width is increased from 50... Increased step by step to 300 The preset pulse width limit is set within a certain range. The calibration control module correlates each pulse width with its corresponding agonist muscle mechanical response and determines the pulse width from the calibration sequence. and If the signal consistency index during closed-loop control continuously indicates a systematic drift in the induced response, the calibration control module 500 can perform a light recalibration after the electrical stimulation output is stopped to refresh the index. , And the baselines of each component formed by calibration.

[0072] In some embodiments, such as Figure 1 and Figure 3 As shown, the predictive control module 200 includes an induced contraction state prediction unit 203. The induced contraction state prediction unit 203 establishes an induced contraction response model to obtain the conversion relationship between pulse width and induced contraction degree from historical control data. The induced contraction response model is a first-order discrete response model with exogenous input.

[0073]

[0074] In the formula, Indicates the first The degree of induced contraction in each discrete control step and Indicating a broad historical perspective, , and This indicates the model parameters that need to be updated online. This represents the model residual. The induced contraction state prediction unit 203 reads the historical pulse width and historical induced contraction degree required by the induced contraction response model from the historical control data, so that the model parameters can be updated as the pulse width-contraction response relationship changes during training.

[0075] In some embodiments, the induced contraction state prediction unit 203 updates the parameters of the induced contraction response model online using a recursive parameter update method. Its update gain, parameter estimates, and parameter covariance are determined according to the following relationships:

[0076]

[0077]

[0078]

[0079]

[0080] In the formula, This represents the parameter vector of the induced contraction response model, with superscript... This indicates the transpose operation. Indicates the parameter update gain; Indicates the first The parameter vector estimates obtained from each discrete control step include , and The estimated value; This represents the covariance matrix of the parameter estimates; ,and This represents the regression vector composed of the degree of historically induced contraction and the historical pulse width; ,and This represents the forgetting factor, which is used to give higher weight to data from more recent discrete control steps in parameter updates; This represents the identity matrix. Whenever a discrete control step generates a new induced contraction level, the induced contraction state prediction unit 203 performs a parameter recursive update, and the updated model parameters are used for the next multi-time recursive prediction.

[0081] In some embodiments, forgetting factor The value is set to 0.97. After each parameter update, the induced contraction state prediction unit 203 determines the prediction error based on the predicted value before the update and the currently obtained induced contraction degree, and continuously counts the prediction error. When the prediction error exceeds the preset prediction error threshold multiple times consecutively, it stops using the current model parameters to continue generating candidate pulse widths and triggers the reinitialization of the induced contraction response model; the preset prediction error threshold is, for example, 15%. This prediction error discrimination is used to prevent models that have significantly deviated from the current muscle response relationship from continuing to enter the constrained optimization process.

[0082] The induced contraction state prediction unit 203 uses the currently formed induced contraction degree and historical pulse width as the starting point for recursion, and uses the updated model parameters to determine the induced contraction degree at subsequent prediction times:

[0083]

[0084] In the formula, Indicates based on the first The data already obtained in the discrete control step, for the... Predicted value of the degree of induced contraction at each predicted time; =1, ..., ; Indicates the length of the prediction time domain; , and This represents the model parameters obtained through online updates. In the recursion, the index is not greater than... The item uses the obtained data, and the index is greater than... The pulse width term is provided by the candidate pulse width sequences to be solved. Multiple Arranged according to the predicted time, a prediction sequence for the degree of induced contraction is formed.

[0085] In another implementation, the response state construction unit 104 can also determine the stimulus contraction switching capability based on the parameters of the induced contraction response model, specifically using the following formula:

[0086]

[0087] In the formula, Indicates the ability to switch between stimulation and contraction. , and These represent the parameters of the induced contraction response model. In the... Each discrete control step, the response state construction unit 104, is based on the online updated... The corresponding stimulus contraction conversion ability is denoted as .

[0088] The response state construction unit 104 extracts spectral features and contraction timing features from different response components of the agonist muscle response state. As one implementation, the spectral features... High-frequency vibration band power With low-frequency vibration band power The ratio, specifically, is expressed by the following formula:

[0089]

[0090] Contraction time series characteristics It is determined by at least one of the contraction rise rate and relaxation time of the muscle deformation response. The response state building unit 104 fuses the stimulus contraction switching capacity, spectral characteristics, and contraction timing characteristics relative to their respective fresh baselines:

[0091]

[0092] In the formula, Indicates the amount of fusion of active muscle forces; , and These represent the baseline of stimulus-contraction switching capability, the baseline of spectral characteristics, and the baseline of contraction timing characteristics formed during the calibration phase, respectively. , and Indicates the fusion weight, and It can make Greater than and The ability to stimulate contraction and conversion was used as the principal component, and its verification was supplemented by spectral changes and contraction time sequence changes.

[0093] In some embodiments, to enable the use of a uniform scale for agonist muscle production states across different training processes, the response state construction unit normalizes the agonist muscle production fusion amount using a fresh-evoked response baseline:

[0094]

[0095] In the formula, Indicates the active muscle production index, This represents the baseline of the fresh-state induced response determined during the calibration phase. Fresh-state corresponding Approaching 100%, when the stimulus-contraction switching ability, spectral characteristics, or contraction timing characteristics decrease, Corresponding changes. As input to the current muscle safety status, it enters the safety output control module and simultaneously enters the active muscle force response model to form safety constraint data in the prediction time domain.

[0096] The agonist muscle productivity index describes the stimulus-contraction transition state of the agonist muscle under the current stimulation condition. The main cause of a decrease in stimulus-contraction transition capacity can be a decline in muscle productivity, or it may be caused by changes in electrode position, contact impedance, or muscle length. The predictive control module 200 can compensate for these changes through online updates of the response relationship, while the safety output control module 300 further incorporates the antagonist muscle response state and signal consistency index to eliminate system-level interference. Since the electrical stimulation electrodes act on the agonist muscle, the agonist muscle productivity index is formed using agonist muscle response characteristics, and the antagonist muscle response state is used for co-activation and artifact detection to avoid diluting the productivity changes of the agonist muscle with the antagonist muscle response.

[0097] In some embodiments, such as Figure 1 , Figure 3 and Figure 5 As shown, the predictive control module 200 also includes an active muscle force state prediction unit 201. The active muscle force state prediction unit 201 reads historical active muscle force indices and historical pulse widths from historical control data and establishes an active muscle force response model. The active muscle force response model adopts the following second-order discrete response form:

[0098]

[0099] In the formula, , , and Indicates parameters to be updated. This represents the model residuals. The agonist muscle force state prediction unit 201 reads historical agonist muscle force index and historical pulse width from historical control data and updates the parameters of the agonist muscle force response model online. After the online update is completed, the updated model parameters and the... Substituting the active muscle production index and required historical data for each discrete control step into the active muscle production response model after omitting model residuals, we obtain... to ,in, Indicates based on the first The data obtained in the discrete control step are related to the first... The predicted values ​​of the active muscle production index at each prediction time are arranged in order of prediction time to form the active muscle production index prediction sequence.

[0100] The response state construction unit 104 determines the co-activation index based on the agonist muscle response state and the antagonist muscle response state. The offset amplitude of the agonist muscle response state relative to its resting baseline can be obtained first. And the magnitude of the shift in the antagonist muscle response state relative to its resting baseline. Then, the co-activation index is determined according to the following relationship:

[0101]

[0102] In the formula, This indicates the co-activation index. The larger the value, the higher the proportion of the antagonist muscle response in the total response of both muscles. When Below the preset minimum response amount At that time, it can be Set to zero or a preset initial value to avoid unstable ratios in the low response range. Enter the predictive control module to adjust the induced contraction tracking weight, and enter the safety output control module 300 to perform admission discrimination and hard interruption discrimination.

[0103] When the antagonist muscle is not directly stimulated by the electrical stimulation electrode, its response state can also be used to verify whether the change in the agonist muscle response is caused by a change in muscle force. If the agonist and antagonist muscles decrease synchronously in the same discrete control step, and the difference between their changes is within a preset consistency range, then the response state construction unit 104 identifies this synchronous change as a system-level disturbance and sets an artifact marker.

[0104]

[0105] In the above artifact discrimination relationship, and These represent the changes in the amplitude of the agonist muscle offset and the amplitude of the antagonist muscle offset relative to the previous discrete control step, respectively. This indicates the threshold for consistency in synchronous changes. Indicates the first Artifact identification for each discrete control step. When At this time, the response state construction unit 104 can freeze the online update of the active muscle force index, and the safety output control module 300 uses the flag to prohibit or interrupt the electrical stimulation output.

[0106] In some embodiments, such as Figure 1 and Figure 3 As shown, the predictive control module 200 includes an optimization control unit 202, which determines the active muscle production safety boundary based on the motion intention confidence level c(k). As one implementation, the active muscle production safety boundary is determined according to the following relationship:

[0107]

[0108] In the formula, Indicates the first Safety boundary of active muscle production for each discrete control step. Indicates the basic security boundary. Indicates the boundary adjustment coefficient. This indicates a preset reliability threshold. Because... It can vary in different discrete control steps. Updated with discrete control steps; this boundary serves as both a constraint on the active muscle force index prediction sequence and for the first... Output admission criteria for the active muscle production index of discrete control steps.

[0109] In some embodiments, the preset reliability threshold can be 0.7, the duration of a single stimulus can be 5 seconds, the total duration of a single training session can be 20 minutes, the maximum number of stimuli in a single training session can be 40, and the prediction time domain length can be... It can be set to 10 to control the time domain length. It can be 3, representing the number of persistent discriminations of the Co-Contraction Index (CCI). The number of times continuous artifact detection can be 3. It can be 5, representing the number of times the production process terminates and de-vibrates. It can be 5. It should be noted that the above value is only used to illustrate one configuration method for the control parameters. This indicates the number of prediction times included in the prediction time domain. This indicates the number of candidate control variables included in the control time domain. and These represent the number of discrete control steps that must be continuously satisfied for the corresponding state. The values ​​written into the parameter configuration interface are used in the corresponding state prediction, output admission, or training termination process.

[0110] To illustrate how motor intent confidence and dual-muscle co-activation affect induced contraction tracking, the optimized control unit 202 can employ the following single-cycle tracking cost:

[0111]

[0112] In the formula, This represents the cost of single-cycle tracking. Indicates the total activation penalty weight. c(k) represents the deviation of the induced contraction from the target contraction, with the superscript 2 indicating the square operation. The higher c(k), the stronger the effect of the tracking bias on the control decision. The higher the value, the lower the corresponding tracking weight, in order to reduce the impact of the contraction amount formed by the joint contraction of agonist and antagonist muscles on control decisions. The weighting relationship in the single-cycle tracking cost is further extended to the prediction time domain.

[0113] Optimize control unit 202 for the first Set time weights for each prediction time. and the confidence level of the movement intention Co-activation index and Together, they are used to determine the induced contraction tracking adjustment parameters. In the first... For the discrete control step, the confidence level of future motion intention and the future co-activation index have not yet been obtained. As one implementation method, the optimized control unit 202 will... The motion intent confidence and co-activation exponent obtained from each discrete control step are used as constant parameters in the prediction time domain. Constrained optimization aims to reduce the tracking deviation of the induced contraction prediction sequence relative to the target contraction, the pulse width variation, and the safety state relaxation. The specific formula is as follows:

[0114]

[0115] In the formula, This represents the cumulative cost in the prediction time domain. Indicates the length of the prediction time domain. Indicates the length of the control time domain. Indicates the first Time weights for each prediction time point Indicates the first Smoothing weights for pulse width variations Indicates the first The change in pulse width at each predicted time point. Indicates the first The safety slack at each predicted time. The penalty weight represents the amount of relaxation in the safety state; Indicates based on the first The data already obtained in the discrete control step, for the... The predicted value is obtained by predicting the degree of induced contraction at each prediction time. and They represent the first The motion intent confidence and co-activation index obtained for each discrete control step The target degree of contraction.

[0116] To perform matrix processing on the induced contraction degree prediction sequence, the induced contraction state prediction unit 203 writes the aforementioned recursive prediction relationship of the induced contraction degree as an affine relationship between the prediction output vector and the control increment vector:

[0117]

[0118] in,

[0119]

[0120]

[0121]

[0122] In the formula, This represents the vector predicting the degree of induced contraction. This represents the vector of pulse width changes within the control time domain; This represents the predicted initial state, which is composed of the current induced contraction level and the most recent historical pulse width. ,and This represents the free response matrix that maps the predicted initial state to the free response. ,and This represents the controlled response matrix that shows the effect of each pulse width change on the degree of contraction induced at each prediction time. and It is formed recursively based on the parameters of the induced contraction response model after online updates.

[0123] The candidate pulse width sequence in the control time domain is obtained by accumulating the pulse width change vector:

[0124]

[0125] In the formula, Indicates absolute pulse width sequence, Represents the vector of pulse width changes. This represents a cumulative matrix where all elements in the lower triangular region are 1. This indicates the most recent historical pulse width. This represents a column vector where all elements are 1. This accumulation relation is used to convert the pulse width change vector into an absolute pulse width sequence corresponding to each prediction time.

[0126] Based on the aforementioned induced contraction degree prediction vector and absolute pulse width sequence, the constrained optimization process adopts the following matrix objective:

[0127]

[0128] In the formula, Indicated by and As an optimization variable, it is used for minimization. Represents the target reference vector. Each component represents the target degree of contraction. ; This represents the tracking weight matrix for induced contraction; This represents the weight matrix for pulse width variation; Indicates the slack amount under each safety state The vector formed; superscript This indicates the transpose operation. Used to represent the tracking deviation of the predicted vector for the degree of induced contraction relative to the target reference vector. Used to represent the change in pulse width Used to represent the amount of relaxation in a safe state. The diagonal elements of the equation correspond to the induced contraction tracking adjustment parameters at different prediction times. The induced contraction tracking adjustment parameters are determined by the first... The motion intention confidence, co-activation index, and time weight of the corresponding prediction time are determined for each discrete control step. The diagonal elements are used to configure the smoothing weights for different pulse width variations in the control time domain. The induced contraction tracking weight matrix takes the following form:

[0129]

[0130] In the formula, This indicates that a diagonal matrix is ​​formed using the items within the parentheses as diagonal elements.

[0131] The pulse width change weight matrix takes the following form:

[0132]

[0133] In the formula, to These represent the smoothing weights for the changes in pulse width within the control time domain.

[0134] Furthermore, substituting the aforementioned predicted output vector relationship into the matrix optimization objective, and according to the pulse width change vector... After simplification, we obtain the quadratic term matrix and the linear term vector:

[0135]

[0136]

[0137] In the formula, This represents the quadratic term matrix under constrained optimization. This represents the vector of terms in a constrained optimization process. From the controlled response matrix Induced contraction tracking weight matrix Pulse width change weight matrix Sure; From free response With the target reference vector The deviation between them is determined. Therefore, the parameters of the induced contraction response model, the first... The confidence level of the motion intention, the co-activation index, and the target contraction degree of each discrete control step are respectively incorporated into the process of determining the candidate pulse width using the corresponding data.

[0138] The optimization control unit 202 solves the following standard form of constrained optimization problem:

[0139]

[0140] In the formula, Represented by the vector of pulse width variation and safety state relaxation vector As an optimization variable, it is used for minimization. and Let them represent the quadratic term matrix and the linear term vector obtained by defining the relationship between the aforementioned quadratic term matrix and linear term vector, respectively; The penalty weight represents the amount of relaxation in the safe state.

[0141] One constraint relationship for the standard quadratic programming objective is as follows:

[0142]

[0143]

[0144]

[0145] In the formula, and These represent the lower and upper limits of the absolute pulse width, respectively. This represents a column vector where all elements are 1. Indicates absolute pulse width sequence, This indicates the upper limit of the pulse width variation. Indicates the first Predicted value of active muscle production index at each predicted time point Indicates the first Safety boundary of active muscle production for each discrete control step. Indicates the slack amount under each safety state The vector formed ≥0 indicates that all components of the vector are non-negative; Take 1 to , Take 0 to . Limit absolute pulse width, Limit the amount of pulse width variation. The predicted value of the agonist muscle productivity index is ensured to be no lower than the boundary of the agonist muscle productivity safety boundary after adjustment by the safety state relaxation amount. After obtaining the optimal pulse width change vector, the optimization control unit takes the pulse width corresponding to the first pulse width change as the first... The candidate pulse widths for the discrete control steps are used; the pulse widths at the remaining predicted times are used to solve for the optimal pulse width change in this step. Each discrete control step is re-solved based on the newly acquired data.

[0146] In some embodiments, the optimization control unit 202 assembles the aforementioned standard quadratic programming objective and its constraints into a convex quadratic programming problem, and uses an online quadratic programming solver to obtain the pulse width variation vector, such as qpOASES (online active set quadratic programming solver) or OSQP (operator splitting quadratic programming solver). Penalty weights for safety state relaxation. Set to be greater than the weight of each time period and each smoothing weight This causes the optimization control unit 202 to prioritize compressing the relaxation of the agonist muscle production safety boundary. The optimization control unit 202 uses only the first component of the optimal pulse width change vector to form a candidate pulse width, and in the next discrete control step, it re-executes the model update and optimization solution based on the newly obtained induced contraction degree, agonist muscle production index, motion intention confidence, and co-activation index, thereby forming rolling control.

[0147] In some embodiments, such as Figure 1 , Figure 4 and Figure 5 As shown, the safety output control module 300 also includes a timing coupling error determination unit 301, an output enable discrimination unit 302, an output termination control unit 303, and a hard interrupt unit 304. The timing coupling error determination unit 301 calculates the absolute value of the difference between the motion intention time information and the corresponding electrical stimulation output time information based on a common time reference to obtain the timing coupling error.

[0148] As an implementation, the preset timing threshold can be 50ms, so that central motor intention events and peripheral stimulation events are within the preset time coupling range.

[0149] In some embodiments, the output permission discrimination unit 302 discriminates each admission condition. A motion intention trigger signal indicating triggering constitutes a trigger admission condition; a motion intention confidence level not lower than a preset confidence threshold constitutes an intention reliability admission condition; an agonist muscle productivity index not lower than the agonist muscle productivity safety boundary constitutes a current productivity state admission condition; an artifact identifier indicating no artifact constitutes a signal reliability admission condition; a co-activation index lower than a preset co-activation threshold constitutes a dual-muscle synergy admission condition; and a timing coupling error not greater than a preset timing threshold constitutes a timing admission condition. When all the above admission conditions are met, the output permission discrimination unit 302 will... Execution trigger signal for each discrete control step Set to execution state and determine the candidate pulse width as the final pulse width; if at least one admission condition is not met, Set to non-executable state. Used to indicate the Each discrete control step determines whether the electrical stimulation output module 400 is allowed to output an electrical stimulation signal. Only the electrical stimulation output module 400... When in execution mode, the electrical stimulation electrode is driven to output an electrical stimulation signal according to the final pulse width.

[0150] As a specific implementation, the frequency of the electrical stimulation signal can be between 20Hz and 50Hz, and can be preset to 35Hz; the pulse width can be between 50μs and 300μs, and the amplitude can be between 0mA and 50mA, with specific values ​​configured according to the target muscle group and the predetermined tolerance state. Keeping the electrical stimulation frequency constant during a training session can reduce the impact of frequency changes on the mechanomotor spectrum characteristics; the predictive control module 200 changes the electrical stimulation output by adjusting the pulse width. After completing one electrical stimulation output and obtaining the corresponding mechanomotor signal, the system writes the pulse width of the electrical stimulation signal, the induced contraction degree determined based on the mechanomotor signal, and the agonist muscle production index into historical control data for the predictive control module 200 to read.

[0151] When the output permission discrimination unit 302 prohibits the current electrical stimulation output due to unmet motion intention confidence, co-activation index, or other admission conditions, the electrical stimulation output module 400 does not output candidate pulse widths, the safety output control module 300 sets the final pulse width to zero, and the prediction control module 200 retains the historical pulse width corresponding to the most recent valid output. When the admission conditions for subsequent discrete control steps are met again, the optimization control unit 202 uses the retained historical pulse width as the accumulation benchmark for pulse width changes to re-determine the candidate pulse width. Therefore, even if the final pulse width is set to zero, it does not affect the determination of subsequent candidate pulse widths, which are still determined based on the historical pulse width corresponding to the most recent valid output.

[0152] The output termination control unit 303 performs continuous counting and threshold comparison for different termination conditions. The first termination condition is that the agonist muscle force index falls below a preset force termination threshold for multiple consecutive discrete control steps; the second termination condition is that the stimulus contraction switching ability determined by historical control data falls below a preset switching ability termination threshold for multiple consecutive discrete control steps; the third termination condition is that the final pulse width continuously reaches a preset pulse width limit, and the induced contraction degree continuously fails to reach the target contraction degree; the fourth termination condition is that the cumulative number of electrical stimulation outputs reaches a preset number threshold; and the fifth termination condition is that the cumulative actual output duration reaches a preset duration threshold. When any termination condition is met, the output termination control unit 303 sets the execution trigger signal to a non-execution state and sets the final pulse width to zero to end the electrical stimulation output for this training session. Setting continuous discrete control step counting for the first three conditions can reduce the probability of false termination due to measurement fluctuations in a single discrete control step.

[0153] In some embodiments, when the agonist muscle production index is below a preset production termination threshold for five consecutive discrete control steps, the output termination control unit 303 determines that the corresponding termination condition is met; the stimulation contraction switching ability can also be determined by counting consecutive discrete control steps to see if it is continuously below a preset switching ability termination threshold. Optionally, when the co-activation index is not below a preset co-activation interruption threshold for three consecutive discrete control steps, the output termination control unit 303 determines the continuous co-activation state as the training termination condition. This continuous co-activation termination is independent of the co-activation admission judgment for a single discrete control step: the admission judgment is used to prohibit the output of the current electrical stimulation, and the continuous co-activation termination is used to end the current training.

[0154] The parameter configuration interface can also receive manual termination commands. In response to a manual termination command, the safety output control module 300 sets the execution trigger signal to a non-execution state and sets the final pulse width to zero. Manual termination is a control to exit the current training session and does not change the motion intent trigger signal, the mechanical response state, or the computational relationship between the two response models.

[0155] The hard interrupt unit 304 monitors persistent artifacts, abnormal co-activation of two muscles, abnormal acquisition of mechanograph signals, and abnormal data transmission links. The first hard interrupt condition is the presence of artifacts after a preset number of consecutive artifact markers; the second hard interrupt condition is that the co-activation index is not lower than a preset co-activation interruption threshold; the third hard interrupt condition is that abnormal acquisition of mechanograph signals is determined based on the aforementioned signal consistency index determined by innovation and innovation covariance, acquisition path status, or data integrity detection; the fourth hard interrupt condition is that abnormal data transmission links are determined based on communication heartbeat, verification results, or timeout detection. When any hard interrupt condition is met, the hard interrupt unit 304, without waiting for the constrained optimization process to complete, directly sets the execution trigger signal to a non-execution state and sets the final pulse width to zero.

[0156] In some embodiments, the hard interrupt unit 304 is also connected to an independent hardware watchdog. The hardware watchdog monitors the output pulse width limit, mechanical response acquisition status, and communication heartbeat of the electrical stimulation output module 400; when it detects that the output pulse width exceeds the absolute limit, data acquisition is continuously missing, or the communication heartbeat times out, the hardware watchdog directly disables the electrical stimulation output. For the continuous discrimination of artifact identifiers, it can be... The condition for persistent artifacts is determined when five consecutive discrete control steps indicate the presence of artifacts. The hardware watchdog's output disable operation does not depend on the solution results of the aforementioned constrained optimization problem.

[0157] Thus, the system establishes a training state, a training termination state, and an abnormal hard interruption path within a single training session. In the training state, the constrained optimization process pre-limits candidate pulse widths based on the agonist muscle production index prediction sequence, and the output allow discrimination unit 302 performs cycle-by-cycle admission based on the current measured state. In the training termination state, the system no longer resumes electrical stimulation output within the same training session; muscle recovery and recalibration occur at the start of the next training session. The abnormal hard interruption path immediately prohibits output by bypassing the constrained optimization process when the state or link becomes unreliable.

[0158] In another embodiment, the system further includes a timing synchronization and storage module 600. The timing synchronization and storage module 600 uses the global timestamp generated by the EEG acquisition unit 101 as a time anchor point to align the raw EEG data, raw mechanomotor data of agonist and antagonist muscles, electrical stimulation start and stop times, admission discrimination results, training termination events, and control parameter distribution records. The raw EEG data, raw mechanomotor data, and event-marked data can be written to three interconnected data files; the event-marked data records at least the discrete control step identifier, motor intention trigger time, electrical stimulation output time, final pulse width, and safety state changes. Historical control data is extracted from the records that have output electrical stimulation signals and formed corresponding mechanical responses, for the predictive control module 200 to update the response model online.

[0159] In another embodiment, the system further includes a Training Effective Quality Score (TEQS) module 700. After the training session concludes, the TEQS module 700 reads the full-cycle data synchronized with the storage module 600 and determines the effective training percentage, fatigue control achievement rate, neural plasticity coupling achievement rate, and clinical operational efficiency optimization rate based on the full-cycle data, and generates a TEQS score. The TEQS module 700 can also output curves showing the changes in agonist muscle production index, pulse width, motor intent confidence, and co-activation index with discrete control steps, as well as a total score on a percentage basis, for subsequent parameter configuration and training recording. The TEQS module performs data statistics after training, thus not altering the aforementioned closed-loop control relationship.

[0160] Example 2:

[0161] This embodiment illustrates a complete operation process of the closed-loop electrical stimulation control system of Embodiment 1.

[0162] like Figures 1 to 5As shown, after the system starts, a common time reference is first established between the EEG acquisition unit 101, the mechanical response acquisition unit 103, and the electrical stimulation output module 400. The system then checks whether the EEG acquisition channel, the mechanical response acquisition paths corresponding to the agonist and antagonist muscles, and the data transmission link are in a usable state. If an acquisition abnormality or link abnormality is detected during the startup check phase, the hard interrupt unit keeps the execution trigger signal in a non-execution state.

[0163] When the acquisition status is normal, the calibration control module 500 outputs a calibration stimulus sequence with progressively varying pulse widths within preset output limits. The response state construction unit 104 determines the maximal contraction induced from the agonist muscle mechanical response. And obtain a fresh baseline of the stimulus-contraction switching capacity, spectral characteristics, and contraction timing characteristics. Induce maximal contraction. As a normalization benchmark for the original induced contraction characteristics, the calibration control module 500 will use the target intensity coefficient The target degree of contraction during operation is determined. Freshness-induced response baseline Used to normalize the fusion amount of agonist muscle production to obtain the agonist muscle production index.

[0164] After entering closed-loop control, the EEG decoding unit 102 in the... Output of each discrete control step and The response state construction unit 104 utilizes the mechanomotor signals of the agonist and antagonist muscles to perform multi-source response state estimation based on the aforementioned state equation and observation equation, and performs signal consistency detection based on innovation and innovation covariance to form... , , and .in, Used to induce contraction tracking Used for current access determination and prediction of security constraints. Used for current tracking weight configuration and co-activation detection. Used for signal reliability assessment.

[0165] The induced contraction state prediction unit 203 reads historical pulse width and historical induced contraction degree from historical control data, updates the parameters of the induced contraction response model based on the aforementioned parameter recursive update relationship, and then performs recursive prediction based on the updated induced contraction response model to form an induced contraction degree prediction sequence. The agonist muscle force state prediction unit 201 reads historical agonist muscle force index and historical pulse width from historical control data, updates the parameters of the agonist muscle force response model, and forms an agonist muscle force index prediction sequence. The response state construction unit 104 simultaneously determines the stimulus contraction transition capability of the corresponding discrete control step based on the aforementioned stimulus contraction transition capability determination relationship, so that it is included in both the calculation of the agonist muscle force index and the subsequent training termination judgment.

[0166] The optimization control unit 202 uses the motion intention confidence of the corresponding discrete control steps. Harmony Activation Index And combined with the time weight of each prediction time. Develop induced contraction tracking adjustment parameters; determine the relationship and based on the aforementioned agonist muscle force safety boundary. Update the safety boundary for agonist muscle production; then combine the two predicted sequences and the target degree of contraction. The problem is a constrained optimization problem defined by the aforementioned standard quadratic programming objective and its constraint relationships. After solving, the pulse width corresponding to the first control increment is taken as the candidate pulse width, so that the candidate pulse width is subject to the joint constraints of the pulse width range, the pulse width change range, and the agonist muscle force state while tracking the induced contraction target.

[0167] After candidate pulse widths are formed, the output permission discrimination unit 302 sequentially checks the trigger admission condition, intention reliability admission condition, current productivity state admission condition, signal credibility admission condition, dual-muscle synergy admission condition, and timing admission condition. When all admission conditions are met, the candidate pulse width is determined as the final pulse width and executed by the electrical stimulation output module 400; if any admission condition is not met, the current electrical stimulation output is prohibited. After completing one electrical stimulation output and obtaining the mechanograph signal corresponding to the electrical stimulation output, the system writes the pulse width of the electrical stimulation signal, the induced contraction degree determined based on the mechanograph signal, and the agonist productivity index into the historical control data, and re-executes signal processing, model updating, state prediction, and candidate pulse width solving in the next discrete control step.

[0168] During the closed-loop control cycle, the output termination control unit 303 continuously checks the agonist muscle production index, stimulation-contraction conversion ability, the relationship between the final pulse width and the target contraction degree, and the cumulative output. The hard interruption unit independently checks the artifact state, co-activation state, mechanograph signal acquisition state, and data transmission link state. When any termination condition is met, the system ends the current operation; when any hard interruption condition is met, the system immediately disables electrical stimulation output. Thus, constrained optimization processing is responsible for prospective regulation in the prediction time domain, the output allowance discrimination unit 302 is responsible for the admission of a single electrical stimulation output, and the output termination control unit 303 and the hard interruption unit 304 are responsible for exit control at the training process level.

[0169] After this operation is completed, the timing synchronization and storage module 600 aligns the raw EEG data, raw mechanograph data, and event-marked data with global timestamps and writes them to the associated data files. The training effectiveness quality scoring module 700 reads the data files, determines the effective training percentage, fatigue control achievement rate, neural plasticity coupling achievement rate, and clinical operation efficiency optimization rate, and generates an evaluation report including the agonist muscle force index, pulse width, motor intention confidence, and coactivation index change curves. The above scoring processing is performed using existing data statistical methods after the closed-loop control is completed.

[0170] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A closed-loop electrical stimulation control system, characterized in that, It includes a signal processing module, a predictive control module, a safety output control module, and an electrical stimulation output module; among which, The signal processing module is configured to: acquire electroencephalogram (EEG) signals and mechanomotor signals corresponding to the target muscle group; determine the motor intention trigger signal and the motor intention confidence level based on the EEG signals; and determine the induced contraction degree, agonist muscle force index, coactivation index, and artifact markers based on the mechanomotor signals. The predictive control module is configured to: acquire historical control data, including historical pulse width, historical induced contraction degree, and historical voluntary muscle production index; determine an induced contraction degree prediction sequence based on the historical pulse width and the historical induced contraction degree; determine a voluntary muscle production index prediction sequence based on the historical pulse width and the historical voluntary muscle production index; determine induced contraction tracking adjustment parameters based on the motor intention confidence and the coactivation index; determine a voluntary muscle production safety boundary based on the motor intention confidence; and perform constrained optimization processing to determine candidate pulse widths based on the induced contraction degree prediction sequence, the voluntary muscle production index prediction sequence, the pre-determined target contraction degree, the induced contraction tracking adjustment parameters, and the voluntary muscle production safety boundary. The safety output control module is configured to: determine the execution trigger signal and the final pulse width based on the motion intention trigger signal, the motion intention confidence, the active muscle force index, the coactivation index, the artifact identifier, and the candidate pulse width; The electrical stimulation output module is configured to: in response to the execution trigger signal, drive the electrical stimulation electrode to output an electrical stimulation signal according to the final pulse width; wherein the historical control data is determined based on the output electrical stimulation signal and the mechanograph signal corresponding to the electrical stimulation signal.

2. The system according to claim 1, characterized in that, The signal processing module includes an EEG acquisition unit, an EEG decoding unit, and a mechanical response acquisition unit; wherein... The EEG acquisition unit is configured to acquire the EEG signals; The EEG decoding unit is configured to: decode the EEG signal to determine the motor intention trigger signal and the motor intention confidence level; The mechanical response acquisition unit includes mechanical response acquisition sections respectively corresponding to the agonist and antagonist muscles of the target muscle group. The mechanical response acquisition section is configured to acquire the mechanical muscle motion signal including muscle deformation response data and vibration response data.

3. The system according to claim 2, characterized in that, The signal processing module further includes a response state construction unit, which is configured as follows: Multi-source response state estimation and abnormal data processing are performed on the mechanograph signals corresponding to the agonist muscle and the antagonist muscle respectively to obtain the response state of the agonist muscle and the response state of the antagonist muscle. The degree of induced contraction is determined based on the agonist muscle response state. Based on the historical control data, the stimulus-contraction switching capability is determined; Based on the stimulus-contraction switching capability, the spectral characteristics and contraction timing characteristics contained in the agonist muscle response state, the agonist muscle force index is determined; Based on the active muscle response state and the antagonist muscle response state, the coactivation index and the artifact identifier are determined.

4. The system according to claim 1, characterized in that, It also includes a calibration control module, which is configured to: Within a preset output limit, the electrical stimulation output module is controlled to output a calibrated stimulation sequence. Based on the active muscle mechanical response corresponding to the calibrated stimulation sequence, the baseline for induced maximal contraction and fresh induced response is determined. The target degree of contraction is determined based on the induced maximum contraction and the preset target intensity coefficient.

5. The system according to claim 1, characterized in that, The prediction control module includes an induced contraction state prediction unit, which is configured as follows: Historical pulse width and historical induced contraction degree are obtained from the historical control data; Based on the historical pulse width and the historical induced contraction degree, the induced contraction response model is updated online; Based on the updated induced contraction response model, the predicted sequence of the degree of induced contraction is determined.

6. The system according to claim 1, characterized in that, The prediction control module further includes an active muscle contraction state prediction unit, which is configured to: Historical active muscle force index and historical pulse width are obtained from the historical control data; The active muscle force response model is updated online based on the historical active muscle force index and the historical pulse width. Based on the updated active muscle force response model, the predicted sequence of the active muscle force index is determined.

7. The system according to claim 1, characterized in that, The predictive control module includes an optimization control unit, which is configured to: The induced contraction tracking adjustment parameters are determined based on the motion intention confidence, the coactivation index, and the time weight of the corresponding prediction time. Based on the confidence level of the movement intention, the safety boundary of the agonist muscle force is determined; An optimization objective is constructed to reduce the tracking bias, pulse width variation, and safety state relaxation of the induced contraction degree prediction sequence relative to the target contraction degree. Under the constraints that the candidate pulse width meets the preset pulse width value range, the pulse width change meets the preset change range, and the predicted sequence of the agonist muscle force index is not lower than the boundary of the agonist muscle force safety boundary after the safety state relaxation amount is adjusted, the constrained optimization process is performed to determine the candidate pulse width.

8. The system according to claim 1, characterized in that, The safety output control module includes a timing coupling error determination unit and an output permission determination unit; wherein... The timing coupling error determination unit is configured to: perform timing alignment of the motion intention trigger signal and the electrical stimulation signal based on a common time reference, and determine the timing coupling error; The output permission discrimination unit is configured to: set the execution trigger signal to the execution state when all admission conditions are met, and determine the candidate pulse width as the final pulse width; The admission criteria include: the motion intention trigger signal indicates triggering; the confidence level of the motion intention is not lower than a preset confidence threshold; the active muscle force index is not lower than the active muscle force safety boundary; the artifact identifier indicates no artifacts; the co-activation index is lower than a preset co-activation threshold; and the temporal coupling error is not greater than a preset temporal threshold.

9. The system according to claim 1, characterized in that, The safety output control module further includes an output termination control unit, which is configured to: set the execution trigger signal to a non-execution state and set the final pulse width to zero when any of the termination conditions are met; The termination conditions include: the agonist muscle production index is continuously lower than a preset production termination threshold; the stimulation contraction switching ability determined based on the historical control data is continuously lower than a preset switching ability termination threshold; within a continuous control cycle, the final pulse width continuously reaches a preset pulse width limit, and the induced contraction degree continuously fails to reach the target contraction degree; the cumulative number of electrical stimulation outputs reaches a preset number threshold; and the cumulative actual output duration reaches a preset duration threshold.

10. The system according to claim 1, characterized in that, The safety output control module further includes a hard interrupt unit, which is configured to: set the execution trigger signal to a non-execution state and set the final pulse width to zero when any one of the hard interrupt conditions is met; The hard interruption conditions include: the artifact identifier continuously indicates the presence of artifacts; the co-activation index is not lower than the preset co-activation interruption threshold; abnormal acquisition of mechanical muscle animation signal is detected; and abnormal data transmission link is detected.

Images