A stroke rehabilitation training dynamic scheduling method based on electroencephalogram fatigue monitoring

Abstract

The application discloses a kind of based on electroencephalogram fatigue monitoring's cerebral apoplexy rehabilitation training dynamic scheduling method, the steps include: S1.in the rehabilitation training process, the electroencephalogram signal and movement performance data of patient are collected;S2.the electroencephalogram signal is preprocessed and sliding window time-frequency analysis is carried out, and cognitive fatigue index I-fatigue and movement endurance index I-endurance are calculated;S3.according to the numerical interval and its change trend where I-fatigue and I-endurance are located, current training state is determined;S4.training parameter scheduling instruction is generated based on the training state output by S3;S5.training execution module executes subsequent training task according to training parameter scheduling instruction;S6.scheduling result is recorded and individualized fatigue threshold database of patient is established or updated.The application can convert the subjective fatigue state of patient into quantifiable, traceable closed-loop scheduling index, improve the safety, continuity and individualized degree of post-stroke rehabilitation training.

Description

Technical Field

[0001] This invention relates to the field of medical rehabilitation technology, specifically to a dynamic scheduling method for stroke rehabilitation training based on brain electrical fatigue monitoring. Background Technology

[0002] Stroke patients often experience varying degrees of motor dysfunction, such as hemiplegia, loss of fine motor skills, abnormal coordination, or decreased exercise tolerance. Rehabilitation training, through repetitive, task-oriented, and actively participated motor exercises, promotes neuroplasticity and is an important means of functional recovery after stroke. With the development of brain-computer interfaces, functional electrical stimulation, and rehabilitation robotics, more and more rehabilitation systems are able to trigger peripheral stimulation or assistive movements based on the patient's motor intentions.

[0003] However, existing stroke neurorehabilitation training guidance systems primarily focus on "whether the action is triggered" and "whether the movement is completed," lacking real-time identification of the patient's fatigue state during training. Clinical training is usually conducted according to a fixed schedule and intensity, with therapists mainly relying on the patient's subjective feedback and visual observation to determine whether to allow rest or reduce intensity. Stroke patients often experience cognitive decline, inattention, decreased sensory feedback, and insufficient exercise tolerance, and their fatigue state may appear before obvious motor failure. If high-intensity training is maintained after fatigue accumulation, it may induce compensatory shoulder shrugging, abnormal synergistic contractions, worsening spasticity, or decreased training compliance; if the training intensity is reduced too early, the effective training dose will be reduced, which is not conducive to neurological remodeling.

[0004] In existing technologies, some stroke neurorehabilitation training guidance systems can record training duration, trigger success rate, or movement trajectory. However, these indicators are usually post-training statistics and cannot directly reflect changes in central cognitive load and exercise tolerance at the time of training. On the other hand, changes in prefrontal theta and alpha waves in EEG signals can reflect cognitive load and fatigue trends, while changes in sensorimotor cortex rhythms can reflect motor readiness and motor control status. If EEG fatigue indicators, motor performance indicators, and functional electrical stimulation response indicators can be combined for real-time scheduling of training parameters, a safer and more individualized closed-loop rehabilitation training system can be formed. Therefore, it is necessary to propose a dynamic scheduling method and system for post-stroke rehabilitation training based on EEG fatigue monitoring. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a dynamic scheduling method and system for post-stroke rehabilitation training based on EEG fatigue monitoring, in order to solve the problems of rigid training load and inability to adapt to the real-time physiological state of patients in existing technologies.

[0006] To achieve the above objectives, the technical solution of the present invention is: a method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring, the steps of which include: S1. During rehabilitation training, collect the patient's electroencephalogram (EEG) signals and motor performance data; S2. Perform preprocessing and sliding window time-frequency analysis on the EEG signals to calculate the cognitive fatigue index I-fatigue and the exercise endurance index I-endurance; S3. Determine the current training state based on the numerical range and trend of I-fatigue and I-endurance; S4. Generate training parameter scheduling instructions based on the training state output by S3; S5. The training execution module executes subsequent training tasks according to the training parameter scheduling instructions; S6. Record scheduling results and establish or update a database of individualized fatigue thresholds for patients; In step S1, the EEG signals include prefrontal cortical EEG signals and sensorimotor cortical EEG signals, acquired via a scalp EEG cap or dry / wet electrodes; the motor performance data includes at least motor imagery triggering results, task completion quality, joint movement data, surface electromyography (EMG) signals, and functional electrical stimulation (fES) evoked response data; the motor imagery triggering results include whether the current trial was successfully triggered; the task completion quality includes at least the joint range of motion achievement rate and movement trajectory deviation; the joint movement data includes at least the joint range of motion, movement initiation time, movement completion time, and movement trajectory; the surface EMG signals are used to determine the EMG initiation time and / or muscle contraction intensity of the target muscle; the fES evoked response data includes the time difference from the time stamp of the fES output command or the time stamp of the stimulation pulse to the time stamp of the surface EMG of the target muscle and / or the time stamp of the joint movement, the time difference being the latency of the fES evoked response.

[0007] In step S2, the cognitive fatigue index I-fatigue is calculated based on the power ratio of theta waves to alpha waves in the prefrontal cortex. The exercise endurance index I-endurance is: I-endurance = 100 × [a × SR + b × Q + c × (1 - ΔLnorm)] ΔLnorm = min{1, max[0, (Lcur - Lbase) / Lbase]} Wherein, SR is the success rate of triggering motor imagery within the most recent M valid trials; Q is the normalized value of task completion quality, ranging from 0 to 1, and Q is at least determined by the joint range of motion achievement rate ROM. rate Trajectory accuracy 1-Etraj The result is obtained by weighting the task completion result C and the normalized value of muscle contraction intensity Semg; when the current trial is successfully completed, C is 1, and when it is not completed, C is 0. traj ΔLnorm is the smaller of 1 and the normalized mean of the relative joint deviation angles. The normalized mean of the relative joint deviation angles is the average of the ratios of the absolute differences between the preset target joint angles and the actual joint angles obtained simultaneously for multiple target joints under the same rehabilitation training task and the range of target joint movement angles for the rehabilitation training task. a, b, and c are weighting coefficients, and a+b+c=1. ΔLnorm is the change in the latency of functional electrical stimulation evoked responses, where Lcur is the average latency of functional electrical stimulation evoked responses in the current analysis window or the most recent M trials, and Lbase is the average latency of functional electrical stimulation evoked responses in the pre-training calibration phase or historical non-fatigue training phase. The latency of functional electrical stimulation evoked responses is the time difference between the functional electrical stimulation output command or stimulation pulse timestamp and the onset time of electromyography on the target muscle surface or the onset time of joint movement. ΔLnorm is used to characterize the degree of increase in the current latency relative to the individualized baseline latency.

[0008] The motion performance data in step S1 is acquired through one or more of the following: training task software, functional electrical stimulation controller, surface electromyography electrodes, inertial measurement unit, angle sensor, pressure sensor, camera device, or rehabilitation robot encoder; wherein, the motion imagery triggering result includes whether a single task was successfully triggered and the triggering delay, the task completion quality includes task completion rate, trajectory deviation, joint range of motion, target achievement rate, or normalized value of muscle contraction intensity (Semg), and the functional electrical stimulation evoked response data includes functional electrical stimulation output timestamp, target muscle surface electromyography initiation time, and joint movement initiation time.

[0009] Furthermore, the EEG signal sampling rate is 250 Hz to 1000 Hz. Impedance checks are performed before acquisition and the electrode impedance is kept below a preset threshold. In the prefrontal cortex region, the acquisition electrodes are placed at one or more of the leads Fp1, Fp2, Fz, F3, and F4. In the sensorimotor cortex region, the acquisition electrodes are placed at one or more of the leads C3, C4, Cz, FC3, FC4, CP3, and CP4.

[0010] Furthermore, the preprocessing includes filtering the EEG signal, removing artifacts from the electrooculogram (EOG) and electromyogram (EMG) signals, and performing time-frequency analysis on the EEG signal through a sliding analysis window to obtain the EEG power in different frequency bands. This includes segmenting the effective EEG data into analysis windows with fixed time intervals, and calculating the power Pθ in the prefrontal cortex θ band and the power Pα in the α band using Welch power spectrum estimation for each analysis window, or obtaining the equivalent power using short-time Fourier transform or wavelet transform. The θ band is 4 Hz to 7 Hz, the α band is 8 Hz to 13 Hz, the μ band is 8 Hz to 13 Hz, and the β band is 13 Hz to 30 Hz.

[0011] Furthermore, the normalized mean of the joint relative deviation angles is: ; The target joint movement angle range or target trajectory length for the rehabilitation training task. The preset target joint angle for the rehabilitation training task software at time t. N represents the actual track angle of the joint collected by the inertial measurement unit, angle sensor, or visual tracking device at time t, and N represents the number of sampling points of the target joint in the rehabilitation training task. Normalized value of task completion quality: Q = q1 × ROM rate + q2 × (1 - E traj ) + q3 × C + q4 × Semg, where q1 + q2 + q3 + q4 = 1.

[0012] Furthermore, I-fatigue = (Pθ / Pα) × K Wherein, Pθ is the average power of the θ band of the prefrontal cortex, Pα is the average power of the α band of the prefrontal cortex, and K is the individualized normalization coefficient; K satisfies K = I0 / Rbase, where Rbase is the mean of Pθ / Pα during the calibration phase, I0 is the preset baseline target value, and I0 ranges from 0.25 to 0.40.

[0013] Furthermore, the states in step S3 include optimal training state, mild fatigue state, moderate fatigue state, and severe fatigue state, and the determination logic includes: When I-fatigue < 0.5 and I-endurance > 80%, it is considered the optimal training state; When 0.5 ≤ I-fatigue < 1.0 and I-endurance > 60%, it is judged as a state of mild fatigue; When 1.0 ≤ I-fatigue < 1.5 and I-endurance > 40%, it is judged as a moderate fatigue state; When I-fatigue ≥ 1.5 or I-endurance ≤ 40%, it is considered a state of severe fatigue. The determination logic also includes a stability confirmation step, which means that the corresponding training state is output only when the same state meets the corresponding conditions in N consecutive analysis windows or M consecutive valid trials, where N is 2 to 5 and M is 3 to 10.

[0014] Furthermore, step S4 includes: when the determination result is the optimal training state, maintaining the current high-intensity training; when the determination result is a mild fatigue state, automatically extending the rest time by 10%-20%; when the determination result is a severe fatigue state, immediately terminating the current training and suggesting ending the course. When the condition is determined to be moderate fatigue, the scheduling decision module further calculates the dominant quantity of cognitive fatigue (Dcog) and the dominant quantity of decreased exercise endurance (Dmotor), where... Dcog = I - fatigue / 1.5 Dmotor = 1 - I-endurance / 100 + ΔLrt; When Dcog - Dmotor ≥ 0.15, prioritize simplifying the task difficulty or reducing cognitive load; when Dmotor - Dcog ≥ 0.15, prioritize reducing the intensity of functional electrical stimulation or shortening the training duration per set; when |Dcog - Dmotor| < 0.15, simultaneously simplify the task difficulty and reduce the intensity of functional electrical stimulation, where ΔLrt is the extension of the latency of the functional electrical stimulation evoked response read in real time at the current scheduling moment.

[0015] Simplifying task difficulty includes reducing the number of targets, reducing the complexity of the motion trajectory, narrowing the range of motion, switching from a three-dimensional space task to a two-dimensional plane task, or switching from continuous compound actions to a single action; Reducing the intensity of functional electrical stimulation includes lowering the current amplitude or pulse width by 10% to 30%, and ensuring that the adjusted stimulation intensity is not lower than the individualized motor threshold.

[0016] The simplified task difficulty includes reducing the number of targets, reducing the complexity of the movement trajectory, narrowing the range of motion, switching from a three-dimensional spatial task to a two-dimensional planar task, or switching from continuous compound movements to a single movement; the reduced functional electrical stimulation intensity includes reducing the current stimulation current or pulse width by 10% to 30%, and the adjusted stimulation intensity is not lower than the individualized movement threshold; when pain, increased spasm, abnormal co-contraction of antagonist muscles, or abnormal increase in electromyography on the surface of the target muscle is detected, a safety lock state is entered and automatic stimulation is stopped.

[0017] A system for implementing the method of the claims includes: an EEG acquisition module, a motor performance acquisition module, a signal preprocessing module, a fatigue calculation module, a state determination module, a scheduling decision module, a training execution module, a recording update module, and a safety lock module; the EEG acquisition module is used to acquire EEG signals from the prefrontal cortex and sensorimotor cortex; the motor performance acquisition module is used to acquire motor imagery triggering results, task completion quality, joint movement data, surface electromyography signals, and functional electrical stimulation evoked response data; the signal preprocessing module is used for filtering, artifact suppression, and time-frequency analysis; the fatigue calculation module is used to calculate I-fatigue and I-endurance; the state determination module is used to perform the training state determination as described in claim 7; the scheduling decision module is used to perform the training parameter scheduling as described in claims 8 to 10; the training execution module is used to execute the adjusted training task; the recording update module is used to synchronously record the training process and update the individualized fatigue threshold database; the safety lock module is used to limit or stop the functional electrical stimulation output when severe fatigue or abnormal response occurs.

[0018] The record update module synchronously records timestamps, EEG power, I-fatigue, I-endurance, training status, training parameters before adjustment, training parameters after adjustment, task execution results, abnormal patient reactions, and manual intervention information after each analysis window, each valid trial, or each parameter adjustment. The individualized fatigue threshold database is updated after the end of a single training session, a single course, or after manual confirmation, and is used for the initialization of K-values, Lbase values, state thresholds, or scheduling strategies for the next training session.

[0019] Beneficial effects: (1) Training fatigue is transformed from subjective judgment into a calculable dual index of EEG and motor function; (2) The collection sources, calculation process and normalization benchmarks of EEG signals and motor performance data are clarified, which facilitates system implementation; (3) The adjusted tasks are executed through the training execution module and the scheduling results are recorded synchronously by the record update module, forming a traceable closed-loop training process; (4) In the state of moderate fatigue, the dominant cognitive fatigue and the dominant decline in exercise endurance are further distinguished, avoiding the simple "one-size-fits-all" reduction of training intensity; (5) Through the update of the individualized fatigue threshold database, subsequent training can better adapt to the patient's recovery stage and tolerance. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the method of the present invention.

[0021] Figure 2 This is a flowchart of the data acquisition process for electroencephalogram (EEG) signals and motor performance in this invention.

[0022] Figure 3This is a flowchart of the calculation process for I-fatigue and I-endurance in this invention.

[0023] Figure 4 This is a logic diagram for training state determination and parameter scheduling in this invention.

[0024] Figure 5 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0026] Terminology Explanation: A trial session refers to the smallest experimental unit in a training or experiment using a stroke neurorehabilitation training guidance system, where a complete task is completed according to a pre-set time schedule. In a classic trial session, each session includes a pre-set timed motor imagery task period and a rest period. During the task period, the patient is prompted visually or auditorily to visualize the movement of a specific limb (e.g., left or right hand). Each training session contains multiple trials. The EEG acquisition device simultaneously collects EEG signals and surface electromyography (EMG) signals during the task and rest periods, and labels the collected EEG and EMG signals according to the action task tags to form a training sample set for subsequent model training, motor intention recognition, and safety lock determination.

[0027] The improved neurorehabilitation training guidance system used in this invention includes an EEG acquisition module, an EEG motor intention recognition module, an EMG acquisition module, a muscle response assessment module, a functional electrical stimulation module, a stimulation parameter adjustment module, a master control calculation module, and a safety protection module.

[0028] The EEG acquisition module is used to acquire EEG signals from the patient's scalp, perform signal conditioning, and transmit the signals to the main control computing module.

[0029] The EEG motor intention recognition module communicates with the EEG acquisition module and the main control computing module. It is used to preprocess, extract features and classify the raw EEG signals, output the motor intention category and its confidence level, and pass it to the main control computing module for decision-making.

[0030] The electromyography (EMG) acquisition module communicates with the muscle response assessment module to acquire surface EMG signals of the target muscle group and related muscle groups (including antagonistic muscles), complete signal conditioning, and transmit the signals to the main control computing module.

[0031] The muscle response assessment module communicates with the stimulation parameter adjustment module and is used to assess the activation level, activation sequence, fatigue state, and abnormal co-contraction of the target muscle group based on the electromyographic data output by the electromyographic acquisition module. It outputs the current muscle response classification result (insufficient response / appropriate / excessive / abnormal antagonist muscle / fatigue trend).

[0032] The functional electrical stimulation module communicates with the main control computing module and is used to generate controllable electrical stimulation pulses based on the stimulation parameter instructions issued by the main control computing module, and apply them to the target muscle group through patch electrodes.

[0033] The stimulation parameter adjustment module communicates with the functional electrical stimulation module to receive muscle response assessment results, calculate updated stimulation parameters (intensity, pulse width, frequency, channel, duration) according to the closed-loop adjustment strategy, and write them into the parameter register area of ​​the main control calculation module for use in the next stimulation cycle.

[0034] The main control computing module is responsible for receiving EEG and EMG data, running all algorithms (intention recognition, muscle assessment, parameter adjustment), managing stimulation parameters and sending control commands to the electrical stimulation module, and coordinating the timing synchronization of each module.

[0035] The safety protection module communicates with the functional electrical stimulation module to prevent electrical stimulation from causing harm to the patient at both the hardware and software levels.

[0036] The electromyography (EMG) acquisition module collects EMG signals from patients in both resting and active movement states. For target muscle groups, it extracts the normalized mean EMG envelope. To characterize the activation level of the target muscle group, for antagonistic muscle groups, the mean normalized electromyographic envelope value was extracted. and the proportion of synchronous activation of antagonistic muscles It is used to characterize the degree of abnormal co-contraction of antagonistic muscles.

[0037] Before the assessment began, reference values ​​were calibrated for each electromyography (EMG) acquisition channel, and the EMG envelope signal was normalized based on the reference values.

[0038] The reference value is the maximum voluntary contraction reference value (MVC). Specifically, the patient is required to perform three maximum isometric contraction tests on the target muscle, each lasting 5 seconds, with a rest interval between adjacent tests, preferably 60 seconds. The electromyographic envelope peak value of each test is extracted, and the mean of the three peak values ​​is calculated as the MVC reference value for the corresponding channel. .

[0039] After the evaluation begins, the normalized electromyographic envelope signal is processed by sliding windowing, and electromyographic features are continuously extracted within each time window. Preferably, a sliding evaluation window period of 200 ms is used, and with a sampling frequency of 1000 Hz, each window contains 200 effective sampling points.

[0040] The features extracted from the preprocessed, artifact-suppressed, envelope-extracted, and normalized electromyographic data by windowing feature extraction include at least those used to characterize the activation levels of the target muscle group and synergistic muscle groups. Within each evaluation window, the normalized mean electromyographic envelope of the target muscle group, synergistic muscle group, and / or antagonistic muscle group is calculated. (%MVC) is used to characterize the average activation level of the corresponding muscle group within the current window.

[0041] in For the first Normalized electromyographic envelope values ​​at each sampling point . It serves as the basis for judging response intensity, assessing activation level, and classifying responses as insufficient, appropriate, or excessive.

[0042] Synchronous activation ratio of antagonistic muscles This is used to characterize the degree of co-contraction of the antagonist muscle relative to the target muscle group. Preferably, within the same evaluation window (200 ms) after effective activation of the target muscle group, the normalized mean electromyographic envelope values ​​of the antagonist and target muscle groups are calculated separately, and the synchronous activation ratio of the antagonist muscle is calculated using the following formula:

[0043] in: and These represent the normalized mean electromyographic envelope (%MVC) of the antagonistic and target muscle groups within the same evaluation window. To avoid distorted results due to ineffective activation of the target muscle group or a denominator close to zero, when... If the effective activation conditions of the target muscle group are not met, the synchronous activation ratio of the antagonistic muscle may not be calculated in this round. Preferably, when When, it is judged to be the normal level of co-contraction; when At that time, it was determined to be a mild abnormal co-contraction; when At that time, it was determined to be a clear abnormal co-contraction.

[0044] Example 1

[0045] A dynamic scheduling system for post-stroke rehabilitation training based on EEG fatigue monitoring includes an EEG acquisition module, a motor performance acquisition module, a signal preprocessing module, a fatigue calculation module, a state determination module, a scheduling decision module, a training execution module, a record update module, and a safety lock module.

[0046] The EEG acquisition module is used to acquire EEG signals from the prefrontal cortex and sensorimotor cortex. The motor performance acquisition module is used to collect data on motor imagery triggering results, task completion quality, joint movement data, surface electromyography signals, and functional electrical stimulation evoked response data. The signal preprocessing module is used to filter, suppress artifacts, and perform time-frequency analysis on the acquired signals. The fatigue calculation module is used to calculate the cognitive fatigue index (I-fatigue) and the exercise endurance index (I-endurance). The status determination module is used to determine the current training status based on the cognitive fatigue index and the exercise endurance index. The scheduling decision module is used to generate training parameter scheduling instructions based on the training status; The training execution module is used to execute the adjusted training tasks; The record update module synchronously records the training process and updates the individualized fatigue threshold database; The safety lock module is used to stop or limit the automatic stimulation output in cases of severe fatigue, pain, increased spasms, or abnormal synergistic contractions.

[0047] Example 2

[0048] A dynamic scheduling method for post-stroke rehabilitation training based on EEG fatigue monitoring is proposed. Its core logic is to use EEG signals as a "physiological sensor" to decode the patient's cognitive and motor fatigue levels in real time. The steps include: Step 1. During rehabilitation training, the patient's EEG signals and motor performance data are simultaneously collected using the EEG acquisition module and the motor performance acquisition module, respectively. Furthermore, the EEG signals include prefrontal cortical EEG signals (responsible for cognitive load) and sensorimotor cortical EEG signals (responsible for motor control). The acquisition electrodes are positioned in the prefrontal cortex regions related to cognitive load and attentional states, preferably including one or more leads from Fp1, Fp2, Fz, F3, and F4, and in the sensorimotor cortex regions related to EEG activity reflecting motor imagery or attempted movement, preferably including one or more leads from C3, C4, Cz, FC3, FC4, CP3, and CP4. Wherein F represents the frontal region, Fp represents the frontal pole region, C represents the central region, FC represents the frontocentral region, CP represents the central parietal region, and z represents the midline of the head; odd numbers are usually on the left side, and even numbers are usually on the right side.

[0049] In this embodiment, the patient wears a scalp EEG cap to collect prefrontal cortex EEG signals in leads Fp1, Fp2, and Fz, while simultaneously collecting sensorimotor cortex EEG signals in leads C3, C4, and Cz to assist in identifying motor imagery or attempting motor-related states. The EEG sampling rate is set to 500 Hz, and the electrode impedance is controlled below 10 kΩ before collection.

[0050] The motor performance data includes at least the results of motor imagery triggering, the quality of task completion, joint movement data, surface electromyography signals, and functional electrical stimulation evoked response data.

[0051] The motor imagery triggering result includes whether the current trial was successfully triggered; the task completion quality includes at least the joint range of motion achievement rate and the movement trajectory deviation; the joint movement data includes at least the joint range of motion, movement initiation time, movement completion time, and movement trajectory; the surface electromyography signal is used to determine the electromyographic initiation time and / or muscle contraction intensity of the target muscle; the functional electrical stimulation evoked response data includes: the time difference from the functional electrical stimulation output command timestamp or stimulation pulse timestamp to the surface electromyographic initiation time of the target muscle and / or the joint movement initiation time, the time difference being the latency of the functional electrical stimulation evoked response.

[0052] Motor performance data is collected synchronously from multiple sources; the training task software records the start time, motor cue time, success of motor imagery triggering, trigger delay, and task completion result for each trial; the functional electrical stimulation controller records the stimulus output command and stimulus pulse timestamps; surface electromyography electrodes are attached to the target muscle to record the onset time of the muscle response and the intensity of muscle contraction after stimulation; inertial measurement units or angle sensors record the range of motion of joints, trajectory deviation, and onset and completion times of movement. All of the above data is aligned using a unified timestamp, enabling the analysis of EEG signals, stimulus output, and motor performance on the same timeline.

[0053] Quantitative calibration of muscle contraction intensity includes: setting a low-load calibration phase before training begins, recording the target muscle's resting surface electromyography (EMG) RMSrest and the RMS values ​​of 3 to 5 standard induced contractions or active maximum tolerable contractions. The median of the stable segment RMS or the mean of the top 10% of stable responses is used as the individualized reference value RMSref. During training, the current root mean square value of surface EMG RMScur is calculated within a preset response window after each functional electrical stimulation pulse or exercise cue. The response window is preferably 50 ms to 500 ms after stimulation or 0 s to 1 s after exercise cue. The normalized value of muscle contraction intensity, Semg, is calculated using the following formula: Here, epsilon is a minimal constant to prevent the denominator from being zero. Therefore, Semig takes a value from 0 to 1, with a larger value indicating that the current target muscle contraction is closer to the individualized reference contraction level.

[0054] Quantitative calculation of motion trajectory deviation includes: denoting the target trajectory preset by the rehabilitation training task software as r. ref (t), where r is the actual trajectory collected by the inertial measurement unit, angle sensor, or visual tracking device. cur (t), both are time-normalized according to the same task start and end time and resampled into N sampling points. Trajectory deviation E traj Calculate using the following formula: in This refers to the range of joint motion angles or the length of the target trajectory for the rehabilitation training task, used to eliminate differences in task scales; when only single joint angles are collected... This is the absolute difference between the actual joint angle and the target joint angle. This represents the normalized mean of the relative joint deviation angles.

[0055] The success rate of motor imagery triggering (SR) is calculated based on the number of successful triggers within the most recent M valid trials. The formula is SR = Nsuccess / M, where Nsuccess is the number of successful motor imagery or attempts at motor intention within the most recent M valid trials, and M is between 5 and 30.

[0056] The normalized value of task completion quality Q is further derived from the joint range of motion achievement rate ROM. rate Trajectory accuracy 1-E traj The result of task completion, C, and the normalized value of muscle contraction intensity, Semg, constitute the optimal formula: Q = q1 × ROM rate +q2 × (1 - E traj ) + q3 × C + q4 × Semg, where q1 + q2 + q3 + q4 = 1, and C is 1 when the current trial is successfully completed and 0 when it is not completed.

[0057] Therefore, Semig is not collected as an isolated value, but is included in the calculation of I-endurance = 100 × [a × SR + b × Q + c × (1 - ΔLnorm)] through Q. This ensures that insufficient actual contraction response of the target muscle can reduce the exercise endurance index and further affect training status assessment and scheduling decisions. The aforementioned SR, Semig, and E... traj Both Q and I are written to the current valid trial with a unified timestamp, so that subsequent I-endurance calculations can reflect the current motion output capability rather than subjective scores.

[0058] Step 2. The signal preprocessing module performs time-frequency analysis on the preprocessed EEG signal and sliding analysis window to obtain the EEG power in different frequency bands, and calculates the cognitive fatigue index I-fatigue and the exercise endurance index I-endurance based on the EEG power, motor performance data and functional electrical stimulation evoked response data. The preprocessing of the EEG signal includes: (1) the original EEG signal is filtered by a 50 Hz notch filter to remove power frequency interference. (2) it is filtered by a 1 Hz to 45 Hz bandpass filter to retain the main EEG rhythms. (3) Artifact suppression: a regression-based method for removing EEG artifacts is used in the online stage. The scalp EEG acquisition device sets an auxiliary EEG reference channel at the frontal electrode (near FCz), and uses least squares regression to estimate the transfer coefficient between each EEG channel and the EEG reference channel. The regression component is subtracted from the EEG signal at each sampling point. The regression-based method for removing EEG artifacts is a classic technique established in the field of EEG signal processing. This method has low computational cost and is suitable for real-time operation. For EMG artifacts, the scalp EEG acquisition device calculates the energy of each channel in the 30-45 Hz frequency band within each sliding window. If the energy of a channel in this frequency band exceeds three times the standard deviation of the mean of the sliding baseline of the channel in the past 60 s, the channel and the window are marked as EMG contamination windows. The feature value of the most recent uncontaminated window of the channel is used to replace it during feature extraction.

[0059] The time-frequency analysis of the EEG signal sliding analysis window to obtain the EEG power of different frequency bands includes: segmenting the effective EEG data into analysis windows with fixed time intervals. In this embodiment, a 2-second analysis window and a 50% overlap rate are used for sliding analysis. For each analysis window, Welch power spectrum estimation is used to calculate the power Pθ in the prefrontal cortex theta band and the power Pα in the alpha band. Alternatively, short-time Fourier transform or wavelet transform can be used to obtain the equivalent power. The theta band ranges from 4 Hz to 7 Hz, the alpha band from 8 Hz to 13 Hz, the μ band from 8 Hz to 13 Hz, and the β band from 13 Hz to 30 Hz. The cognitive fatigue index I-fatigue is obtained by calculating the ratio of the theta wave power to the alpha wave power in the prefrontal cortex. When a patient is fatigued, the theta wave power increases while the alpha wave power decreases, and the cognitive fatigue index increases accordingly.

[0060] The cognitive fatigue index (I-fatigue) is calculated as follows: I-fatigue = (Pθ / Pα) × K Wherein, Pθ is the average power of the theta band in the prefrontal cortex, Pα is the average power of the α band in the prefrontal cortex, and K is the individualized normalization coefficient. Before training begins, 60 seconds of resting-state or low-load calibration EEG data are collected from the patient. The mean Pθ / Pα ratio Rbase during the calibration phase is calculated, and K is set to I0 / Rbase, where Rbase is the mean Pθ / Pα ratio during the calibration phase, and I0 is the preset baseline target value, with I0 ranging from 0.25 to 0.40. When historical valid training data is available, Rbase can also be obtained by weighted averaging of the training baselines from the most recent 3 to 5 training sessions in which no severe fatigue state occurred. Thus, the I-fatigue of different patients at the start of training is normalized to a level close to the same baseline, and the subsequent increase in I-fatigue reflects the increase in fatigue relative to their own baseline.

[0061] The exercise endurance index (I-endurance) is calculated based on changes in motor imagery trigger success rate, task completion quality, and functional electrical stimulation evoked response latency. It reflects a patient's ability to maintain effective motor output during training. A lower I-endurance indicates a decrease in motor imagery trigger success rate, task completion quality, or stimulus-evoked response efficiency, and a weaker ability to sustain training tasks.

[0062] I-endurance calculation formula: I-endurance = 100 × [a × SR + b × Q + c × (1 - ΔLnorm)] ΔLnorm = min{1, max[0, (Lcur - Lbase) / Lbase]} Wherein, SR is the success rate of motor imagery triggering within the most recent M valid trials, where M ranges from 5 to 30; Q is the normalized value of task completion quality, which comprehensively considers the joint range of motion achievement rate, movement trajectory deviation, task completion result, and the normalized value of muscle contraction intensity Semg, with a value range of 0 to 1; a, b, and c are weighting coefficients, and a+b+c=1; ΔLnorm is the change in the latency of functional electrical stimulation evoked responses, where Lcur is the average latency of functional electrical stimulation evoked responses in the current analysis window or the most recent M trials, and Lbase is the average latency of functional electrical stimulation evoked responses in the pre-training calibration phase or the historical non-fatigue training phase. The functional electrical stimulation evoked response latency is the time difference between the output command or stimulation pulse timestamp of the functional electrical stimulation and the onset time of electromyography on the target muscle surface or the onset time of joint movement; ΔLnorm is used to characterize the increase of the current latency relative to the individualized baseline latency. If (Lcur-Lbase) / Lbase is less than 0, then ΔLnorm=0; if (Lcur-Lbase) / Lbase is greater than 1, then ΔLnorm=1; otherwise, ΔLnorm is equal to the relative increase of the current latency relative to the baseline latency.

[0063] In this embodiment, a, b, and c are taken as 0.4, 0.3, and 0.3, respectively. When the patient's trigger success rate decreases, the quality of task completion declines, or the latency of muscle response after stimulation prolongs, I-endurance decreases accordingly.

[0064] S3. Determine the current training state based on the numerical range and trend of I-fatigue and I-endurance; the training state includes the optimal training state, mild fatigue state, moderate fatigue state, and severe fatigue state. Based on the aforementioned dual-exponential model, training states are divided into four levels and differentiated scheduling is implemented. This classification is based on: I-fatigue reflects the degree of increase in the prefrontal cortex θ / α power ratio relative to the individualized baseline, characterizing cognitive load and attentional fatigue; I-endurance integrates the success rate of motor imagery triggering, task completion quality, target muscle contraction intensity, trajectory deviation, and the latency of functional electrical stimulation-evoked responses, characterizing the patient's ability to maintain effective motor output during the current training phase. Both I-fatigue and I-endurance respectively cover two dimensions: central cognitive fatigue and peripheral motor endurance decline. Therefore, "low I-fatigue and high I-endurance" can be defined as the optimal training state; a slight deviation between the two can be defined as a mild fatigue state; increased cognitive fatigue or significantly decreased motor endurance without reaching the safe termination condition can be defined as a moderate fatigue state; and a significantly increased I-fatigue or I-endurance falling below the preset lower limit can be defined as a severe fatigue state. The technical effect of the above four-level division is that it enables the system to convert continuously changing EEG and motor performance data into discrete, executable scheduling instructions, which correspond to maintaining training, extending rest or slightly reducing load, moderately adjusting parameters, and terminating training or locking safety, thereby avoiding frequent false triggers caused by single threshold control and improving the real-time performance, safety and individualization of rehabilitation training scheduling.

[0065] When I-fatigue < 0.5 and I-endurance > 80%, it is considered the optimal training state; When 0.5 ≤ I-fatigue < 1.0 and I-endurance > 60%, it is judged as a mild fatigue state; When 1.0 ≤ I-fatigue < 1.5 and I-endurance > 40%, it is judged as a moderate fatigue state; When I-fatigue ≥ 1.5 or I-endurance ≤ 40%, it is considered a state of severe fatigue.

[0066] For boundary or crossover states, the state determination module adopts a conservative priority principle: the optimal training state requires both I-fatigue and I-endurance to meet optimal conditions simultaneously; when the cognitive fatigue index is still within the optimal range but the exercise endurance index has not reached the optimal range, the training state is output according to the lower level corresponding to the decline in exercise endurance. Therefore, when I-endurance=75% and I-fatigue=0.4, although I-fatigue<0.5, I-endurance does not exceed 80%, so the system does not determine it as the optimal training state, but rather as a mild fatigue state or a mild decline in exercise endurance, and implements a scheduling strategy of extending the rest time by 10% to 20% or reducing the duration of a single training set by 5% to 10%.

[0067] The determination logic also includes a stability confirmation step, which means that the corresponding training state is output only when the same state meets the corresponding conditions in N consecutive analysis windows or M consecutive valid trials, where N is 2 to 5 and M is 3 to 10.

[0068] Step 4. Generate training parameter scheduling instructions based on the training status, wherein the training parameters include training duration, task difficulty, FES stimulus intensity, and rest interval; For optimal training, maintain current training parameters or slightly increase task difficulty within a preset safe range; for mild fatigue, extend rest intervals by 10% to 20% or reduce single-set training duration by 5% to 10% to avoid overexertion; for severe fatigue, immediately terminate training, prompt rest, and disconnect automatic functional electrical stimulation (FES) output. This closed-loop scheduling mechanism ensures that each training session is conducted within the patient's individual optimal stimulation zone, avoiding both overload and underextension. FES stimulation intensity includes one or more parameters among stimulation current amplitude, pulse width, stimulation frequency, and duty cycle.

[0069] When a state of moderate fatigue is determined, the scheduling decision module no longer simply chooses to reduce the stimulus or the task difficulty, but further calculates the cognitive fatigue dominant quantity Dcog and the exercise endurance decline dominant quantity Dmotor. The calculation formula is as follows: Dcog = I - fatigue / 1.5 Dmotor = Here, ΔLrt is the normalized increment of the functional electrical stimulation evoked response latency obtained in real time at the current scheduling moment in step 2, rather than simply calling the historical sliding window mean from the early training stage or the previous stage. Specifically, step 2 continuously updates Lcur and ΔLnorm with a 2-second analysis window and a 50% overlap rate; the state determination module outputs the current training state after satisfying the same state for N consecutive analysis windows or M consecutive valid trials; when the scheduling decision module receives this state output, it synchronously reads the ΔLnorm of the most recently confirmed analysis window or the most recent M valid trials as ΔLrt, and sets Dmotor = 1 - I-endurance / 100 + ΔLrt.

[0070] To reduce real-time scheduling lag, the system also sets a time validity constraint: if the interval between the end time of the latest valid analysis window used to calculate ΔLrt and the current scheduling time is greater than the preset time limit Tmax, the value is not used directly, but the system waits for the next valid analysis window or updates exponentially based on the most recent valid value; Tmax is preferably 3 s to 10 s. Thus, ΔLnorm serves as the basic definition for the latency normalization increment in step 2, and ΔLrt, as the real-time value read at the current scheduling time in step 4, participates in the calculation of the dominant quantity of exercise endurance decline, thereby balancing the consistency of calculation definitions and the needs of real-time scheduling.

[0071] When Dcog - Dmotor ≥ 0.15, it indicates that attentional load or cognitive fatigue is more prominent, and it is advisable to simplify the task difficulty or reduce the cognitive load. When Dmotor - Dcog ≥ 0.15, it indicates a more pronounced decline in peripheral exercise endurance, and the intensity of functional electrical stimulation should be reduced or the duration of a single training session should be shortened. When |Dcog - Dmotor| < 0.15, the task difficulty is simplified and the intensity of functional electrical stimulation is reduced.

[0072] The 0.15 value is the threshold value of the preferred embodiment. In other embodiments, this threshold value can be adjusted between 0.1 and 0.2 based on clinical expert experience or individual patient differences. Furthermore, simplifying task difficulty includes reducing the number of targets, reducing the complexity of motion trajectories, narrowing the range of motion, switching from three-dimensional space tasks to two-dimensional planar tasks, or switching from continuous compound actions to single actions.

[0073] Reducing the intensity of functional electrical stimulation includes lowering the current amplitude or pulse width by 10% to 30%, and ensuring that the adjusted stimulation intensity is not lower than the individualized motor threshold.

[0074] When the patient reports increased pain or spasms, or when abnormal co-contraction of antagonistic muscles or abnormally elevated electromyography on the surface of the target muscle is detected, the system enters a safety lockout state and stops automatically increasing stimulation.

[0075] Abnormal co-contraction of antagonistic muscles or abnormally elevated electromyographic activity on the surface of target muscles is determined by the muscle response assessment module in the neurorehabilitation training guidance system.

[0076] S5. The training parameter scheduling instruction is sent to the training execution module, which then executes subsequent training tasks according to the adjusted training parameters.

[0077] S6. The record update module synchronously records the scheduling results, training execution results, and patient responses during the training execution process, and updates the patient's individualized fatigue threshold database after the training phase ends.

[0078] During training, the recording and updating module works synchronously with the training execution module. After each analysis window, each valid trial, or each parameter adjustment, the recording and updating module saves the current timestamp, EEG power, I-fatigue, I-endurance, training status, parameters before adjustment, parameters after adjustment, task execution results, and information on abnormal patient reactions and manual interventions.

[0079] After a single training session or course, the individualized fatigue threshold database is updated based on the time point of the first occurrence of mild, moderate, and severe fatigue states during the training, the corresponding fatigue index (I-fatigue) and exercise endurance index (I-endurance), training parameters before and after scheduling, patient abnormal reactions, and manual confirmation results. The individualized fatigue threshold database stores at least: patient identification, training date, mean Pθ / Pα values ​​(Rbase) during the calibration phase, individualized normalization coefficient (K), baseline latency of functional electrical stimulation evoked responses (Lbase), optimal / mild / moderate / severe fatigue state thresholds, the time of first occurrence of each state, the sequence of I-fatigue and I-endurance from the most recent training sessions, task difficulty level, training duration, rest interval, FES stimulation current / frequency / pulse width / duty cycle, initial task intensity, safety event markers, and manual confirmation results. The K value is updated using the weighted average of Rbase from the calibration phase, where no severe fatigue occurred and training tolerance was manually confirmed to be good in the last 3 to 5 training sessions, and Knew = I0 / Rbase,new is set. Lbase is updated using the median or weighted average of the effective induced response latency within the last 3 to 5 non-fatigue windows. The state threshold is adjusted in small steps based on the time of the first occurrence of fatigue state, manual confirmation labels, and safety events. If mild fatigue occurs too early, the initial intensity of the next task is reduced or the rest ratio is increased. If fatigue does not occur in multiple consecutive training sessions and the completion quality is stable, the initial task intensity is slightly increased within a safe range. At the start of the next training session, the system initializes the training scheme based on the updated K, Lbase, state threshold, and initial task intensity, thereby achieving individualized closed-loop scheduling.

[0080] Example 3: Adaptive training scheduling for stroke patients in the recovery period The patient was a 56-year-old male with right-sided hemiplegia, diagnosed 3 months after a left basal ganglia infarction, Brunnstrom stage IV. During system initialization, EEG electrodes were attached to Fp1, Fp2, Fz, C3, C4, and Cz. Initial training parameters were: task difficulty level 3 (three-dimensional grasping task); functional electrical stimulation current 20 mA; 2 minutes of training per set, followed by a 30-second rest period.

[0081] Within 0 to 10 minutes, the patient's condition was good, with I-fatigue maintained at around 0.3 and I-endurance above 85%, indicating optimal training status, and the original parameters were maintained. From 10 to 20 minutes, the patient's attention began to wander, I-fatigue rose to 0.7, and I-endurance dropped to 70%, indicating mild fatigue. The rest time was extended from 30 seconds to 40 seconds, and the patient was prompted to rest on the screen. From 20 to 30 minutes, I-fatigue rose to 1.2, and I-endurance dropped to 55%, indicating moderate fatigue. Because Dmotor was higher than Dcog, the system preferentially reduced the functional electrical stimulation current to 15mA and lowered the task difficulty to level 2. The patient was able to continue training without significant compensatory shoulder shrugging.

[0082] Example 4: Fatigue warning in patients with cognitive impairment and stroke The patient, a 68-year-old female, presented with cognitive decline and right-sided cerebral infarction, exhibiting poor training compliance. Five minutes after training commenced, the system detected a rapid increase in prefrontal theta wave power, with I-fatigue reaching 1.8 and I-endurance dropping to 35%. The system determined this to be a state of severe fatigue, and the safety lock module immediately restricted functional electrical stimulation output. The training execution module ceased the current task and sent a notification to the therapist's workstation stating, "Cognitive overload; it is recommended to end this training session or shorten the duration of each session." Based on the system report, the therapist shortened subsequent training sessions from 30 minutes to 15 minutes and added rest and interaction prompts between tasks.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring, characterized in that, The steps include: S1. During rehabilitation training, collect the patient's electroencephalogram (EEG) signals and motor performance data; S2. Perform preprocessing and sliding window time-frequency analysis on the EEG signals to calculate the cognitive fatigue index I-fatigue and the exercise endurance index I-endurance; S3. Determine the current training state based on the numerical range and trend of I-fatigue and I-endurance; S4. Generate training parameter scheduling instructions based on the training state output by S3; S5. The training execution module executes subsequent training tasks according to the training parameter scheduling instructions; S6. Record scheduling results and establish or update a database of individualized fatigue thresholds for patients; In step S1, the EEG signals include prefrontal cortical EEG signals and sensorimotor cortical EEG signals, acquired through a scalp EEG cap or dry / wet electrodes; the motor performance data includes at least the motor imagery triggering result, task completion quality, joint movement data, surface electromyography (EMG) signals, and functional electrical stimulation (fES) evoked response data; the motor imagery triggering result includes whether the current trial was successfully triggered; the joint movement data includes at least the joint range of motion, movement initiation time, movement completion time, and movement trajectory; the surface EMG signals are used to determine the EMG initiation time and / or muscle contraction intensity of the target muscle; the fES data includes the time difference from the time stamp of the fES output command or the time stamp of the stimulation pulse to the time stamp of the surface EMG of the target muscle and / or the time stamp of the joint movement, the time difference being the latency of the fES. In step S2, the cognitive fatigue index I-fatigue is calculated based on the power ratio of theta waves to alpha waves in the prefrontal cortex. The exercise endurance index I-endurance is: I-endurance = 100 × [a × SR + b × Q + c × (1 - ΔLnorm)] ΔLnorm = min{1, max[0, (Lcur - Lbase) / Lbase]} Wherein, SR is the success rate of triggering motor imagery within the most recent M valid trials; Q is the normalized value of task completion quality, ranging from 0 to 1, and Q is at least determined by the joint range of motion achievement rate ROM. rate Trajectory accuracy 1-E traj The result is obtained by weighting the task completion result C and the normalized value of muscle contraction intensity Semg; when the current trial is successfully completed, C is 1, and when it is not completed, C is 0. traj ΔLnorm is the smaller of 1 and the normalized mean of the relative joint deviation angles. The normalized mean of the relative joint deviation angles is the average of the ratios of the absolute differences between the preset target joint angles and the actual joint angles obtained simultaneously for multiple target joints under the same rehabilitation training task and the range of target joint movement angles for the rehabilitation training task. a, b, and c are weighting coefficients, and a+b+c=1. ΔLnorm is the change in the latency of functional electrical stimulation evoked responses, where Lcur is the average latency of functional electrical stimulation evoked responses in the current analysis window or the most recent M trials, and Lbase is the average latency of functional electrical stimulation evoked responses in the pre-training calibration phase or historical non-fatigue training phase. The latency of functional electrical stimulation evoked responses is the time difference between the functional electrical stimulation output command or stimulation pulse timestamp and the onset time of electromyography on the target muscle surface or the onset time of joint movement. ΔLnorm is used to characterize the degree of increase in the current latency relative to the individualized baseline latency.

2. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, The EEG signal sampling rate is 250 Hz to 1000 Hz. Impedance checks are performed before acquisition and the electrode impedance is kept below a preset threshold. In the prefrontal cortex region, the acquisition electrodes are placed at one or more of the leads Fp1, Fp2, Fz, F3, and F4. In the sensorimotor cortex region, the acquisition electrodes are placed at one or more of the leads C3, C4, Cz, FC3, FC4, CP3, and CP4.

3. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, Preprocessing includes filtering the EEG signal and removing artifacts from the electrooculogram (EOG) and electromyogram (EMG) signals. Time-frequency analysis of the EEG signal using a sliding analysis window to obtain the EEG power in different frequency bands includes: segmenting the effective EEG data into fixed-time analysis windows; calculating the prefrontal cortex θ-band power Pθ and α-band power Pα for each analysis window using Welch power spectrum estimation, or obtaining the equivalent power using short-time Fourier transform or wavelet transform; wherein the θ-band is 4 Hz to 7 Hz, the α-band is 8 Hz to 13 Hz, the μ-band is 8 Hz to 13 Hz, and the β-band is 13 Hz to 30 Hz.

4. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, The normalized mean of the joint relative deviation angles is: ; The target joint movement angle range or target trajectory length for the rehabilitation training task. The preset target joint angle for the rehabilitation training task software at time t. N represents the actual track angle of the joint collected by the inertial measurement unit, angle sensor, or visual tracking device at time t, and N is the number of sampling points of the target joint in the rehabilitation training task. Normalized value of task completion quality: Q = q1 × ROM rate + q2 × (1 - E traj ) + q3 × C + q4 ×Semg, where q1 + q2 + q3 + q4 = 1.

5. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, I-fatigue = (Pθ / Pα) × K Wherein, Pθ is the average power of the θ band of the prefrontal cortex, Pα is the average power of the α band of the prefrontal cortex, and K is the individualized normalization coefficient; K satisfies K = I0 / Rbase, where Rbase is the mean of Pθ / Pα during the calibration phase, I0 is the preset baseline target value, and I0 ranges from 0.25 to 0.

40.

6. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, The states in step S3 include optimal training state, mild fatigue state, moderate fatigue state, and severe fatigue state. The determination logic includes: When I-fatigue < 0.5 and I-endurance > 80%, it is considered the optimal training state; When 0.5 ≤ I-fatigue < 1.0 and I-endurance > 60%, it is judged as a mild fatigue state; When 1.0 ≤ I-fatigue < 1.5 and I-endurance > 40%, it is judged as a moderate fatigue state; When I-fatigue ≥ 1.5 or I-endurance ≤ 40%, it is considered a state of severe fatigue. The determination logic also includes a stability confirmation step, which means that the corresponding training state is output only when the same state meets the corresponding conditions in N consecutive analysis windows or M consecutive valid trials, where N is 2 to 5 and M is 3 to 10.

7. The method for dynamic scheduling of post-stroke rehabilitation training based on EEG fatigue monitoring according to claim 1, characterized in that, Step S4 includes: when the determination result is the optimal training state, maintain the current high-intensity training; when the determination result is a mild fatigue state, automatically extend the rest time by 10%-20%; when the determination result is a severe fatigue state, immediately terminate the current training and suggest ending the course. When the condition is determined to be moderate fatigue, the scheduling decision module further calculates the dominant quantity of cognitive fatigue (Dcog) and the dominant quantity of decreased exercise endurance (Dmotor), where... Dcog = I - fatigue / 1.5 Dmotor = 1 - I-endurance / 100 + ΔLrt; When Dcog - Dmotor ≥ 0.15, prioritize simplifying the task difficulty or reducing cognitive load; when Dmotor - Dcog ≥ 0.15, prioritize reducing the intensity of functional electrical stimulation or shortening the training duration per set; when |Dcog - Dmotor| < 0.15, simultaneously simplify the task difficulty and reduce the intensity of functional electrical stimulation, where ΔLrt is the extension of the latency of the functional electrical stimulation evoked response read in real time at the current scheduling moment. Simplifying task difficulty includes reducing the number of targets, reducing the complexity of the motion trajectory, narrowing the range of motion, switching from a three-dimensional space task to a two-dimensional plane task, or switching from continuous compound actions to a single action; Reducing the intensity of functional electrical stimulation includes lowering the current amplitude or pulse width by 10% to 30%, and ensuring that the adjusted stimulation intensity is not lower than the individualized motor threshold.

8. A system for implementing the method according to any one of claims 1-7, characterized in that, include: The EEG acquisition module is used to acquire EEG signals from the prefrontal cortex and sensorimotor cortex. The motion performance acquisition module is used to collect motion imagery triggering results, task completion quality, joint motion data, surface electromyography signals, and functional electrical stimulation evoked response data; the central processing module is used to preprocess the acquired data, calculate the cognitive fatigue index and exercise endurance index, determine the current training status, and generate training parameter scheduling instructions based on the training status and dominant relationship. The training execution module is used to execute adjusted rehabilitation training tasks according to scheduling instructions; The record update module is used to synchronously record training process data and establish or update an individualized fatigue threshold database; The safety lockout module is used to limit or stop the output of functional electrical stimulation in the event of severe fatigue or abnormal response.

Images