A no-training neuro-rehabilitation system combining peripheral visual stimulation and motor imagery and methods of use thereof

Abstract

This invention discloses a training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery, and its usage method. The system includes: a stimulation module, an SSVEP data acquisition module, an SSVEP classification and recognition module, a motor imagery data acquisition module, a motor imagery decoding model training module, a comparison module, and a motor imagery recognition accuracy statistics module. The system uses peripheral visual stimulation as the SSVEP signal induction paradigm. After each visual stimulus, SSVEP signal classification and recognition are performed, and the classification results are converted into control commands for rehabilitation equipment to drive limb training. Within 5 seconds thereafter, an active motor imagery task is performed, while motor imagery data is collected. The SSVEP signal classification results are used as training sample labels. Once the motor imagery data meets the training sample size, the motor imagery decoding model is trained, and the model's performance in recognizing motor intentions is evaluated, gradually reducing its dependence on peripheral visual stimulation.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface technology, and in particular to a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, and its method of use. Background Technology

[0002] Neurological diseases such as stroke often lead to limb motor dysfunction, the root cause of which lies in damage to the motor neural pathways. Timely rehabilitation treatment can effectively stimulate the motor function areas of the brain, thereby promoting the remodeling of damaged nerves and restoring limb motor function.

[0003] Brain-computer interface (BCI) technologies used in rehabilitation therapy mainly include steady-state visual evoked potentials (SSVEP) and motor imagery. SSVEP features stable signals, a high signal-to-noise ratio, and widespread usability for the vast majority of people. Good recognition accuracy can be achieved using training-free algorithms such as canonical correlation analysis (CCA) and filter bank canonistic correlation analysis (FBCCA). Motor imagery BCI technology can greatly promote active brain participation during rehabilitation therapy, effectively improving the efficacy of neurorehabilitation, and therefore has the potential for widespread application in rehabilitation. Currently, motor imagery EEG decoding often employs trained deep learning or machine learning supervised algorithms, thus generally divided into a training data acquisition phase and a testing phase. In the training data acquisition phase, training paradigms are designed and motor imagery training samples are collected for algorithm model training; in the testing phase, the trained model is used to identify and classify EEG signals for left hand, right hand, and both feet.

[0004] Currently, SSVEP brain-computer interfaces often use a stimulation paradigm of black background and flashing white squares. During the process, the subject or patient needs to keep their eyes focused on the flashing stimulus, which can easily cause visual fatigue. Furthermore, the high-contrast flashing stimulus has the risk of inducing epilepsy. For stroke patients, the brain lacks active motor intention, so the stimulation of the damaged motor nerves is insufficient, and the rehabilitation effect is generally poor.

[0005] The training sample collection phase of motor imagery brain-computer interfaces typically lasts about 15-20 minutes. During this process, subjects or patients are required to be in a completely resting state, avoiding any electromyographic artifacts caused by limb movements, and actively perform motor imagery tasks. However, this requirement can easily lead to fatigue, posing a significant challenge, especially for stroke patients or those whose motor control is compromised.

[0006] Most subjects or patients who lack experience with brain-computer interfaces do not perform well in the initial stages of using motor imagery. The accuracy rate of left-hand and right-hand binary classification may only be 50%-60%, and frequent incorrect motor recognition may easily discourage patients' confidence in rehabilitation treatment. Summary of the Invention

[0007] To address the issues of low accuracy in recognizing motor intentions during the initial use of existing motor imagery brain-computer interfaces (BCIs) and the fatigue that easily occurs in patients with movement disorders during BCI rehabilitation, this invention proposes a training-free neurorehabilitation system and its usage method that combines peripheral visual stimulation and motor imagery. This system can improve upon the low accuracy of motor imagery BCIs in the initial stages of use and reduce fatigue in patients with movement disorders during BCI rehabilitation.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] This invention proposes a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, comprising:

[0010] The stimulation module includes a peripheral visual stimulation submodule and a motor imagery stimulation submodule. The peripheral visual stimulation submodule is used to generate peripheral visual stimulation to induce SSVEP EEG signals, and the peripheral visual stimulation adopts a circular stimulation. The motor imagery stimulation submodule is used to play left and right hand movement videos inside the circular stimulation to induce motor imagery EEG signals.

[0011] The SSVEP data acquisition module is used to acquire 8-lead EEG data from the occipital region as SSVEP data.

[0012] The SSVEP classification and identification module is used to classify and identify SSVEP signals based on the collected 8-lead EEG data of the occipital region through filter bank-based canonical correlation analysis (FBCCA).

[0013] The motion imagery data acquisition module is used to collect EEG data from the 9 leads of the parietal lobe as motion imagery data;

[0014] The motion imagery decoding model training module is used to train the motion imagery decoding model, which is based on deep learning technology and uses convolutional neural networks and self-attention mechanisms to process motion imagery data.

[0015] The comparison module is used to take the SSVEP classification and recognition results as the real category labels, use the trained motor imagery decoding model to identify the motor intention of the collected motor imagery data, and compare it with the real category labels. If the results are consistent, the collected motor imagery data is used to fine-tune the motor imagery decoding model. If they are inconsistent, the collected motor imagery data is discarded, and rehabilitation training is carried out based on the SSVEP classification and recognition results.

[0016] The accuracy statistics module for motor imagery recognition is used to calculate the accuracy of motor imagery recognition within a certain period. If the overall accuracy reaches 70%, peripheral visual stimulation can be canceled, and rehabilitation training can be carried out solely through motor imagery. If the accuracy is insufficient, motor imagery data will continue to be collected for model fine-tuning.

[0017] Furthermore, the circular stimuli on the left and right sides flash at fixed frequencies of 12Hz and 14Hz, respectively.

[0018] Furthermore, in the SSVEP data acquisition module, when acquiring EEG data from 8 leads in the occipital region, the leads are set to P5, Pz, P6, PO7, O1, Oz, O2, and PO8, and the lead positions are placed according to the international 10-20 lead standard.

[0019] Furthermore, in the SSVEP classification and recognition module, FBCCA uses 1 second as the decoding data length, and the data shape is (8, 250), where 8 is the number of leads and 250 is the number of sampling points. A reference signal with the same frequency as the flicker stimulus is preset, and the original signal is decomposed into sub-band signals of different frequency bands through a bandpass filter bank. Then, canonical correlation analysis is used to calculate the correlation coefficient between each sub-band signal and the reference signal and the weighted summation is performed. Finally, the category of the reference signal with the highest correlation is used as the recognition result.

[0020] Furthermore, in the motion imagery data acquisition module, when acquiring EEG data from the 9 leads of the parietal lobe, the leads are set as FC3, FCZ, FC4, C3, CZ, C4, CP3, CPZ, and CP4, and the lead positions are placed according to the international 10-20 lead standard.

[0021] Furthermore, in the motion imagery decoding model training module, the motion imagery data is first subjected to one-dimensional convolution with a convolution kernel of scale (1,25) to extract local refinement features; then it is subjected to convolution along the lead channels with a convolution kernel of scale (ch,1), where ch is the number of leads, compressing the data into one-dimensional data of (1,Ns), where Ns is the number of sample points after convolution; then the data is segmented into (1,40) segments to conform to the input of the self-attention mechanism, and the global dependency of the EEG signal context is extracted by the self-attention mechanism.

[0022] Another aspect of the present invention proposes a method of using a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, comprising:

[0023] The peripheral visual stimulation submodule generates a ring-shaped stimulus to induce SSVEP EEG signals. The motor imagery stimulation module plays videos of left and right hand movements inside the ring-shaped stimulus to induce motor imagery EEG signals.

[0024] EEG data from 8 leads in the occipital region were collected using the SSVEP data acquisition module and used as SSVEP data.

[0025] The SSVEP signal classification and identification was performed on the collected 8-lead EEG data of the occipital region using the SSVEP classification and identification module based on the filter bank canonical correlation analysis (FBCCA).

[0026] EEG data from the 9 leads of the parietal lobe were collected using the motion visualization data acquisition module as motion visualization data.

[0027] The motion imagery decoding model is trained through the motion imagery decoding model training module. The motion imagery decoding model is based on deep learning technology and uses convolutional neural networks and self-attention mechanisms to process motion imagery data.

[0028] The comparison module uses the SSVEP classification and recognition results as the real category labels. The trained motor imagery decoding model is used to identify the motor intention of the collected motor imagery data and compare it with the real category labels. If the results are consistent, the collected motor imagery data is used to fine-tune the motor imagery decoding model. If they are inconsistent, the collected motor imagery data is discarded and rehabilitation training is carried out based on the SSVEP classification and recognition results.

[0029] The accuracy of motor imagery recognition is statistically analyzed over a certain period using the motor imagery recognition accuracy statistics module. If the overall accuracy reaches 70%, peripheral visual stimulation can be canceled, and rehabilitation training can be conducted solely through motor imagery. If the accuracy is insufficient, motor imagery data will continue to be collected for model fine-tuning.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] This invention utilizes brain-computer interface technology that combines peripheral visual stimulation with motor imagery. Peripheral visual stimulation is used as the SSVEP signal elicitor paradigm. Initially, after each visual stimulus, the system classifies and identifies the SSVEP signal, converting the classification results into control commands sent to the rehabilitation device to guide the patient's limb training. During this period, the patient is required to perform active motor imagery tasks, while the system collects EEG signals as training samples, using the SSVEP classification results as training sample labels. Once the motor imagery data meets the training sample size, a motor imagery decoding model is trained. Subsequently, the model's performance in recognizing motor intentions is evaluated over a certain period, gradually reducing its reliance on peripheral visual stimulation.

[0032] The present invention can bring the following benefits:

[0033] 1. Visual stimulation elicits a strong signal response, and the SSVEP signal recognition accuracy is high, which improves the shortcomings of the low recognition accuracy in the early stages of use of existing motor imagery brain-computer interfaces, making it plug-and-play.

[0034] 2. Peripheral visual stimulation was used to reduce visual fatigue.

[0035] 3. A method combining peripheral visual stimulation and motor imagery is adopted. The SSVEP classification results are used as labels for motor imagery training samples. Limb training is performed within 5 seconds thereafter, and motor imagery training data is collected, which serves as a complete labeled training sample. This process is repeated initially until the training set size is sufficient before starting the training of the motor imagery decoding model. This method avoids the tedious and lengthy stage of collecting motor imagery training samples.

[0036] 4. During a period of use after the motion imagery decoding model has been trained, the SSVEP classification results can be compared with the motion imagery decoding results to evaluate the performance of the motion imagery decoding model, and samples with consistent results can be collected to gradually improve the decoding performance of motion intent. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the architecture of a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, according to an embodiment of the present invention.

[0038] Figure 2 A flowchart illustrating a method of using a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, according to an embodiment of the present invention.

[0039] Figure 3 This is one of the schematic diagrams of the stimulus paradigm provided in the embodiments of the present invention;

[0040] Figure 4 This is a second schematic diagram of the stimulus paradigm provided in an embodiment of the present invention;

[0041] Figure 5 The third schematic diagram of the stimulus paradigm provided in the embodiments of the present invention. Detailed Implementation

[0042] For ease of understanding, the following explanations are provided for some of the terms used in the specific embodiments of this invention:

[0043] SSVEP: Steady-state visual evoked potentials, which are observable electroencephalograms (EEGs) that elicit a corresponding frequency response in the brain through a stimulation module that flashes at a fixed frequency.

[0044] Peripheral visual stimulation: A peripheral circular ring was used as the stimulation target to induce SSVEP signals.

[0045] Motor imagery: refers to imaginary behavior performed solely through the brain without actual physical movement. The patient's motor intentions can be identified by decoding EEG signals.

[0046] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments:

[0047] like Figure 1 As shown, a training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery includes:

[0048] The stimulation module includes a peripheral visual stimulation submodule and a motor imagery stimulation submodule. The peripheral visual stimulation submodule generates peripheral visual stimulation to induce SSVEP EEG signals. The peripheral visual stimulation uses circular stimulation, with the left and right circular stimuli flashing at fixed frequencies of 12Hz and 14Hz, respectively. The motor imagery stimulation submodule plays left and right hand movement videos within the circular stimulation, respectively, to activate the patient's mirror neurons and perform mental imagery tasks related to the movements, thereby inducing motor imagery EEG signals.

[0049] The SSVEP data acquisition module is used to acquire 8-lead EEG data from the occipital region using an EEG acquisition device; the sampling frequency is 250Hz, and the lead selection is P5, Pz, P6, PO7, O1, Oz, O2, and PO8. The lead positions are placed according to the international standard of 10-20 leads, and the acquired EEG data is continuously sent to the SSVEP classification and recognition module.

[0050] The SSVEP classification and recognition module is used to classify and recognize SSVEP signals based on the acquired 8-lead EEG data from the occipital region using FBCCA. FBCCA uses 1 second as the decoding data length, and the data shape is (8,250), where 8 is the number of leads and 250 is the number of sampling points. A reference signal with the same frequency as the flicker stimulus is preset, and the original signal is decomposed into sub-band signals of different frequency bands through a bandpass filter bank ([5,90],[14,90],[22,90],[30,90],[38,90]). Then, CCA calculates the correlation coefficient between each sub-band signal and the reference signal and sums them with weights. Finally, the category of the reference signal with the highest correlation is used as the recognition result.

[0051] The motor imagery data acquisition module is used to collect 9-lead EEG data from the parietal lobe as motor imagery data using an EEG acquisition device. The sampling frequency is also 250Hz, and the specific leads are set as follows: FC3, FCZ, FC4, C3, CZ, C4, CP3, CPZ, and CP4, with the leads positioned according to the international standard of 10-20 leads. Considering the brain's reaction time, only data within 0.5-4.5 seconds is used, i.e., the sample data shape is (9, 1000), where 9 is the number of leads and 1000 is the number of sample points. After a period of training, when the motor imagery data reaches the preset training sample size, the motor imagery decoding model training module trains the motor imagery decoding model.

[0052] The motion imagery decoding model training module is used to train the motion imagery decoding model, which is based on deep learning technology and uses a convolutional neural network (CNN) and a self-attention mechanism to process motion imagery data. The motion imagery data is first subjected to one-dimensional convolution with a kernel of scale (1, 25) to extract local refinement features; then, it is subjected to convolution along the lead channels with a kernel of scale (ch, 1), where ch is the number of leads (9). After this convolution process, the data is compressed into one-dimensional data of scale (1, Ns), where Ns is the number of sample points after convolution; then, the data is segmented into segments of scale (1, 40) to conform to the input of the self-attention mechanism, which extracts the global contextual dependencies of the EEG signal, further improving the model's decoding performance.

[0053] The comparison module is used to take the SSVEP classification and recognition results as the real category labels, use the trained motor imagery decoding model to identify the motor intention of the collected motor imagery data, and compare it with the real category labels. If the results are consistent, the collected motor imagery data is used to fine-tune the motor imagery decoding model. If they are inconsistent, the collected motor imagery data is discarded, and rehabilitation training is carried out based on the SSVEP classification and recognition results.

[0054] The accuracy statistics module for motor imagery recognition is used to calculate the accuracy of motor imagery recognition within a certain period. If the overall accuracy reaches 70%, peripheral visual stimulation can be canceled, and rehabilitation training can be carried out solely through motor imagery. If the accuracy is insufficient, motor imagery data will continue to be collected for model fine-tuning.

[0055] This system enables the conversion of peripheral visual stimuli into motor imagery, and avoids the tedious process of collecting training samples, efficiently collecting labeled motor imagery data.

[0056] Based on the above embodiments, such as Figure 2 As shown, the present invention also proposes a method for using a training-free neurorehabilitation system that combines peripheral visual stimulation and motor imagery, comprising:

[0057] The patient wears an EEG acquisition device and connects to a PC via TCP. The patient then uses the rehabilitation equipment according to the rules. Specifically, this neurorehabilitation system can be combined with various rehabilitation devices, such as exoskeleton robotic arms and hand function rehabilitation training devices.

[0058] The peripheral visual stimulation submodule generates bilateral peripheral visual stimuli by flashing at fixed frequencies of 12Hz and 14Hz, respectively. The motor imagery stimulation module plays left and right hand movement videos within the bilateral peripheral visual stimuli. Patients select a target on one side based on their movement intention, focusing their gaze on the central motor stimulus. The left and right sides represent left and right hand rehabilitation training, respectively. The stimulation paradigm is as follows: Figure 3 As shown.

[0059] The SSVEP data acquisition module collects 8-lead EEG data from the occipital region using an EEG acquisition device at a sampling frequency of 250Hz. The leads selected are P5, Pz, P6, PO7, O1, Oz, O2, and PO8, and the lead positions are placed according to the international standard of 10-20 leads. The EEG data is continuously transmitted from the EEG acquisition device to the PC via TCP. The PC backend uses the SSVEP classification and recognition module and employs FBCCA to perform real-time decoding and analysis of the SSVEP signals.

[0060] In the SSVEP classification and recognition module, FBCCA uses 1 second as the decoding data length, with a data shape of (8,250), where 8 represents the number of leads and 250 represents the number of sampling points. A reference signal with the same frequency as the flickering stimulus is preset, and the original signal is decomposed into sub-band signals of different frequency bands through a bandpass filter bank ([5,90],[14,90],[22,90],[30,90],[38,90]). Then, CCA calculates the correlation coefficient between each sub-band signal and the reference signal and performs a weighted summation. Finally, the category of the reference signal with the highest correlation is used as the recognition result.

[0061] SSVEP signal classification and recognition are completed. Peripheral visual stimulation flashing is paused for 1 second. The classification result is recorded as a label for motor imagery samples and saved. The patient is then reminded to perform the corresponding motor imagery action, such as... Figure 4 .

[0062] Peripheral visual stimulation disappears, while internal motor stimulation related to the SSVEP recognition result continues to play, activating the patient's mirror neurons. Playback on the other side stops. The SSVEP recognition result is converted into control commands and sent to the rehabilitation device, which then guides the patient in limb training. This process lasts 5 seconds, during which the patient's EEG signals are collected as sample data for motor imagery. Figure 5 .

[0063] The motor imagery data acquisition module collects EEG data from nine leads in the parietal lobe as sample data for motor imagery, with a sampling frequency of 250Hz. The specific leads are set as follows: FC3, FCZ, FC4, C3, CZ, C4, CP3, CPZ, and CP4. Considering the patient's brain reaction time, only data within the range of 0.5-4.5 seconds is used, i.e., the sample data shape is (9, 1000), where 9 represents the number of leads and 1000 represents the number of sample points. After a period of training, when the motor imagery data reaches a preset training sample size, the motor imagery decoding model is trained.

[0064] The motion imagery decoding model is trained using a motion imagery decoding model training module. This model employs deep learning technology, utilizing a convolutional neural network (CNN) and a self-attention mechanism. Motion imagery data is first convolved one-dimensionally with a kernel of scale (1, 25) to extract local refinement features. Then, it undergoes convolution along the lead channels with a kernel of scale (ch, 1), where ch represents the number of leads (9). After this convolution, the data is compressed into one-dimensional data of scale (1, Ns), where Ns is the number of sample points after convolution. The data is then segmented into (1, 40) segments to conform to the input of the self-attention mechanism, which extracts the global contextual dependencies of the EEG signal, further improving the model's decoding performance.

[0065] After model training is completed, SSVEP classification and recognition are performed first. The results are used as the true category labels by the comparison module. The trained motor imagery decoding model is then used to identify the movement intentions of the collected motor imagery data and compared with the true category labels. If the results match, the motor imagery data is used for model fine-tuning. If they do not match, the motor imagery data is discarded, and rehabilitation training is based on the SSVEP classification and recognition results.

[0066] The accuracy of motor imagery recognition is statistically analyzed over a certain period using the motor imagery recognition accuracy statistics module. If the overall accuracy reaches 70%, peripheral visual stimulation can be canceled, and rehabilitation training can be conducted solely through motor imagery. If the accuracy is insufficient, motor imagery data will continue to be collected for model fine-tuning.

[0067] In summary, the peripheral visual stimulation proposed in this invention can significantly reduce visual fatigue in patients compared to the traditional SSVEP stimulation paradigm. The SSVEP classification and recognition uses a training-free algorithm, combining peripheral visual stimulation with motor imagery. This allows for convenient acquisition of labeled motor imagery data, avoiding the tedious process of training sample collection and improving upon the initial poor recognition performance of motor imagery brain-computer interfaces, achieving a plug-and-play effect. During patient use, the motor imagery recognition results can be compared with the real labels generated by SSVEP analysis. If the accuracy meets the usage requirements, the transition from SSVEP to motor imagery can be completed; if the accuracy does not meet the requirements, the motor imagery decoding model can be fine-tuned step-by-step to ensure the performance of the rehabilitation system.

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery, characterized in that, include: The stimulation module includes a peripheral visual stimulation submodule and a motor imagery stimulation submodule. The peripheral visual stimulation submodule is used to generate peripheral visual stimulation to induce SSVEP EEG signals, and the peripheral visual stimulation adopts a circular stimulation. The motor imagery stimulation submodule is used to play left and right hand movement videos inside the circular stimulation to induce motor imagery EEG signals. The SSVEP data acquisition module is used to acquire 8-lead EEG data from the occipital region as SSVEP data. The SSVEP classification and identification module is used to classify and identify SSVEP signals based on the collected 8-lead EEG data of the occipital region through filter bank-based canonical correlation analysis (FBCCA). The motion imagery data acquisition module is used to collect EEG data from the 9 leads of the parietal lobe as motion imagery data; The motion imagery decoding model training module is used to train the motion imagery decoding model, which is based on deep learning technology and uses convolutional neural networks and self-attention mechanisms to process motion imagery data. The comparison module is used to take the SSVEP classification and recognition results as the real category labels, use the trained motor imagery decoding model to identify the motor intention of the collected motor imagery data, and compare it with the real category labels. If the results are consistent, the collected motor imagery data is used to fine-tune the motor imagery decoding model. If they are inconsistent, the collected motor imagery data is discarded, and rehabilitation training is carried out based on the SSVEP classification and recognition results. The accuracy statistics module for motor imagery recognition is used to calculate the accuracy of motor imagery recognition within a certain period. If the overall accuracy reaches 70%, peripheral visual stimulation can be canceled, and rehabilitation training can be carried out solely through motor imagery. If the accuracy is insufficient, motor imagery data will continue to be collected for model fine-tuning.

2. The training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery according to claim 1, characterized in that, The circular stimuli on the left and right sides flash at fixed frequencies of 12Hz and 14Hz, respectively.

3. The training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery according to claim 1, characterized in that, In the SSVEP data acquisition module, when acquiring EEG data from 8 leads in the occipital region, the leads are set to P5, Pz, P6, PO7, O1, Oz, O2 and PO8, and the lead positions are placed according to the international 10-20 lead standard.

4. The training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery according to claim 1, characterized in that, In the SSVEP classification and recognition module, FBCCA uses 1 second as the decoding data length, and the data shape is (8, 250), where 8 is the number of leads and 250 is the number of sampling points. A reference signal with the same frequency as the flicker stimulus is preset, and the original signal is decomposed into sub-band signals of different frequency bands through a bandpass filter bank. Then, canonical correlation analysis is used to calculate the correlation coefficient between each sub-band signal and the reference signal and the weighted summation is performed. Finally, the category of the reference signal with the highest correlation is used as the recognition result.

5. The training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery according to claim 1, characterized in that, In the motion imagery data acquisition module, when acquiring EEG data from the 9 leads of the parietal lobe, the leads are set as FC3, FCZ, FC4, C3, CZ, C4, CP3, CPZ, and CP4, and the lead positions are placed according to the international 10-20 lead standard.

6. The training-free neurorehabilitation system combining peripheral visual stimulation and motor imagery according to claim 1, characterized in that, In the training module of the motion imagery decoding model, the motion imagery data is first convolved in one dimension by a convolution kernel of scale (1,25) to extract local refinement features; then it is convolved along the lead channel by a convolution kernel of scale (ch,1), where ch is the number of leads, compressing the data into one-dimensional data of (1,Ns), where Ns is the number of sample points after convolution; then the data is segmented into (1,40) segments to conform to the input of the self-attention mechanism, and the global dependency of the EEG signal context is extracted by the self-attention mechanism.

Images