Eeg-based rehabilitation corpus screening method, system, and medium

Abstract

The application provides an EEG-based rehabilitation corpus screening method, system and medium, which provides corpus screening for a language function rehabilitation system based on a motor imagery brain-computer interface. Through screening, the questions in the original question bank can be classified, and the question bank with a higher activation degree of the motor cortex is screened out, which is used for corpus stimulation of the language function rehabilitation system based on the motor imagery brain-computer interface. The motor cortex of the rehabilitation patient is better activated from the corpus stimulation part, the functional connection of the motor cortex is activated, the recovery and reconstruction of the language function are helped, and the difficulty of imagination when the stimulation is expected can be reduced, so that a better rehabilitation effect is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of language function rehabilitation system based on brain-computer interface, in particular to a rehabilitation corpus used by a rehabilitation system, and more particularly to an EEG-based rehabilitation corpus screening method, system and computer readable storage medium. BACKGROUND

[0002] Post-stroke aphasia refers to a language disorder syndrome caused by brain damage, which is one of the common complications of acute stroke patients, and has a serious impact on the quality of life and prognosis of patients. Among various types of aphasia, motor aphasia patients often show normal auditory comprehension ability, but have difficulty in oral expression. Specific symptoms include limited language, only simple language, patients cannot distinguish sentence structure, and there are difficulties in repeating and writing.

[0003] The classic language pathway model considers the main language pathway as follows: starting from the Wernicke area (posterior temporal lobe), which receives information from the auditory and visual cortex and performs semantic understanding; the arcuate fasciculus connects the Wernicke area with the Broca area (posterior inferior frontal lobe); the Broca area is responsible for producing meaningful language and outputting to the motor cortex to initiate complex muscle movements required for speaking. Therefore, the damaged part of the brain of motor aphasia patients is mostly the Broca area and the motor cortex.

[0004] The motor function rehabilitation system based on motor imagery brain-computer interface is a new rehabilitation method for brain injury patients using brain-computer interface technology. By collecting and analyzing the brain electrical signals of patients, the characteristics related to motor imagery are identified and converted into computer instructions to stimulate the motor nerves and muscles of patients, promote the neural plasticity of brain injury patients, and help patients to realize the remodeling of brain function network and promote the recovery of motor function.

[0005] The language function rehabilitation system based on motor imagery brain-computer interface is targeted at the characteristics of motor aphasia, and makes patients perform motor imagery after receiving corpus stimulation during the language function rehabilitation reconstruction process. Starting from both ends of the classic language pathway model, it affects the Broca area and the motor cortex, and promotes neural plasticity. However, the corpus used by various language rehabilitation systems does not consider the specificity when stimulating motor aphasia patients, and the imagination difficulty of patients after receiving the stimulus is large, and the language function rehabilitation and reconstruction effect is not good. SUMMARY

[0006] In view of the defects and deficiencies of the prior art, the purpose of the present application aims to provide a corpus screening method for a language function rehabilitation system based on a motor imagery brain-computer interface, and screen a question bank with a high activation degree on the motor cortex, so that the motor cortex can be activated to a certain extent when a patient is stimulated by the corpus provided by the language function rehabilitation system, and the language rehabilitation process based on motor imagery can better activate the functional connection of the motor cortex, which is beneficial to promoting the remodeling of the brain functional network and helping the recovery and reconstruction of the language function.

[0007] According to a first aspect of the object of the present application, a rehabilitation corpus screening method based on EEG is provided, comprising:

[0008] obtaining a question bank containing a first number of questions;

[0009] randomly sampling a second number of questions from the question bank, and obtaining EEG data according to corpus stimulation of the subject object based on the second number of questions;

[0010] training a classification network based on the second number of questions and classification data related to the activation of the motor cortex in the EEG data;

[0011] predicting the first number of questions in the question bank based on the classification network, and screening a question bank with a first activation degree on the motor cortex.

[0012] As an optional implementation, the training of the classification network based on the second number of questions and the classification data related to the activation of the motor cortex in the EEG data comprises:

[0013] obtaining motor cortex relative energy proportion data based on the EEG data;

[0014] classifying C motor cortex relative energy proportion data to obtain a first classification label with a first activation degree on the motor cortex and a second classification label with a second activation degree on the motor cortex;

[0015] mapping the second number of questions in the question bank to token sequences to obtain C token sequence tensors;

[0016] using the C token sequence tensors and the first classification label and the second classification label corresponding to the C questions as training data to train a model to obtain a classification network based on a Transformer model.

[0017] According to a second aspect of the object of the present application, a computer system is provided, comprising:

[0018] one or more processors, and

[0019] a memory for storing instructions that can be executed;

[0020] wherein the instructions, when executed by one or more processors, implement the foregoing process of the EEG-based rehabilitation corpus screening method.

[0021] According to a third aspect of the object of the present application, a computer-readable medium storing a computer program is provided, the computer program comprising instructions executable by one or more processors, the instructions, when executed by the one or more processors, implementing the foregoing process of the EEG-based rehabilitation corpus screening method.

[0022] In combination with the technical solutions of the various aspects above, the EEG-based rehabilitation corpus screening method proposed by the present application has the following significant advantages:

[0023] For the language function rehabilitation training of the motor aphasia patient in the motor function rehabilitation system based on the motor imagery brain-computer interface, when the traditional question bank is used to stimulate the corpus, the motor cortex of the motor aphasia patient cannot be activated to a certain extent, and the language recovery rehabilitation effect is difficult to improve. The EEG-based rehabilitation corpus screening method proposed by the present application can screen the question bank with high activation degree of the motor cortex, so that the motor cortex of the patient can be activated to a certain extent when the patient is stimulated by the corpus provided by the language function rehabilitation system, the functional connection of the motor cortex is better activated, the brain function network remodeling is promoted, the motor imagery difficulty of the patient is reduced, and the language function reconstruction effect is improved.

[0024] It should be understood that all combinations of the foregoing concepts and additional concepts described in greater detail below (provided such concepts are not mutually inconsistent) are contemplated as being part of the inventive subject matter. Additionally, all combinations of claimed subject matter are contemplated as part of the inventive subject matter.

[0025] The foregoing and other aspects, embodiments and features of the present teachings can be better understood from the following description of the present teachings taken in conjunction with the accompanying drawings. Other features of the present teachings, such as exemplary embodiments thereof, will be apparent from the following description of the present teachings, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of the EEG-based rehabilitation corpus screening method provided according to an embodiment of the present application.

[0027] Figure 2 is a flowchart of the process of training and generating a classification network based on the second number of questions and the classification data related to the activation of the motor cortex in the EEG data according to an embodiment of the present application. is a flowchart of the process of training and generating a classification network based on the second number of questions and the classification data related to the activation of the motor cortex in the EEG data according to an embodiment of the present application.

[0028] Figure 3 is a flowchart of a process for obtaining motor cortex relative energy proportion data by preprocessing a tensor obtained in a Q-wheel EEG motor imagery experiment according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to better understand the technical content of the present application, specific embodiments are described below in conjunction with the accompanying drawings.

[0030] Aspects of the present application are described in the disclosure by reference to the accompanying drawings, in which a number of illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present application. It should be understood that various concepts and embodiments introduced above and those described in more detail below can be implemented in any of numerous ways, as the concepts and embodiments disclosed herein are not limited to any one implementation. Additionally, some aspects of the present application can be utilized independently, or in any suitable combination with other aspects of the present application.

[0031] Embodiment 1

[0032] In conjunction with Figure 1 The EEG-based rehabilitation corpus screening method of the illustrated embodiment includes the following steps:

[0033] Step S100: Obtain a question bank containing a first number of questions;

[0034] Step S200: Randomly sample a second number of questions from the question bank, and obtain EEG data by stimulating the corpus of the subject according to the second number of questions;

[0035] Step S300: Train a classification network based on the second number of questions and classification data related to motor cortex activation in the EEG data;

[0036] Step S400: Predict the first number of questions in the question bank based on the classification network, and screen out questions with a first activation degree on the motor cortex.

[0037] Therefore, the question bank with a first activation degree on the motor cortex (i.e., the question bank with a high activation degree) screened according to the method of the present application can activate the motor cortex of the language rehabilitation patient to a certain extent when the screened questions are used for corpus stimulation in the language function rehabilitation system, better activate the functional connectivity of the motor cortex, promote the remodeling of the brain functional network, and both reduce the difficulty of motor imagery of the patient and improve the effect of language function reconstruction.

[0038] In conjunction with Figure 1The flowchart shown and the example of the method described above, the EEG-based rehabilitation corpus screening method is configured to be implemented through four stages, i.e. construction of a question bank, corpus stimulation experiment based on a question sample, classification network training based on EEG data under the question sample and corpus stimulation, and question bank screening based on the classification network.

[0039] Construction of a question bank

[0040] As an optional example, in step S100, the operation of obtaining the question bank includes:

[0041] selecting at least one corpus;

[0042] segmenting sentences in the selected corpus;

[0043] Based on certain grammar and sentence pattern rules, the content of the sentences in the corpus is hidden to obtain a first number of questions, and a question bank is constructed, the questions being used to pre-stimulate the subjects in subsequent experiments to collect their EEG data (i.e. brain electrical data).

[0044] For ease of illustration, in the following embodiments, the number of questions contained in the question bank, i.e. the value of the first number, is represented by the parameter N.

[0045] The corpus can be selected from a general corpus, such as at least one of a Chinese corpus and an English corpus.

[0046] In another embodiment, the aforementioned selected corpus can also be selected from, but not limited to, a French corpus, a Russian corpus, a German corpus, a Japanese corpus, a Korean corpus, a Korean corpus, an Indian corpus, a Spanish corpus, a Portuguese corpus, an Italian corpus, a Dutch corpus, an Arabic corpus, a Thai corpus, a Malay corpus, an Indian corpus, a Polish corpus, etc.

[0047] The corpus can be a monolingual corpus or an aligned multilingual corpus (i.e. a corpus composed of two or more languages), such as a Chinese-English parallel bilingual corpus, a Chinese-Japanese parallel bilingual corpus, a Chinese-Korean parallel bilingual corpus, etc. The total number of sentences in the corpus can generally reach tens of thousands.

[0048] It should be understood that the question bank obtained through the method of the present application can be used in a suitable manner by the language function rehabilitation system, integrated in the language function rehabilitation system, and help the language function rehabilitation of patients with language function loss through corpus stimulation. Therefore, in optional embodiments, the selection of the corpus can correspond to the language of the language function rehabilitation system and the region where it is used, and the language of the rehabilitation object. In some embodiments, the language function rehabilitation system can integrate multiple question banks of different languages selected through the method of the present application for selection and use.

[0049] Corpus stimulation experiment based on question samples

[0050] As an optional example, in step S200, randomly sampling a second number of questions in the question bank means randomly selecting a plurality of questions from the first number of questions as samples. For ease of description, the value of the second number is represented by a parameter C. The value of the second number C is much smaller than the value of the first number N.

[0051] As an optional example, in step S200, the corpus stimulation of the test object according to the second number of questions obtains EEG data, including:

[0052] For each test object, the corpus stimulation is performed according to the first number of questions, the prompt time of each question is (n-m)s, the EEG signal during the prompt and m seconds after the prompt is collected, and the EEG data of ns is obtained;

[0053] Each test object completes Q rounds of experiments, and the EEG data of the test object under the corpus stimulation of the first number of questions is obtained, which is recorded as the first tensor [K, Q, C, Ch, n*f]; wherein K represents the total number of test objects, Q represents the number of experimental rounds for each test object, C represents the first number, Ch represents the number of EEG channels, f represents the EEG sampling frequency, and n represents the total time length of each question corpus stimulation.

[0054] In the examples of the present application, the aforementioned corpus stimulation can be realized by using the existing technology of electroencephalogram experiment paradigm, for example, through voice and vision for corpus stimulation, and through electroencephalogram acquisition electrodes or electroencephalogram acquisition caps integrated with electroencephalogram acquisition electrodes to collect the electroencephalogram signals of the test object under the corpus stimulation.

[0055] Classification network training based on question samples and EEG data under corpus stimulation

[0056] In optional examples, in step S300, the classification network is trained based on the second number of questions and the classification data related to the activation of the motor cortex in the EEG data, including:

[0057] Step S301: Obtain motor cortex relative energy proportion data based on the EEG data;

[0058] Step S302: Classify according to the C motor cortex relative energy proportion data, obtain a first classification label with a first activation degree to the motor cortex and a second classification label with a second activation degree to the motor cortex;

[0059] Step S303: Map the second number of questions in the question bank to the token sequence to obtain C token sequence tensors;

[0060] Step S304: Perform model training with the C token sequence tensors and the first classification label and the second classification label corresponding to the C questions as training data to obtain a classification network.

[0061] In an optional example, based on the EEG data, the motor cortex relative energy proportion data is obtained, including:

[0062] Based on the first tensor [K, Q, C, Ch, n*f], the total energy of each EEG channel in the 5-30Hz range is obtained;

[0063] The proportion of the motor cortex channel energy in the total energy is calculated to obtain C motor cortex relative energy proportion data.

[0064] As an optional example, in the method of the application, the C motor cortex relative energy proportion data is obtained by preprocessing the first tensor [K, Q, C, Ch, n*f] and energy proportion calculation, specifically including the following processes:

[0065] Step S3011: Calculate the average of the first tensor [K, Q, C, Ch, n*f] obtained by Q rounds of experiments and record it as a second tensor [K, C, Ch, n*f];

[0066] Step S3012: Based on the second tensor [K, C, Ch, n*f], Fourier transform and take the absolute value of the signal of each EEG channel to obtain a first amplitude spectrum tensor [K, C, Ch, F];

[0067] Step S3013: Based on the first amplitude spectrum tensor [K, C, Ch, F], calculate the average of the amplitude spectrum of the K subjects to obtain a second amplitude spectrum tensor [C, Ch, F];

[0068] Step S3014: Based on the second amplitude spectrum tensor [C, Ch, F], calculate the total energy of each EEG channel in the 5-30Hz range to obtain a tensor [C, Ch]; and

[0069] Step S3015: Calculate the proportion of the motor cortex channel energy in the total energy to obtain C motor cortex relative energy proportion data.

[0070] In an optional example, in step S302, the C motor cortex relative energy proportion data is classified to obtain a first classification label with a first activation degree to the motor cortex and a second classification label with a second activation degree to the motor cortex, including:

[0071] The unsupervised classification algorithm is used to perform cluster analysis on the C motor cortex relative energy proportion data to obtain two types of classification labels and a classification determination threshold J, wherein the motor cortex relative energy proportion data higher than the determination threshold J indicates a first activation degree to the motor cortex, and vice versa.

[0072] As an optional example, taking KNN algorithm as an example, two types of classification labels are obtained by KNN clustering, and a corresponding classification determination threshold J is obtained, wherein the motor cortex relative energy proportion data higher than the determination threshold J indicates a first activation degree to the motor cortex, and vice versa.

[0073] Thus, based on the EEG data obtained in the corpus stimulation experiment, the proportion of the motor cortex channel energy in the total energy of each EEG channel in the range of 5-30 Hz is calculated to obtain C motor cortex relative energy proportion data. Based on the motor cortex relative energy proportion data, cluster analysis is performed by a classification algorithm to obtain a first classification label with a high activation degree to the motor cortex and a second classification label with a low activation degree to the motor cortex, which is used for subsequent classification network training.

[0074] In an optional example, in step S303, the second number of questions in the question bank are mapped to token sequences to obtain C token sequence tensors, including:

[0075] Randomly sample the second number of questions to map them to token sequences, each token has a length of M, and the hidden content in each token is filled with a mask token to form C sequence samples;

[0076] Each of the C sequence samples is uniformly padded to a length L to obtain a token sequence tensor, denoted as [N, L, M], and the padding length L is the length corresponding to the longest sequence in the N sequence samples.

[0077] The mapping of the first quantity of questions in the question bank to the token sequence can be set to be based on an open-source Embedding model, and C questions randomly sampled from the question bank are mapped to token sequences.

[0078] In an optional example, each of the C sequence samples is uniformly padded to a length L, including:

[0079] Each of the C sequence samples is padded using a padding token, and the length of the padded sample is L.

[0080] In an embodiment of the present application, the foregoing classification network is configured to be implemented based on a deep learning model, including but not limited to RNN (Recurrent Neural Network), RetNet (Retentive Networks), and Transformer.

[0081] In step S304, the constructed data set is divided into a training set and a test set, for example, 70% of the data set is used as the training set to train the classification network based on the Transformer model, and 30% of the data set is used as the test set to test the performance of the model. The training process uses a ten-fold cross-validation method to verify the output results of the model and optimizes the network parameters to obtain a reliable and stable classification model.

[0082] In an embodiment of the present application, taking the Transformer model as an example, the classification network based on the Transformer model is trained to complete the prediction and output of the questions with the first activation degree (i.e., high activation degree) and the second activation degree (i.e., low activation degree) of the motor cortex, and to realize the screening of the question bank.

[0083] Question bank screening based on classification network

[0084] In an optional example, the classification network is used to predict the first quantity of questions in the question bank, and the question bank with the first activation degree of the motor cortex is screened, including:

[0085] Each question in the question bank is mapped to a token sequence to obtain N token sequence tensors, which are input into the classification network for prediction, and the token sequence with the first classification label and the corresponding question are output to screen the question bank with the first activation degree of the motor cortex.

[0086] It should be understood that in the implementation of this example, the way of mapping the first number of questions in the question bank to the token sequence to obtain N token sequence tensors is the same as the way of mapping the second number of questions to the token sequence to obtain C token sequence tensors as described above, and will not be repeated here.

[0087] So far, through the foregoing process, the EEG data is obtained based on the corpus stimulation of the selected C questions as samples, and after the EEG data is preprocessed and the model is trained, a binary classification network model is obtained, which can be used to classify the questions with high and low activation levels of the motor cortex, determine the questions with high activation levels of the motor cortex, construct a question bank with high activation levels of the motor cortex, and be used for corpus stimulation of a language function rehabilitation system based on motor imagery brain-computer interface. The motor cortex is better activated from the corpus stimulation part, and the difficulty of motor imagery for patients is reduced, and the effect of language function recovery and reconstruction for motor aphasia patients is improved.

[0088] Embodiment 2

[0089] In combination Figure 2 With the flowchart shown, we will more specifically illustrate the exemplary implementation of each aspect of the rehabilitation corpus screening method based on EEG in this embodiment.

[0090] Step 1, taking the Chinese corpus as an example, a question bank is formed using a public Chinese corpus. First, the software tool is used to segment the sentences in the corpus, and then some content in the sentences is hidden through certain grammar and sentence pattern rules, and finally all the questions are combined to form a question bank, and the number of samples in the question bank is N.

[0091] For example, using the BFSU DiSCUSS corpus of Beijing Foreign Studies University, first, the sentences in the corpus are segmented using the tool in the ltp library of Python, and then some content in the sentences is hidden through pre-set grammar and sentence pattern rules, such as the original sentence is I go to eat, and the hidden sentence is I go ____ or go to eat. The sentence after hiding the content is saved in order as a question to form a question bank, and the number of samples in the question bank is N, usually greater than 10000;

[0092] Step 2, randomly sample C samples in the question bank, C << N, recruit K subjects for the experiment, for example, in this embodiment, 20 questions are selected as samples.

[0093] Step 3, use the C samples to conduct electroencephalogram imagination experiments on the recruited K subjects.

[0094] The experimental paradigm of the electroencephalogram imagination experiment includes a preparation phase and a collection phase.

[0095] Preparation phase: Each subject was placed in a quiet room, seated in a comfortable chair, and wearing an EEG cap identical to that used in the motor imagery-based brain-computer interface language rehabilitation system. After the EEG signal stabilized, the acquisition phase began.

[0096] Acquisition phase: Using the same language stimulation method as the language function rehabilitation system based on motor imagery brain-computer interface, the selected questions are presented to the subjects as language prompts. The prompt time is 2 seconds, and EEG data are collected during the prompt period and 1 second after the prompt, that is, 3 seconds of EEG signals are collected until all C samples are presented.

[0097] During the experiment, each subject performed three rounds of experiments, and the obtained experimental data were recorded as a tensor [K, 3, C, Ch, 3*f].

[0098] Step 3: Preprocess the tensor [K, 3, C, Ch, 3*f] obtained from the EEG imagery experiment to obtain the relative energy share data of C motor cortices.

[0099] In this embodiment, combined with Figure 3 As shown in Figure 2, the calculation process for obtaining the relative energy proportion data of C motor cortices is as follows:

[0100] After calculating the average value of the data obtained from the three rounds of experiments for all subjects, a tensor of the form [K, C, Ch, 3*f] is obtained;

[0101] Furthermore, the signal of each EEG channel is subjected to a fast Fourier transform (e.g., based on 2F point FFT) and the absolute value is taken to obtain a first amplitude spectrum tensor of the form [K, C, Ch, F];

[0102] Furthermore, the amplitude spectra of K subjects are averaged to obtain a second amplitude spectrum tensor of the form [C, Ch, F].

[0103] Furthermore, the total energy of each EEG channel in the range of 5-30 Hz is calculated to obtain a tensor of the form [C, Ch];

[0104] Furthermore, the proportion of motor cortex channel energy to the total energy is calculated to obtain the relative energy proportion data of C motor cortices.

[0105] Step 4: Use the KNN clustering algorithm to cluster the C motor cortex relative energy proportion data into two categories, and obtain the classification judgment threshold. A value above the classification judgment threshold is defined as a high degree of activation of the motor cortex, and vice versa.

[0106] Step 5, map all N questions in the question bank to token sequences using an open-source Chinese Embedding model, each token has a length of M, and the hidden part is filled with mask tokens to form N sequence samples;

[0107] Further, all sequence lengths are padded to the length L of the longest sequence in the N samples using padding tokens, obtaining a token sequence tensor in the form of [N, L, M].

[0108] Step 6, form a data set by combining the token sequence tensor in the form of [C, L, M] corresponding to the C samples obtained by sampling in the brain electrical imagination experiment and the class labels formed by KNN clustering in the foregoing steps, and train the classification network.

[0109] Among them, 30% of the constructed data set is used to test the model performance, and 70% is used to train the classification network based on the Transformer, and the results of ten-fold cross-validation are used as the standard to optimize the network parameters, and a stable and reliable binary classification network is obtained.

[0110] Step 7, use the trained binary classification network to predict all token sequences generated by mapping in the question bank, determine the token sequences with the first classification label and the corresponding questions, and thus filter to obtain a question bank with high activation degree of the motor cortex.

[0111] Therefore, by screening, the questions in the original question bank can be classified, and a question bank with higher activation degree of the motor cortex is obtained, which is used for corpus stimulation of a language function rehabilitation system based on motor imagination brain-computer interface, reduces the difficulty of imagination when anticipating stimulation, and better activates the motor cortex from the corpus stimulation part, thereby achieving better rehabilitation effect.

[0112] Embodiment 3

[0113] In combination with the implementation of the above embodiment of the EEG-based rehabilitation corpus screening method, according to the disclosed embodiments of the present application, a computer system is also proposed, comprising one or more processors and a memory for storing instructions that can be operated.

[0114] Among them, the instructions realize the process of the EEG-based rehabilitation corpus screening method of any of the foregoing embodiments when executed by one or more processors.

[0115] Embodiment 4

[0116] In combination with the implementation of the EEG-based rehabilitation corpus screening method of the above embodiments, according to the embodiments disclosed in the present application, a computer readable medium storing a computer program is also proposed, the computer program includes instructions executable by one or more processors, and the instructions, when executed by the one or more processors, implement the process of the EEG-based rehabilitation corpus screening method of any of the foregoing embodiments.

[0117] As optional examples, the aforementioned computer readable medium can be implemented by including, but not limited to, random access memory, read-only memory, electrically erasable programmable memory, optical disk memory, magnetic disk memory, and a combination of the aforementioned types of computer readable storage media.

[0118] Although the present application has been disclosed in the above preferred embodiments, it is not intended to limit the present application. Those skilled in the art, without departing from the spirit and scope of the present application, can make various modifications and improvements. Therefore, the protection scope of the present application shall be subject to the scope defined by the claims.

Claims

1. A rehabilitation corpus screening method based on EEG, characterized in that: include: Obtaining a question bank, wherein the question bank includes a first number of questions; randomly sampling a second number of questions from the question bank, stimulating the subject with corpus data according to the second number of questions, and obtaining EEG data; training a classification network based on a second number of questions and classification data related to motor cortex activation in the EEG data; Predicting a first number of questions in the question bank based on the classification network, and screening out question banks having a first activation degree on the motor cortex; The step of stimulating the subject with corpus data according to the second number of questions to obtain EEG data includes: Each subject is stimulated with speech and visual language materials according to the first number of questions. The prompt time for each question is (nm) seconds. EEG signals are collected during the prompt period and m seconds after the prompt period to obtain n seconds of EEG data. Each subject completed Q rounds of experiments, and the EEG data of the subject under the stimulation of the first number of question corpus were obtained and recorded as the first tensor [K, Q, C, Ch, n*f]; where K represents the total number of subjects, Q represents the number of experimental rounds for each subject, C represents the value of the first quantity, Ch represents the number of EEG channels, f represents the EEG sampling frequency, and n represents the total duration of each question corpus stimulation; Furthermore, the training and generating of a classification network based on the second number of questions and the classification data related to motor cortex activation in the EEG data includes: Based on the EEG data, obtaining motor cortex relative energy proportion data; Classify the C motor cortexes according to their relative energy proportion data to obtain a first classification label with a first activation degree for the motor cortex and a second classification label with a second activation degree for the motor cortex; Map the second number of questions in the question bank to token sequences to obtain C token sequence tensors; Model training is performed using the C token sequence tensors and the first classification labels and the second classification labels corresponding to the C questions as training data to obtain a classification network based on the Transformer model.

2. The EEG-based rehabilitation corpus screening method according to claim 1, characterized in that: The step of obtaining motor cortex relative energy proportion data based on the EEG data includes: Based on the first tensor [K, Q, C, Ch, n*f], obtain the total energy of each EEG channel in the range of 5-30 Hz; The proportion of the motor cortex channel energy in the total energy is calculated to obtain relative energy proportion data of C motor cortices.

3. The EEG-based rehabilitation corpus screening method according to claim 1, characterized in that: The classifying according to the C motor cortex relative energy proportion data to obtain a first classification label with a first activation degree for the motor cortex and a second classification label with a second activation degree for the motor cortex includes: An unsupervised classification algorithm is used to perform cluster analysis on the relative energy proportion data of C motor cortexes to obtain two types of classification labels and a classification judgment threshold J, wherein: if the relative energy proportion data of the motor cortex is higher than the judgment threshold J, it indicates that the motor cortex has a first degree of activation, otherwise it indicates that the motor cortex has a second degree of activation.

4. The EEG-based rehabilitation corpus screening method according to claim 1, characterized in that: Mapping the second number of questions in the question bank to token sequences to obtain C token sequence tensors includes: Map the second number of randomly sampled questions into a token sequence, where each token has a length of M. The hidden content in each token is filled with a mask token to form C sequence samples; Each of the C sequence samples is padded uniformly according to the length L to obtain a token sequence tensor, which is recorded as [N, L, M]. The padded length L is the length corresponding to the longest sequence in the N sequence samples.

5. The EEG-based rehabilitation corpus screening method according to claim 4, characterized in that: The padding of each of the C sequence samples uniformly according to the length L includes: Use padding token to fill each sample in the C sequence samples, and the length of the padded sample is L.

6. The EEG-based rehabilitation corpus screening method according to any one of claims 1 to 5, characterized in that: The step of predicting a first number of questions in the question bank based on the classification network and selecting a question bank having a first activation degree for the motor cortex includes: Map each question in the question bank to a token sequence to obtain N token sequence tensors, input the classification network for prediction, and output a token sequence with a first classification label and its corresponding question to screen out the question bank with a first activation degree for the motor cortex.

7. A computer system, characterized in that: include: one or more processors, and a memory for storing instructions that can be operated; Wherein, when the instructions are executed by the one or more processors, the process of the EEG-based rehabilitation corpus screening method according to any one of claims 1 to 6 is implemented.

8. A computer-readable medium storing a computer program, characterized in that: The computer program includes instructions that can be executed by one or more processors, and when the instructions are executed by the one or more processors, the process of the EEG-based rehabilitation corpus screening method according to any one of claims 1 to 6 is implemented.

Images