Epilepsy electroencephalogram signal recognition method, system and equipment and storage medium

Abstract

The invention discloses an epilepsy electroencephalogram signal recognition method, system and device and a storage medium, and the method comprises the steps: obtaining an electroencephalogram signal, and calculating phase synchronization information between different channels according to the electroencephalogram signal as an edge feature; filtering the electroencephalogram signals, respectively calculating and extracting time domain features, dynamic features and phase amplitude coupling features corresponding to different wavebands of each channel, and fusing the features into node features; and integrating the obtained edge features and node features into a graph, performing classification by using a graph attention network model, and performing judgment and recognition on the electroencephalogram signals according to a classification result. The invention provides a classification method which is high in performance, has priori interpretability and is used for epilepsy signal identification and prediction, not only considers an interaction relationship among different channels, but also pays attention to and extracts deep features of time domain, dynamics, cross-frequency coupling and the like of nodes, and is more accurate in classification of electroencephalogram signals in different periods.

Description

Technical Field

[0001] The present invention belongs to the field of cognitive neuroscience and information technology, and relates to an epileptic electroencephalogram signal recognition method, system, device and storage medium. Background Art

[0002] The diagnosis of epilepsy has always been a hot topic of concern. As a complex neurological disease, accurate and timely diagnosis of epilepsy is of vital importance to the patient's quality of life, the formulation of treatment plans and prognosis assessment. Traditional epilepsy diagnosis is highly dependent on electroencephalogram (EEG), among which EEG data is an important basis for judging epilepsy. The diagnosis of epilepsy based on EEG generally requires experts to make judgments based on professional knowledge, which requires a lot of manpower and material resources as a cost, and is also affected by multiple factors such as doctor's experience, fatigue and subjective judgment. It may lead to inconsistency and delay in diagnostic results, and it is difficult to predict in advance.

[0003] In recent years, with the rapid development of computer science and artificial intelligence technology, the introduction of machine learning, especially deep learning, has made it possible to automatically diagnose and predict epilepsy. Deep learning models can automatically learn complex feature representations from massive EEG data and make predictions based on them, which is expected to significantly improve the accuracy and efficiency of diagnosis and reduce the burden on medical resources. However, despite significant progress in this field, it still faces many challenges and limitations of existing technologies.

[0004] First, EEG data itself has a high degree of complexity and nonlinear characteristics, and the functional connection between different brain regions is the key to understanding the mechanism of epileptic seizures. However, existing studies have regarded the EEG signals of different channels as independent of each other, or only considered the spatial position relationship between different channels in the EEG signals, without noticing that there are functional connections across multiple brain regions between different brain regions. This simplified processing method is difficult to fully capture the dynamic evolution of epileptic activity in the brain, limiting the accuracy and reliability of the diagnostic model. Secondly, each EEG channel contains rich physiological and pathological information, which is crucial for identifying the unique electrophysiological patterns of epilepsy. However, most studies on epilepsy EEG diagnosis and prediction based on graph neural networks (GNNs) focus more on the connection relationship between different channels, and pay very limited attention to the information of the channels themselves, which easily ignores the spatial and functional connections between different channels. As a result, the model may lose key information in the feature extraction stage, which in turn affects the subsequent diagnostic prediction performance. Furthermore, the interpretability of the model is poor. Although the deep learning model can achieve impressive prediction results, its decision-making process is often like a "black box", and it is difficult for clinicians to directly use its results as a reference. Summary of the invention

[0005] The purpose of the present invention is to provide a method, system, device and storage medium for identifying epileptic EEG signals to solve the problem of lack of accuracy and reliability in diagnosis based on existing deep learning models.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A method for identifying epileptic EEG signals, comprising: Obtain EEG signals, and calculate phase synchronization information between different channels as edge features based on the EEG signals; Filter the EEG signal, calculate and extract the time domain features, dynamic features, and phase-amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The obtained edge features and node features are integrated into a graph, classified using the graph attention network model, and the EEG signals are judged and identified based on the classification results.

[0007] Furthermore, the phase synchronization information between different channels is

[0008] In the formula, represents the length of the time series, , Representation Node The instantaneous phase of the corresponding time series, represents the imaginary part of the phase difference, is a symbolic function.

[0009] Furthermore, the time domain features include the maximum amplitude and energy of the EEG signal; The maximum amplitude of the EEG signal is

[0010] In the formula, represents the time series of the corresponding band, Represent four different bands respectively; The energy of the EEG signal is

[0011] In the formula, Indicates the sampling frequency.

[0012] Furthermore, the calculation method of the dynamic characteristics is: The time series Refactored to dimensional vector sequence:

[0013] in, , based on the reconstruction , calculate its corresponding :

[0014] in, Indicates the set threshold. is the Heaviside function, we get Then, calculate the entire vector sequence :

[0015] Set the dimension to , repeat the above process to get , and calculate the time series of length L dimension, Sample entropy under threshold conditions:

[0016] Calculate the sample entropy of a single channel in different bands , and finally the entropy feature matrix of the channel is obtained .

[0017] Furthermore, the calculation method of the phase amplitude coupling characteristic is: Filter the original time series x into a high-frequency series With low frequency sequence Two frequency bands, and then use Hilbert transform to calculate the instantaneous phase of the low-frequency sequence The amplitude envelope of the high frequency sequence , divide the range into K partitions evenly, for each , the partition corresponds to The average , and get the normalized value :

[0018] Get sequence , and calculate the Shannon entropy of the sequence:

[0019] It is obtained by the following formula :

[0020] Calculate the MI index of the low-frequency δ band for the θ, α, and β bands respectively , get the phase amplitude coupling characteristic matrix of each channel .

[0021] Furthermore, the graph attention network model includes a graph attention network and a node feedforward network; The working mechanism of the graph attention network is: First, calculate the attention index of different nodes

[0022] Where W∈RF′×F is a learnable parameter matrix, For the shared attention mechanism;

[0023] In the formula, ∈R 2F′, For transposition, For the merge operation;

[0024] In the formula, Representation Node The first-order neighbor nodes of

[0025] Where h represents the hth attention operation, Represents a merge operation. represents the activation function; The mechanism of action of the node feedforward network is:

[0026] Where, W1,W2∈RF×F′,RF′×F′′, ∈RF′, RF′′ are the learnable coefficient matrix and bias respectively, F represents the dimension of input data, and F′ represents the dimension of output data. Represented as Sigmoid activation function.

[0027] Furthermore, the EEG signal includes a pre-ictal signal, a pre-ictal signal and an ictal signal.

[0028] An epileptic EEG signal recognition system, comprising: An edge feature calculation module, wherein the edge feature calculation module is used to obtain an EEG signal and calculate phase synchronization information between different channels as an edge feature according to the EEG signal; A node feature calculation module, which is used to filter the EEG signal, respectively calculate and extract the time domain features, dynamic features and phase amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The judgment and recognition module is used to integrate the obtained edge features and node features into a graph, use the graph attention network model to perform classification, and judge and recognize the EEG signal according to the classification results.

[0029] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0030] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method are implemented.

[0031] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for identifying epileptic EEG signals, which calculates the phase synchronization information between different channels as edge features based on EEG signal data, and filters and divides the EEG signals into multiple bands, respectively calculates and extracts the time domain features, dynamic features and phase amplitude coupling features on different bands of each channel, and merges them into node features. The calculated edge features and node features are integrated into a graph, and the graph attention network model is used for classification, and the EEG signals are judged to be non-ictal, pre-ictal and ictal signals, respectively, which plays a role in epilepsy prediction. The present invention uses a deep learning model based on GAT to propose a high-performance, a priori interpretable classification method for epilepsy signal detection and prediction. Compared with the existing research on epilepsy detection, the present invention not only considers the interaction between different channels in the EEG signal, but also pays attention to and extracts the deep-level features of the node itself, such as time domain, dynamics, and cross-frequency coupling, and integrates these two types of information into a graph signal. The deep-level information in the EEG signal can be more comprehensively mined, so that the model can classify EEG signals of different periods more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 It is a schematic diagram of the overall structure of the present invention.

[0034] Figure 2 This is a diagram of the phase amplitude coupling index calculation process of the present invention.

[0035] Figure 3 It is a schematic diagram of the network structure of the present invention.

[0036] Figure 4 Graph showing the loss rate and accuracy of the model of Example 1 of the present invention.

[0037] Figure 5 This is a confusion matrix diagram of the accuracy of the last round of training model of Example 1 of the present invention.

[0038] Figure 6 This is a schematic diagram of the structure of an epileptic EEG signal recognition system according to a preferred embodiment of the present invention.

[0039] Figure 7 It is a schematic diagram of the structure of an electronic device according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0041] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0043] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0044] It should be noted that the terminals involved in the embodiments of the present application may include but are not limited to mobile phones, personal digital assistants (PDAs), wireless handheld devices, tablet computers, personal computers (PCs), MP3 players, MP4 players, wearable devices (for example, smart glasses, smart watches, smart bracelets, etc.), smart home devices and other smart devices.

[0045] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0046] The present invention is further described in detail below in conjunction with the accompanying drawings: See also Figure 1 The present invention provides a method for identifying epileptic EEG signals, which specifically comprises the following steps: Step 1: First, calculate the phase synchronization information between different channels based on the original EEG data samples As an edge feature:

[0047] In the formula, represents the length of the time series, , Representation Node The instantaneous phase of the corresponding time series, represents the imaginary part of the phase difference, is a sign function. The range of PLI is between 0 and 1. The larger the value, the stronger the phase synchronization between the two regions, that is, the closer the functional connection between the two regions is. Conversely, the relationship between the two regions is more independent.

[0048] Based on the PLI index, the connection relationship between different nodes is calculated, and the functional connection binary matrix corresponding to the original EEG signal is constructed. , each element in the matrix The calculation formula is as follows:

[0049] in, is the set threshold. When the PLI between two nodes exceeds the threshold, it is considered that there is a connection between the two channels. Otherwise, there is no connection. ∈RN×N, N is the number of channels of the EEG signal.

[0050] Step 2: Filter the original EEG signal sample signal into For the four bands, the time domain features, dynamic features, and phase-amplitude coupling features corresponding to different bands of each channel are calculated and extracted respectively, and they are fused into node features.

[0051] Time domain characteristics: For EEG signals in different bands, calculate their maximum amplitudes respectively and energy As the time domain characteristics of the signal, the calculation methods are as follows

[0052]

[0053] In the formula, is the time series of the corresponding band, and its maximum absolute value is taken as the amplitude information of the band ; is the sampling frequency, The time integral of the square of is taken as the energy information of the band.

[0054] Dynamic characteristics: For a time series , refactoring it into dimensional vector sequence:

[0055] in, , based on the reconstruction , calculate its corresponding :

[0056] in, is the threshold value set, is the Heaviside function, we get Then, calculate the entire vector sequence , the calculation formula is as follows:

[0057] Next, set the dimensions to , repeat the above operation to get , and calculate the time series of length L dimension, The sample entropy under threshold conditions is:

[0058] set up is 2, For time series 0.2 times the standard deviation, calculate the sample entropy of a single channel in different bands , and finally the entropy feature matrix of the channel is obtained , where the subscripts represent the different bands of the channel.

[0059] Phase amplitude coupling characteristics: Filter the original time series x into a high-frequency series With low frequency sequence Two frequency bands, and then use Hilbert transform to calculate the instantaneous phase of the low-frequency sequence The amplitude envelope of the high frequency sequence , where the range of the instantaneous phase is [-180°, 180°]. Divide the range into K partitions evenly. , the partition corresponds to The average , and get the normalized value , the calculation process is as follows Figure 2 shown.

[0060]

[0061] This gives the sequence , and calculate the Shannon entropy of the sequence:

[0062] Finally, we get :

[0063] In the formula, is the instantaneous phase of the low-frequency sequence, is the amplitude envelope of the high frequency sequence, , set K=18, and calculate the MI index of the low-frequency δ band for the θ, α, and β bands respectively , get the phase amplitude coupling characteristic matrix of each channel .

[0064] The original EEG signal is reconstructed into graph structure data through the functional connectivity, time domain, dynamics, and phase amplitude coupling characteristics obtained by the above calculations. , where the edge The binary PLI function matrix calculated in step 1 ,node ,in .

[0065] Step 3: Integrate the calculated edges and nodes into a graph and input it into the graph attention network model. The graph attention network model mainly includes the graph attention network and the node feedforward network. The specific mechanism of the graph attention network is shown in the following formula: First, calculate the attention index of different nodes

[0066] Where W∈RF′×F is a learnable parameter matrix, It is a shared attention mechanism.

[0067]

[0068] In the formula, ∈R 2F′, For transposition, For merge operation.

[0069]

[0070] In the formula, Representation Node The first-order neighbor nodes

[0071] Where h represents the hth attention operation, Represents a merge operation. Represents the activation function.

[0072] The mechanism of the node feedforward network is as follows:

[0073] Where, W1,W2∈RF×F′,RF′×F′′, ∈RF′, RF′′ are the learnable coefficient matrix and bias respectively, F represents the dimension of input data, and F′ represents the dimension of output data. Represented as Sigmoid activation function.

[0074] Step 4: The specific structural diagram of the classification algorithm is as follows Figure 3 As shown in the figure, the data processed by the above steps is input into the attention network model of the figure to obtain the final classification result, and to determine whether the original EEG signal is a pre-ictal, pre-ictal or ictal signal.

[0075] Embodiment 1: This embodiment provides a method for identifying epileptic EEG signals, comprising the following steps: 1) Obtain preprocessed EEG signal data. In this embodiment, the dataset used is the CHB-MIT dataset, an open source scalp EEG signal dataset recorded at Children's Hospital Boston. These scalp EEG signals are measured by multiple electrodes placed at fixed positions on the subject's scalp, and are used to record multi-channel electrophysiological signals of brain activity. The dataset records a total of 198 continuous EEG data of epilepsy records, which were recorded from a total of 22 patients aged between 1.5 and 22 years old, including 17 female patients and 5 male patients. Among them, the data of CHB21 was obtained by retesting the CHB01 patient 1.5 years later, and some information of CHB24 is unknown. The data was measured at a resolution of 256 Hz and 16 bit, and the position of the electrodes followed the international 10~20 system. In this embodiment, we selected the CHB01 patient as the research subject, and for the channel selection of the brain area, we selected 22 channels that were included in most of the records.

[0076] 2) Bandpass filtering: In this embodiment, a 3rd order Butterworth filter is used to perform a 0.1-40 Hz bandpass filtering on the original EEG signal to eliminate the DC and high frequency signal interference in the EEG signal.

[0077] 3) This embodiment uses a window size of 5s and a sliding step size of 3s to segment the EEG signal data of different periods, and obtains nearly 10,000 EEG signal samples of different categories with a time length of 5s.

[0078] 4) Calculate the functional connection binary matrix between different channels of the original EEG signal data sample As edge feature ; and filter the original EEG sample signal into Four bands, respectively calculate and extract the time domain, dynamics, phase amplitude coupling and other information corresponding to different bands of each channel, and merge them into node features Based on the calculated node information and edge information, the original sample is reconstructed into a graph structure data sample for subsequent model training. .

[0079] 5) Integrate the calculated edges and nodes into a graph and input it into the classification algorithm based on the graph attention network. The structural diagram is as follows: Figure 3 As shown, the final classification results are obtained, and the original EEG signals are judged to be non-ictal, pre-ictal and ictal signals respectively.

[0080] like Figure 4As shown in the figure, it represents the loss rate and accuracy of the model. "Training loss rate" and "Validation loss rate" represent the loss rate of the model on the training set and validation set respectively. It is an indicator to measure the difference between the model prediction results and the actual results. The lower the loss rate, the closer the model prediction results are to the actual results, and the better the model performance. "Training accuracy rate" and "Validation accuracy rate" represent the accuracy of the model on the training set and validation set respectively. It is also an indicator to measure the difference between the model prediction results and the actual results. The closer to 1, the more accurate the classification results. Figure 4 The loss rate of the model finally stabilizes and stabilizes at a smaller value, and the accuracy of the model finally stabilizes and stabilizes at a higher value. The following points can be drawn: ① The model has converged: During the training process, the value of the loss function gradually decreases and tends to be stable, indicating that the model has learned the main features of the data and is no longer significantly improved.

[0081] ② Training has been completed: During the optimization process, the value of the loss function becomes smaller and smaller, and finally stabilizes at a smaller value, which means that the model parameters have found a better solution.

[0082] ③Stable model performance: The stability of the loss rate indicates that the performance of the model has reached a stable level, which means that the error has been minimized.

[0083] ④Good generalization ability: The stability of loss rate and accuracy mainly reflects the performance of the model, because overfitting models usually show unstable loss rate and accuracy during training.

[0084] ⑤ Reasonable parameter setting: The stability of loss rate and accuracy indicates that the parameters (learning rate, batch size, etc.) are properly set, and the model can converge to a better solution within a limited number of rounds.

[0085] like Figure 5 As shown, the confusion matrix of the model is shown. In the confusion matrix, "true value" represents the actual label of the data, and "predicted value" represents the label predicted by the model. The numbers in the nine cells represent the sample counts of the model's prediction results that match the corresponding labels. The "accuracy" numerical evaluation indicator in the matrix represents the accuracy, which represents the model's ability to distinguish each type of sample. The average accuracy of the model is 93.81%. The "precision" numerical evaluation indicator in the matrix represents the accuracy, which measures the accuracy of the model in predicting positive samples. Figure 5 The accuracy and precision of the model are both above 90%, indicating that the model has good overall prediction performance for the data set, high reliability, and can capture positive samples well, thereby better distinguishing EEG signals at different periods.

[0086] Embodiment 2: Embodiment 2 provided by the present invention is an embodiment of the epilepsy EEG signal recognition system provided by the present invention, such as Figure 6 As shown, the embodiment of the system includes: an edge feature calculation module, a node feature calculation module and a judgment and identification module.

[0087] An edge feature calculation module, wherein the edge feature calculation module is used to obtain an EEG signal and calculate phase synchronization information between different channels as an edge feature according to the EEG signal; A node feature calculation module, which is used to filter the EEG signal, respectively calculate and extract the time domain features, dynamic features and phase amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The judgment and recognition module is used to integrate the obtained edge features and node features into a graph, use the graph attention network model to perform classification, and judge and recognize the EEG signal according to the classification results.

[0088] It can be understood that the epilepsy EEG signal recognition system provided by the present invention corresponds to the epilepsy EEG signal recognition method provided by the aforementioned embodiments. The relevant technical features of the epilepsy EEG signal recognition system can refer to the relevant technical features of the epilepsy EEG signal recognition method, which will not be repeated here.

[0089] Another object of the present invention is to provide an electronic device, such as Figure 7 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the steps of the epileptic EEG signal recognition method.

[0090] The epileptic EEG signal recognition method comprises the following steps: Obtain EEG signals, and calculate phase synchronization information between different channels as edge features based on the EEG signals; Filter the EEG signal, calculate and extract the time domain features, dynamic features, and phase-amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The obtained edge features and node features are integrated into a graph, classified using the graph attention network model, and the EEG signals are judged and identified based on the classification results.

[0091] A fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the epileptic EEG signal recognition method are implemented.

[0092] The epileptic EEG signal recognition method comprises the following steps: Obtain EEG signals, and calculate phase synchronization information between different channels as edge features based on the EEG signals; Filter the EEG signal, calculate and extract the time domain features, dynamic features, and phase-amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The obtained edge features and node features are integrated into a graph, classified using the graph attention network model, and the EEG signals are judged and identified based on the classification results.

[0093] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying epileptic EEG signals, characterized in that: include: Obtain EEG signals, and calculate phase synchronization information between different channels as edge features based on the EEG signals; Filter the EEG signal, calculate and extract the time domain features, dynamic features, and phase-amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The obtained edge features and node features are integrated into a graph, classified using the graph attention network model, and the EEG signals are judged and identified based on the classification results.

2. The method for identifying epileptic EEG signals according to claim 1, characterized in that: The phase synchronization information between different channels is: In the formula, represents the length of the time series, , Representation Node The instantaneous phase of the corresponding time series, represents the imaginary part of the phase difference, is a symbolic function.

3. The method for identifying epileptic EEG signals according to claim 1, characterized in that: The time domain features include the maximum amplitude and energy of the EEG signal; The maximum amplitude of the EEG signal is In the formula, represents the time series of the corresponding band, Represent four different bands respectively; The energy of the EEG signal is In the formula, Indicates the sampling frequency.

4. The method for identifying epileptic EEG signals according to claim 1, characterized in that: The calculation method of the dynamic characteristics is: The time series Refactored to dimensional vector sequence: in, , based on the reconstruction , calculate its corresponding : in, Indicates the set threshold. is the Heaviside function, we get Then, calculate the entire vector sequence : Set the dimension to , repeat the above process to get , and calculate the time series of length L dimension, Sample entropy under threshold conditions: Calculate the sample entropy of a single channel in different bands , and finally the entropy feature matrix of the channel is obtained .

5. The epileptic EEG signal recognition method according to claim 1, characterized in that: The calculation method of the phase amplitude coupling characteristic is: Filter the original time series x into a high-frequency series With low frequency sequence Two frequency bands, and then use Hilbert transform to calculate the instantaneous phase of the low-frequency sequence The amplitude envelope of the high frequency sequence , divide the range into K partitions evenly, for each , the partition corresponds to The average , and get the normalized value : Get sequence , and calculate the Shannon entropy of the sequence: It is obtained by the following formula : Calculate the MI index of the low-frequency δ band for the θ, α, and β bands respectively , get the phase amplitude coupling characteristic matrix of each channel .

6. The epileptic EEG signal recognition method according to claim 1, characterized in that: The graph attention network model includes a graph attention network and a node feedforward network; The working mechanism of the graph attention network is: First, calculate the attention index of different nodes Where W∈RF′×F is a learnable parameter matrix, For the shared attention mechanism; In the formula, ∈R 2F′, For transposition, For the merge operation; In the formula, Representation Node The first-order neighbor nodes of Where h represents the hth attention operation, Represents a merge operation. represents the activation function; The mechanism of action of the node feedforward network is: Where, W1,W2∈RF×F′,RF′×F′′, ∈RF′, RF′′ are the learnable coefficient matrix and bias respectively, F represents the dimension of input data, and F′ represents the dimension of output data. Represented as Sigmoid activation function.

7. The method for identifying epileptic EEG signals according to claim 1, characterized in that: The EEG signals include pre-ictal signals, pre-ictal signals and ictal signals.

8. An epileptic EEG signal recognition system, characterized in that: include: An edge feature calculation module, wherein the edge feature calculation module is used to obtain an EEG signal and calculate phase synchronization information between different channels as an edge feature according to the EEG signal; A node feature calculation module, which is used to filter the EEG signal, respectively calculate and extract the time domain features, dynamic features and phase amplitude coupling features corresponding to different bands of each channel, and fuse them into node features; The judgment and recognition module is used to integrate the obtained edge features and node features into a graph, use the graph attention network model to perform classification, and judge and recognize the EEG signal according to the classification results.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images