Epileptic seizure prediction method and system based on power spectrum and phase spectrum of electroencephalogram

Abstract

The invention provides an epileptic seizure prediction method and system based on a power spectrum and a phase spectrum of an electroencephalogram, and the system comprises an electroencephalogram signal obtaining module which is configured to obtain a multi-channel electroencephalogram signal and carry out the preprocessing of the multi-channel electroencephalogram signal; a power spectrum and phase spectrum extraction module configured to divide the electroencephalogram signal into a plurality of segments of set time, transform each segment of set time, and extract a power spectrum and a phase spectrum based on the transform; the model construction module is configured to construct a double-flow convolutional neural network model, and the model comprises a first network flow and a second network flow; the feature fusion module is configured to splice the output features of the first network flow and the second network flow in a channel dimension, fuse power and phase information, and obtain a fused feature map; and the prediction module is configured to convert the fused feature map into a one-dimensional feature vector through further processing, and calculate the probability of epileptic seizure and non-seizure according to the one-dimensional feature vector to output the probability of epileptic seizure and non-seizure.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical signal processing and artificial intelligence technology, and in particular relates to an epileptic seizure prediction method and system based on the power spectrum and phase spectrum of an electroencephalogram. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Epilepsy is a common neurological disease that affects tens of millions of people worldwide. Accurate detection of epileptic seizures is crucial for the diagnosis, treatment, and management of patients. Traditional epileptic seizure detection mainly relies on manual analysis of electroencephalogram (EEG) signals by neurologists, but this method is time-consuming and easily affected by subjective factors, making it difficult to detect epileptic seizures in a timely and accurate manner.

[0004] In order to improve the efficiency and accuracy of epileptic seizure detection, researchers have developed various automated detection methods. Among them, time-frequency analysis methods such as short-time Fourier transform (STFT) and wavelet transform (WT) have been widely used to extract the time-frequency features of EEG signals. However, these methods have a trade-off between time and frequency resolution, and it is difficult to obtain high time and high frequency resolution at the same time.

[0005] As an improved time-frequency analysis method, S-transform combines the advantages of STFT and WT, and has a frequency-dependent Gaussian window function, which can provide better time-frequency resolution while maintaining phase information.

[0006] In recent years, deep learning, especially convolutional neural networks (CNNs), has made significant progress in the fields of image classification and signal processing. CNNs can automatically learn the deep features of data, reducing the reliance on manual feature extraction.

[0007] However, in epileptic seizure detection, how to effectively combine the power spectrum and phase spectrum information of EEG signals to build a high-performance deep learning model remains a challenge.

[0008] In the prior art, for example: CN113397561B_An EEG detection system based on S transform and Bayesian convolutional neural network and its working method and device; Paper, Shandong University, Research on automatic detection of epilepsy based on S transform and bidirectional LSTM network. Most methods only use the power spectrum information of EEG signals, ignoring the importance of phase spectrum in epileptic seizure detection. Summary of the invention

[0009] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides an epileptic seizure prediction method based on the power spectrum and phase spectrum of the electroencephalogram (EEG). This method can make full use of the power and phase information of the EEG signal, and combine it with deep learning technology to improve the accuracy and robustness of epileptic seizure detection.

[0010] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0011] In a first aspect, an epileptic seizure prediction system based on the power spectrum and phase spectrum of an electroencephalogram is disclosed, comprising:

[0012] The EEG signal acquisition module is configured to: acquire multi-channel EEG signals and perform preprocessing;

[0013] The power spectrum and phase spectrum extraction module is configured to: divide the preprocessed electroencephalogram signal into a plurality of segments of set time, transform each segment of the set time, and extract the power spectrum and phase spectrum based on the transformation;

[0014] The model building module is configured to: build a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes the power spectrum, and the second network stream processes the phase spectrum;

[0015] The feature fusion module is configured to: splice the output features of the first network flow and the second network flow in the channel dimension, fuse the power and phase information, and obtain a fused feature map;

[0016] The prediction module is configured as follows: the fused feature map is further processed to be converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and outputted for the one-dimensional feature vector.

[0017] As a further technical solution, in the EEG signal acquisition module, multi-channel EEG signals are acquired and preprocessed, specifically including:

[0018] Collect EEG signal data at a set sampling rate and extract signal segments at each set time;

[0019] Wavelet is used to perform discrete wavelet transform, detail coefficients are selected, and a filter signal is reconstructed. Based on the filter signal, a signal segment to remove electrooculogram and electromyography artifacts is filtered.

[0020] As a further technical solution, in the model construction module, the first network flow and the second network flow have the same structure, which are used to extract deep features, and both include: multi-layer convolutional layers, batch normalization layers, activation function layers and pooling layers;

[0021] Among them, the first convolution layer receives the power spectrum or phase spectrum. After a series of convolution and pooling operations, the size of the output feature map is gradually reduced and the number of channels is gradually increased.

[0022] As a further technical solution, in the prediction module, the fused feature map is further processed and converted into a one-dimensional feature vector. The specific process is as follows:

[0023] The fused feature map is further subjected to convolution and pooling operations, and finally converted into a one-dimensional feature vector through a global average pooling layer.

[0024] As a further technical solution, in the prediction module, the probability of epileptic seizure and non-seizure is calculated and outputted for the one-dimensional feature vector, specifically:

[0025] The global average pooling layer is connected to a fully connected layer, using an activation function, to output the probability of epileptic seizure and non-seizure.

[0026] As a further technical solution, the constructed two-stream convolutional neural network model adopts a leave-one-out strategy during the model training process to ensure that the data of the test subjects are not involved in the training.

[0027] As a further technical solution, the invention further includes: a post-processing module configured to: perform post-processing on the output probability, specifically: add the output scores of a plurality of adjacent set time segments to obtain a comprehensive score;

[0028] The ratings are filtered using a sliding average filter to smooth the rating sequence;

[0029] A threshold is set to discriminate the smoothed score, and if the score is higher than the threshold, it is considered an epileptic seizure.

[0030] In a second aspect, a method for predicting epileptic seizures based on the power spectrum and phase spectrum of an electroencephalogram is disclosed, comprising:

[0031] Acquire multi-channel EEG signals and perform preprocessing;

[0032] Dividing the preprocessed electroencephalogram signal into a plurality of segments of set time, transforming each segment of the set time, and extracting a power spectrum and a phase spectrum based on the transformation;

[0033] Constructing a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes a power spectrum, and the second network stream processes a phase spectrum;

[0034] The output features of the first network flow and the second network flow are spliced ​​in the channel dimension, and the power and phase information are fused to obtain a fused feature map;

[0035] The fused feature map is further processed and converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and output for the one-dimensional feature vector.

[0036] One or more of the above technical solutions have the following beneficial effects:

[0037] The present invention comprehensively extracts the time-frequency feature information of the power spectrum and phase spectrum by performing S-transformation on the EEG signal. The two-stream convolutional neural network is used to process the power and phase features respectively and fuse them at the feature level, overcoming the deficiency of the traditional method that only relies on the power spectrum and ignores the phase spectrum information. This not only enriches the feature expression and improves the model's ability to capture the complex patterns of epileptic seizures, but also significantly improves the accuracy and sensitivity of epileptic seizure detection.

[0038] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0040] Figure 1 This is an overall flow chart of the epileptic seizure detection method based on S transform and dual-stream convolutional neural network according to an embodiment of the present invention;

[0041] Figure 2 Schematic diagram of the structure of the two-stream convolutional neural network described in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0043] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0044] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0045] Embodiment 1

[0046] See attached Figure 1 As shown, this embodiment discloses an epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram, including:

[0047] The EEG signal acquisition module is configured to: acquire multi-channel EEG signals and perform preprocessing;

[0048] The power spectrum and phase spectrum extraction module is configured to: divide the preprocessed electroencephalogram signal into a plurality of segments of set time, transform each segment of the set time, and extract the power spectrum and phase spectrum based on the transformation;

[0049] The model building module is configured to: build a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes the power spectrum, and the second network stream processes the phase spectrum;

[0050] The feature fusion module is configured to: splice the output features of the first network flow and the second network flow in the channel dimension, fuse the power and phase information, and obtain a fused feature map;

[0051] The prediction module is configured as follows: the fused feature map is further processed to be converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and outputted for the one-dimensional feature vector.

[0052] Firstly, in the electroencephalogram signal acquisition module, multi-channel EEG signals are acquired and preprocessed.

[0053] In this module, the specific implementation is: collect 18 channels of EEG data at a sampling rate of 256Hz, extract the signal segments of non-overlapping signals every 4 seconds, and each segment contains 1024 sampling points. Then, for each segment, DWT decomposition and reconstruction are performed on each EEG signal segment to achieve filtering. Specifically: Daubechies-4 (Db4) wavelet is used for 5-layer discrete wavelet transform (DWT), and the formula is:

[0054]

[0055] Among them, ψ j,k (t) is scaled to 2 j , the Db4 wavelet function with a translation of k. Through DWT, the signal is decomposed into approximation coefficients and five detail coefficients.

[0056] In order to remove high-frequency noise and artifacts such as electrooculogram (EOG) and electromyography (EMG), the detail coefficients d3 (16–32 Hz), d4 (8–16 Hz), and d5 (4–8 Hz) are selected to reconstruct the filtered signal with a frequency range of 4–32 Hz. Reconstructed signal x recon The formula for (t) is:

[0057]

[0058] Among them, c j,kis the detail coefficient corresponding to the scale and position.

[0059] Secondly, the filtered signal is divided into four 1-second segments, each containing 256 sampling points. For each 1-second segment, S-transform is performed to extract the power spectrum and phase spectrum.

[0060] The definition of S transform is as follows:

[0061]

[0062] Wherein, p is a parameter controlling the width of the Gaussian window, p=0.5 to strike a balance between time and frequency resolution, x(t) is the time signal, τ is the time variable, and f is the frequency variable.

[0063] Calculate the power spectrum |S(τ,f)| 2 And the phase spectrum φ(τ,f):

[0064] |S(τ,f)| 2 =|S x (τ,f)| 2 =[Re(S x (τ,f))] 2 +[Im(S x (τ,f))] 2

[0065]

[0066] Through the above calculations, the power spectrum and phase spectrum of each 1-second segment are obtained. The size of the power spectrum and phase spectrum is a two-dimensional time-frequency feature matrix with a dimension of 32×32×18. Among them, 32 represents the frequency dimension, 32 represents the time dimension, and 18 represents the number of EEG channels.

[0067] Then, a two-stream convolutional neural network model is constructed to process the power spectrum and phase spectrum respectively. The two network streams have the same structure, and each network stream has multiple convolutional layers, batch normalization layers, activation function layers, and pooling layers to extract deep features.

[0068] See attached Figure 2 As shown, the specific structure of the two-stream CNN model is:

[0069] Input layer: The input size is 32×32×18.

[0070] First convolutional layer: The convolution kernel size is 3×3, the number of convolution kernels is 32, the stride is 1, and the padding is "same". The output size is 32×32×32.

[0071] Batch Normalization Layer: Normalizes the convolution output.

[0072] ReLU activation function layer: applies non-linear activation.

[0073] Max pooling layer: The pooling window is 2×2 and the stride is 2. The output size is 16×16×32.

[0074] Second convolutional layer: The convolution kernel size is 3×3, the number of convolution kernels is 64, the stride is 1, and the padding is “same”. The output size is 16×16×64.

[0075] Batch Normalization layer.

[0076] ReLU activation function layer.

[0077] Max pooling layer: The pooling window is 2×2 and the stride is 2. The output size is 8×8×64.

[0078] Dropout layer: Dropout probability is 0.5.

[0079] The third convolutional layer: The convolution kernel size is 3×3, the number of convolution kernels is 128, the stride is 1, and the padding is "same". The output size is 8×8×128.

[0080] Batch Normalization layer.

[0081] ReLU activation function layer.

[0082] At the same time, a shortcut connection is added, including a convolution layer (the convolution kernel size is 3×3 and the number is 128) and a maximum pooling layer (the pooling window is 4×4 and the step size is 4), which directly reduces the input from 32×32×18 to 8×8×128 and adds it to the main branch to improve the feature expression capability.

[0083] Feature fusion and classification: The output features of the two network streams, power spectrum and phase spectrum, are concatenated in the channel dimension to obtain a fused feature map of size 8×8×256.

[0084] Fourth convolutional layer: The convolution kernel size is 3×3, the number of convolution kernels is 256, the stride is 1, and the padding is "same". The output size is 8×8×256.

[0085] Batch Normalization layer.

[0086] ReLU activation function layer.

[0087] Max pooling layer: The pooling window is 2×2 and the stride is 2. The output size is 4×4×256.

[0088] Fifth convolutional layer: The convolution kernel size is 3×3, the number of convolution kernels is 512, the stride is 1, and the padding is "same". The output size is 4×4×512.

[0089] Batch Normalization layer.

[0090] ReLU activation function layer.

[0091] Max pooling layer: The pooling window is 4×4 and the stride is 4. The output size is 1×1×512.

[0092] Global Average Pooling (GAP) layer: converts the feature map into a feature vector of length 512.

[0093] Fully connected layer: The number of neurons is 2, corresponding to the epileptic seizure and non-seizure categories.

[0094] SoftMax activation function: outputs the probability of each category.

[0095] Specifically, see again Figure 2 As shown in the figure, the first convolutional layer receives an input feature map of size 32×32×18, which is a power spectrum or phase spectrum. After a series of convolution and pooling operations, the size of the output feature map is gradually reduced and the number of channels is gradually increased. In order to prevent overfitting, a Dropout layer is added to the network.

[0096] Next, the output features of the two network streams are concatenated in the channel dimension to fuse the power and phase information. The fused feature map undergoes further convolution and pooling operations, and finally passes through a global average pooling (GAP) layer to convert the feature map into a one-dimensional feature vector. Then, a fully connected layer is connected, and the SoftMax activation function is used to output the probability scores of epileptic seizures and non-seizures.

[0097] During the model training process, the Leave-One-Subject-Out (LOSO) strategy was adopted to test each subject individually, and their data did not participate in the training, to ensure that the test subject's data did not participate in the training and improve the generalization ability of the model. The optimizer used was Adam, with a learning rate of 0.001, a batch size of 512, a maximum number of training rounds of 320, and an L2 regularization coefficient of 0.0001.

[0098] During the model training process, the order of the input EEG channels is randomly shuffled in each small batch of data to prevent the model from overfitting to a specific channel order.

[0099] Finally, the probability scores output by the model were post-processed. The output scores of four adjacent 1-second segments were added together to obtain a comprehensive score. In order to smooth the score sequence, a sliding average filter with a length of 7 was used to filter the scores. A threshold was set to discriminate the smoothed scores. If the score was higher than the threshold, it was determined to be an epileptic seizure. The threshold Thr was adjusted according to the patient's EEG signal characteristics to achieve the best balance between sensitivity and specificity. Taking into account the delay caused by filtering, the segments determined to be epileptic seizures were expanded by a certain number of points on the time axis.

[0100] In an implementation example, specifically, score accumulation: the output probability scores of four adjacent 1-second segments are added together, and the comprehensive score ranges from -4 to +4.

[0101] Moving average filtering: A moving average filter with a length of 7 (N=3) is used, and the formula is:

[0102]

[0103] Among them, x(i) is the i-th comprehensive score.

[0104] Threshold judgment: Set the threshold Thr. When the smoothed score Z(i)>Thr, it is judged as an epileptic seizure.

[0105] Boundary compensation: For segments judged as epileptic seizures, 12 points are expanded on both sides of the time axis to compensate for the delay caused by filtering, thereby improving the sensitivity index of epilepsy detection.

[0106] Through the above method, the present invention achieves high-accuracy detection of epileptic seizures, fully utilizes the power and phase characteristics of EEG signals, and has important clinical application value.

[0107] The sub-technical solution of this embodiment can improve the accuracy and robustness of epileptic seizure detection. The method makes full use of the power spectrum and phase spectrum information of the electroencephalogram (EEG) signal, combined with deep learning technology, to overcome the defects of insufficient use of phase information and low detection accuracy in traditional methods.

[0108] Application of the model on the CHB-MIT dataset:

[0109] The model was trained and tested on the CHB-MIT database, where the epilepsy database contains EEG data of 23 patients with a total length of approximately 979.93 hours, including EEG records of epileptic seizures and non-seizure states. The model performance evaluation indicators include segment-level accuracy, segment-level sensitivity, segment-level specificity, event-level sensitivity, and event-level false detection rate (FDR). These evaluation indicators are defined as follows:

[0110] Segment level accuracy (Accuracy):

[0111]

[0112] Segment Level Sensitivity:

[0113]

[0114] Segment-level specificity:

[0115]

[0116] Event-level sensitivity: the percentage of correctly detected seizure events to the total number of seizure events.

[0117] Event-level false detection rate (FDR): The average number of false detections per hour.

[0118] The experimental results on the CHB-MIT database are shown in Table 1: average accuracy: 92.00%, average sensitivity: 79.59%, average specificity: 92.23%, average event level sensitivity: 85%, and average false positive rate: 2.52 times / hour.

[0119] Table 1. Epileptic EEG detection results of the model on the CHB-MIT database

[0120]

[0121]

[0122] Embodiment 2

[0123] See again the attached Figure 1 As shown, the purpose of this embodiment is to provide an epileptic seizure prediction method based on the power spectrum and phase spectrum of the electroencephalogram, including:

[0124] Acquire multi-channel EEG signals and perform preprocessing;

[0125] Dividing the preprocessed electroencephalogram signal into a plurality of segments of set time, transforming each segment of the set time, and extracting a power spectrum and a phase spectrum based on the transformation;

[0126] Constructing a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes a power spectrum, and the second network stream processes a phase spectrum;

[0127] The output features of the first network flow and the second network flow are spliced ​​in the channel dimension, and the power and phase information are fused to obtain a fused feature map;

[0128] The fused feature map is further processed and converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and output for the one-dimensional feature vector.

[0129] Embodiment 3

[0130] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0131] Embodiment 4

[0132] The purpose of this embodiment is to provide a computer-readable storage medium.

[0133] A computer-readable storage medium stores a computer program, which executes the steps of the above method when executed by a processor.

[0134] Embodiment 5

[0135] The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments.

[0136] The steps involved in the apparatus of the above embodiment correspond to the method embodiment 2, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method of the present invention.

[0137] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0138] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. An epileptic seizure prediction system based on the power spectrum and phase spectrum of electroencephalogram, characterized in that: include: The EEG signal acquisition module is configured to: acquire multi-channel EEG signals and perform preprocessing; The power spectrum and phase spectrum extraction module is configured to: divide the preprocessed electroencephalogram signal into a plurality of segments of set time, transform each segment of the set time, and extract the power spectrum and phase spectrum based on the transformation; The model building module is configured to: build a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes the power spectrum, and the second network stream processes the phase spectrum; The feature fusion module is configured to: splice the output features of the first network flow and the second network flow in the channel dimension, fuse the power and phase information, and obtain a fused feature map; The prediction module is configured as follows: the fused feature map is further processed to be converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and outputted for the one-dimensional feature vector.

2. The epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram as claimed in claim 1, characterized in that: In the EEG signal acquisition module, multi-channel EEG signals are acquired and preprocessed, specifically including: Collect EEG signal data at a set sampling rate and extract signal segments at each set time; Wavelet is used to perform discrete wavelet transform, detail coefficients are selected, and a filter signal is reconstructed. Based on the filter signal, a signal segment to remove electrooculogram and electromyography artifacts is filtered.

3. The epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram as claimed in claim 1, characterized in that: In the model building module, the first network flow and the second network flow have the same structure and are used to extract deep features, and both include: multi-layer convolutional layers, batch normalization layers, activation function layers and pooling layers; Among them, the first convolution layer receives the power spectrum or phase spectrum. After a series of convolution and pooling operations, the size of the output feature map is gradually reduced and the number of channels is gradually increased.

4. The epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram as claimed in claim 1, characterized in that: In the prediction module, the fused feature map is further processed and converted into a one-dimensional feature vector. The specific process is as follows: The fused feature map is further subjected to convolution and pooling operations, and finally converted into a one-dimensional feature vector through a global average pooling layer.

5. The epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram as claimed in claim 1, characterized in that: In the prediction module, the probability of epileptic seizure and non-seizure is calculated and outputted for the one-dimensional feature vector, specifically: The global average pooling layer is connected to a fully connected layer, using an activation function, to output the probability of epileptic seizure and non-seizure.

6. The epileptic seizure prediction system based on the power spectrum and phase spectrum of the electroencephalogram as claimed in claim 1, characterized in that: The constructed two-stream convolutional neural network model adopts the leave-one-out strategy during the model training process to ensure that the test subject’s data is not involved in the training; Preferably, it further comprises: a post-processing module configured to: perform post-processing on the output probability, specifically: add the output scores of a plurality of adjacent set time segments to obtain a comprehensive score; The ratings are filtered using a sliding average filter to smooth the rating sequence; A threshold is set to discriminate the smoothed score, and if the score is higher than the threshold, it is considered an epileptic seizure.

7. An epileptic seizure prediction method based on the power spectrum and phase spectrum of the electroencephalogram, characterized in that: include: Acquire multi-channel EEG signals and perform preprocessing; Dividing the preprocessed electroencephalogram signal into a plurality of segments of set time, transforming each segment of the set time, and extracting a power spectrum and a phase spectrum based on the transformation; Constructing a two-stream convolutional neural network model, the model includes a first network stream and a second network stream, the first network stream processes a power spectrum, and the second network stream processes a phase spectrum; The output features of the first network flow and the second network flow are spliced ​​in the channel dimension, and the power and phase information are fused to obtain a fused feature map; The fused feature map is further processed and converted into a one-dimensional feature vector, and the probability of epileptic seizure and non-seizure is calculated and output for the one-dimensional feature vector.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method described in claim 7 is implemented.

9. A computer 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 program, the steps of the method described in claim 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to claim 7 are performed.

Images