Study On Seizure Classification Method Based On Graph Convolutional Network

Epilepsy is a brain disorder caused by hyper-synchronous discharges of neurons in the brain and affects approximately 75 million people worldwide.Because of the uncertainty of seizures,epilepsy is not only physically and psychologically damaging,but also has a social impact,and timely and effective detection of seizures is important for both the individual and society.A variety of seizure classification methods exist,but there is still room for improvement in their performance.In this paper,a graphical convolutional network model is used to further improve the performance of seizure classification.Firstly,the basic structure of the brain is analysed,the characteristics of the EEG signals in normal and seizure states are explored separately,and the basis for the classification of EEG signals in epileptic patients is described.Seizures are the result of the synergistic action of several brain regions,then there is a certain relationship between different channels of EEG,which lays the foundation for the establishment of a valid seizure classification model.Second,a multi-channel seizure classification model based on a mutual information graph convolutional neural network was developed.While the Pearson method used in the traditional graph convolutional neural network approach determines the connectivity between channels,the model uses the mutual information connectivity of EEG signals to determine the pattern of functional connectivity of brain regions in epileptic patients.Experiments were also conducted on the CHB-MIT dataset,and the results showed that the established network model could improve the classification accuracy with an accuracy of 99.11%,which is superior to other methods for the epilepsy classification problem.Finally,a deep learning seizure classification model that fuses graph convolutional neural networks and long-short term memory networks was developed.In the fusion model,multiple graph convolutional neural networks and long-short-term memory neural networks were used to extract the graph features and time-domain features of multichannel EEG signals respectively,and the seizure classification was performed by a multilayer perceptron for the period.Experiments were also conducted on the CHB-MIT dataset,and the results showed that the model achieved an accuracy of 99.75%,providing better performance compared with other methods.