Research On Epilepsy Detection Based On Brain Functional Networks

Epilepsy is a neurological disease that causes huge burden and physical great pain to the patient.Clinically,medical staff observe electro-encephalogram(EEG)to detect epilepsy,but this process is time-consuming and laborious,so automatic detection of epilepsy is of great significance.However,studies have ignored the transmission of neural electrical activity between various brain regions and rarely focused on the construction and analysis of brain functional network.Therefore,it is very important to construct a brain network that can show the connection strength between brain regions during epileptic seizures.In addition,the existing eeg feature research is limited to the time domain and frequency domain,which are only limited to the local or global features of the EEG.Therefore,it is very important to automatically learn the key features of the EEG signal for epilepsy detection.When a seizure occurs,the electrodes placed on the scalp near the epileptic focus will have obvious and consistent voltage changes,so how to convolute the adjacent electrodes to take advantage of the graph structure between channels is the key work.Based on the public CHB-MIT data set,a data pre-processing scheme was designed for the dataset.Effective brain connectivity(EBC)is used to analyze brain networks,and it is found that the small-world properties of brain network become prominent during the course of epileptic seizures.Then,the key features of epilepsy EEG detection were extracted through the autoencoder,and the support vector machine and convolutional neural network were used as a classifier to compare the presence or absence of auto-encoder features to verify the effect of auto-encoder.Finally,a new network structure is proposed which is composed of hierarchy graph network and light graph convolutional network.In this process,the complex graph structure relationship is first integrated through a hierarchy graph network,and then graph convolution is used to combine the graph structure and features to finally achieve classification.Comparing with other experiments,the model proposed in this paper has achieved,better experimental results in terms of sensitivity and specificity.