Epilepsy prediction based on PTE and TE of EEG signals using DSC-CNN

Abstract

Epilepsy is a disease caused by abnormal discharge of neurons, which can seriously endanger people's life. Therefore, it is necessary to predict the occurrence of epilepsy timely. The main content of epilepsy prediction is to distinguish the preictal and interictal periods of electroencephalography (EEG) signals. The transfer entropy (TE) and phase transfer entropy (PTE) of EEG signals is computed and sliced to form the features of the EEG signals. Then, these features are inputted to a depthwise separable convolutional neural network (DSC-CNN), which has a small amount of parameters and computation, to classify the two periods of EEG signals. The CHB-MIT scalp EEG dataset are used to evaluate the performance of this scheme. The computation results are also compared to other state-of-the-art algorithms to verify its advantages. Experimental results show that the prediction time with this method reaches 22.6 minutes. The prediction specificity is 99.26% for the prediction time of 15 minutes, and it is 99.52% for the prediction time of 22.6 minutes. Moreover, the DSC-CNN has a small amount of parameters and short running time.