A New Approach for Classification of Epilepsy EEG Signals Based on Temporal Convolutional Neural Networks

Abstract

Epilepsy, a chronic disease, whose symptoms are sudden onset of brain neurons by the abnormal discharge, leads to transient cerebral dysfunction. In recent years, the observation of EEG signals plays an important role in the diagnosis of epilepsy, as EEG signals contain much useful detailed information which helps doctors to diagnose epilepsy. However, artificial analyses rely heavily on doctors' experience and may produce erroneous and missed diagnosis. Thus many researchers focus on methods that can automate the process of EEG detection. According to recent studies, the mainstream approaches mainly consist of two steps: signal pre-processing and classification. In this paper, our method uses a deep convolutional network called Temporal Convolutional Neural Networks (TCN) to classify the EEG signals. The proposed approach can automatically learn feature representation from the raw EEG signals without any pre-processing. To evaluate the effectiveness of our approach, we design and train a classifier base on TCN to classify EEG data. The best performance of the 14 different combinations of two-class epilepsy detection gives an accuracy of 100.00%. Our results show that TCN is able to accurately distinguish epileptic data from non epileptic data. In terms of computation cost and accuracy, this novel approach proposed in this paper is more competitive than others, such as wavelet transform, Naive Bayesian and KNN etc.