Deep learning for predicting epileptic seizures using EEG signals

Abstract

Epileptic seizure prediction is a challenging problem which consists in identifying a seizureonset using electroencephalogram (EEG) signals, either by an experienced neurologist or automaticallyusing machine learning techniques. In this work, we will take advantage from recent advances of deeplearning techniques and propose two architectures. The first model is based on one dimensional convolutional neural network (1-D CNN) architecture that learns hierarchical representations of EEG with nopreprocessing steps. The second model is based on long term recurrent convolutional network (LRCN)which has the capacity to learn different representations of the spatiotemporal structure of EEG signal.Experimental results demonstrate the reliability of the models by achieving a mean Area Under Curve(AUC) Receiver Operating Characteristics (ROC) of 0.848 for 1-D CNN model, and 0.873 for LRCN model.