EEG-Based Detection and Prediction of Epileptic Seizures Using One-Dimensional Convolutional Neural Networks

Abstract

Seizure detection and prediction using electroencephalogram (EEG) signals is still challenging. The accurate detection and prediction of seizures will improve the quality of life and reduce the suffering for people with epilepsy. In this work, deep learning (DL) related techniques are applied. Through the analysis of DL related techniques combined with EEG signals, this dissertation aims to explore the efficient seizure detection and prediction methods or algorithms. Considering the one-dimensional characteristics of EEG signals (time series), our work mainly focuses on the application of one-dimensional convolutional neural networks (1DCNN) for seizure detection and prediction. Moreover, since the combination of channel selection and 1D-CNN is less studied in seizure prediction, the methods of channel selection strategy combined with 1D-CNN are also proposed for the analysis of seizure prediction. In the first article, we analyzed a short-term EEG dataset for seizure detection. This work simultaneously used 1D-CNN and two-dimensional convolutional neural networks (2D-CNN) to test the short-term Bonn EEG dataset and achieved remarkable results. In the second article, we further studied the seizure detection by using two long-term EEG datasets, the CHB-MIT sEEG and the SWEC-ETHZ iEEG datasets. In this work, a stacked 1D-CNN model was applied to test these two different datasets. In the third article, our goal was to study the seizure prediction using EEG signals. Therefore, based on the long-term Freiburg iEEG dataset, we proposed a novel method of 1D-CNN combined with channel selection strategy for seizure prediction. Since the third article only considered 9 channel cases (total 63 channel cases) for the best channel case selection, the fourth article further discussed channel selection strategy based on all channel cases for seizure prediction. In the fifth article, a novel method of channel increment strategy-based 1D-CNN was proposed for seizure prediction based on the same iEEG dataset. In conclusion, our work successfully applied CNNs in the short- and longterm sEEG and iEEG signals for the analysis of seizure detection and prediction, and the methods of channel selection strategy combined with 1D-CNN also showed remarkable performances in seizure prediction. Keywords: Epilepsy, electroencephalogram (EEG), seizure detection, seizure prediction, one-dimensional convolutional neural networks (1D-CNN)