Epileptic Seizure Detection and Prediction using Deep Learning
Abstract
Neurologists perform visual inspection to detect the presence of epileptic seizure which is a laborious process. Moreover, approximately 50 million people are affected by epileptic seizure. Thus a system is required that automatically detect and predict seizure. Neural patterns of Electroencephalogram (EEG) signals of patients in seizure state and preictal state can be used to detect and predict seizures respectively. Seizure prediction is important because if seizures get predicted at early stages, then they can be suppressed using electric stimulations. The input EEG signals are fed as input to the 15 layers Convolutional neural network. One of the biggest challenges while detecting and predicting seizures using EEG signal as input is robust feature extraction. In our approach, we use the power of Convolutional neural networks to extract the robust features of electroencephalogram signals. We evaluate our model on Bonn university database by performing multiclass classification i.e. Preictal vs. Ictal vs. healthy and have achieved an overall accuracy of 96.67%.