Epileptic Seizure Detection Using Convolution Neural Networks

Abstract

Epilepsy is a central nervous system disorder that affects a substantial number of world's population and disrupts the quality of life of the sufferers. A number of diagnostic techniques evolved over the years for the detection of epileptic seizures using encephalograms. The subject paper presents design and implementation of a classification model based on convolution neural networks that is capable of detecting epileptic seizures using computational methods utilizing encephalogram data. The study used convolution neural networks that have unique characteristics for recognizing patterns and images and in classifying their features. The neural network architecture proposed herein comprises of layers for input and output with several hidden convolution layers. The electroencephalogram database that was used in this work is the freely accessible CHB-MIT scalp encephalogram database. The developed approach was implemented using the 22 subject database and testing was carried out on patients a few days after the withdrawal of the anti-seizure medications. The test subjects were composed of 5 males and 17 females from various age groups. It was observed that the suggested algorithm could detect about 94.6 percent of the 198 tested seizure records, indicating a good performance of the proposed seizure detection algorithm.