Implementation of Convolutional Neural Network for Epileptic Seizure Detection

Abstract

Epilepsy is a neurological disorder and disability in which the brain activity becomes abnormal causing seizures. A seizure is characterized by the unprovoked sudden alteration of the electrical activity of the brain. It is defined by a prolonged inclination to cause epileptic seizures and by the pathophysiological, psychological, cognitive, and social ramifications of this state, thus early detection of epileptic seizures is crucial. In this work, the convolutional neural network (CNN) is used to extract the important spatial information from Electroencephalogram (EEG) signals and a classification task using pre-trained networks by transfer learning is performed on the extracted features to detect the onset of a seizure. An accuracy of 95.09% is achieved using MATLAB. Pre-trained neural networks involve an enormous number of computations. Hence, a novel 30-30-10-10 neural network is devised as a feedforward fully connected neural network to reduce computational complexity. Simulation is performed in Verilog using Xilinx Vivado, achieving an accuracy of 96.6667%.