Epileptic detection using artificial neural networks

Abstract

For the diagnosis of epilepsy, electroencephalogram (EEG) signal plays an important role. EEG recordings of an epileptic patient obtained from 'ambulatory recording systems' contain a large volume of EEG data. A time consuming analysis of the entire length of EEG data by an expert is required to detect the epileptic activity. This paper discusses an automated diagnostic method using artificial neural networks for the detection of epilepsy. Experimental results obtained with four different types of neural networks, namely, multi-layer perceptron, Elman network, probabilistic neural network and learning vector quantization is presented. It is found that the Elman network performs better than the other three neural networks. It is also shown that the performance of the Elman network with a single attribute as the input is almost identical to that of the recently reported LAMSTAR network, which uses two attributes as inputs.