A Multistage System for Automatic Detection of Epileptic Spikes

A multistage automatic detection system for epileptic spikes is introduced as an assistant tool for epileptic analysis and diagnosis based on electroencephalogram (EEG). The system consists of four stages: preprocessing, feature extraction, classifier and expert system. Multiple state-of-the-art signal processing and machine learning techniques including wavelet transform, spectral filtering, artificial neural network are utilized in order to improve the ability of the overall system stage by stage. Compared to other works, our contributions are three-fold: peaks in the EEG recording are categorized into two groups of non-epileptic spikes and possible epileptic spikes by a committee of three perceptrons; appropriate mother wavelet and wavelet scales are selected for the best system performance; and, based on the neurological fact that an epileptic spike is usually followed by a slow wave, a simple expert system is presented to eliminate pseudo-spikes which are closely analogous to true epileptic spikes. Experimental results show that the proposed system is capable of detecting epileptic spikes efficiently.