The classification of spikes in EEG recordings using features derived from ICA

Abstract

Electroencephalogram is a complex signal which, until now, has required expert human visual analysis to be used as a medical diagnostic test. An efficient method for automated analysis would provide substantial time saving in practice. A new approach for the automatic detection of epileptic activity in the form of spike and sharp waves in EEG recordings is presented. It comprises feature extraction derived from independent component analysis and a quadratic classifier. Optimisation with respect to correction for eyeblinks and occipital alpha rhythm was cross-evaluated on sharp and spike waves from 7 EEG recordings of total duration 123 minutes previously labelled by an expert. With a training set discrimination threshold of 70%, the sensitivity was 64.1% with 36.1 false positives per minute. After correction for eyeblinks and occipital alpha, sensitivity and false positives reduced to 64% and 59% and 29.7 and 16.6 per min, respectively. Optimisation is still in development; planned improvements include consideration of spatial patterning of epileptic activity and correction for other background activity and artefacts. (4 pages)