The self-organising feature map in the detection of epileptiform transients in the EEG

A system is proposed which utilises Kohonen´s self-organising feature map (SOFM) to help detect the presence of epileptiform transients in the EEG. The system consists of a feature extractor, which employs a mimetic approach to detect candidate epileptiform transients (CETs) on individual channels in the multichannel recording, followed by a trained SOFM. The SOFM system is proposed as a single channel module of a larger system to detect multichannel epileptiform events by incorporating the outputs of sixteen such modules. The SOFM was trained with CETs obtained from 16-channel bipolar EEG segments of nine patients and fine-tuned through Kohonen´s learning vector quantisation technique (LVQ2). Measures of the sensitivity and specificity of the trained map were obtained by presenting a subset of CETs to the SOFM which had been graded by two to three electroencephalographers as being true or false. The results obtained show that the trained SOFM has a sensitivity of 63% and a specificity of 79% for a 16×16 SOFM