Separating EEG spike-clusters in epilepsy by a growing and splitting net

Abstract

The presurgical evaluation of epilepsy patients relies on an exact localization and delineation of the generators of epileptic seizures. During the registration of the electroencephalogram (EEG) sharp transient signals called spikes can be observed. These spikes give hints for the so called epileptogenic zone in the brain. In order to decide whether these spikes derive from single or multiple generators an incrementing topology preserving map with insertion and deletion of units was trained for the EEG data of individual patients. By deleting of units the net was separated into subnets. Thus it could be further used for vector quantization. The spatial distributions of the peak amplitude of the spikes in all channels as well as the time differences of their peaks were used as input signals. The separation of spatio-temporal clusters of the spikes was compared with those clusters identified by a human reviewer.