Extracting the spike process from the EEG by spatially constrained ICA

Epileptic patients often show interictal epileptic discharges (IED´s) in the electroencephalogram (EEG) recorded between seizures. This epileptiform activity is in many cases related to the location of the seizure onset, and is believed to reflect the frequency of the seizures. We present a fully automated technique that is able to extract the IED´s from the EEG, despite the obscuring artifacts. The presented technique is based on a multi-objective optimization by maximizing the signal´s kurtosis and minimizing its distance to a defined template. Preliminary results show that this technique automatically extracts a source on which spike detection techniques should perform better than on the regular channel selection procedure