Constrained ICA for seizure onset analysis in the EEG

The extraction of epileptic seizure waveform from the electroencephalogram (EEG) using Independent Component Analysis (ICA) was demonstrated by James and Lowe. A recent variation, Constrained ICA, allows a supplied reference signal to select a single component to be extracted. We show how this algorithm can be applied to the problem of seizure waveform extraction from EEG signals prior to the visible onset of a seizure. The algorithm also allows information such as topographic maps of the seizure waveforms to be obtained. Automatic early detection of such waveforms could lead to a real time system for seizure prediction