We present a novel method using Independent Component Analysis (ICA) that is capable of isolating epileptiform discharges (EDs) from unaveraged background EEG, separating the EDs into their spatio-temporal components. No assumptions of the underlying generator model are necessary. We performed a simulation study where the epileptiform activity (EA) was assumed to be emanating from two (different) dipole generators in close physical proximity and orientation in the brain. The degree of correlation between spatio-temporal components of the generators and those extracted with ICA was measured to assess performance. ICA was also compared to Principal Component Analysis (PCA). In every data-set ICA managed to isolate the two underlying components. In each case the time varying EDs were almost identical to the models' especially in their spatial distribution. Conversely PCA never managed to isolate the two separate underlying generators of the EDs. The method was also tested on real unaveraged data segments of epileptiform EEGs. For every patient tested, ICA managed to isolate two separate components from the EA with a consistent time course for each patient. These results convince us that ICA is a useful tool in decomposing unaveraged EEG into its underlying components for further assessment of the spatio-temporal components.
|Number of pages||6|
|Journal||IEE Colloquium (Digest)|
|Publication status||Published - Oct 6 1999|
ASJC Scopus subject areas
- Electrical and Electronic Engineering