Isolation of epileptiform discharges from unaveraged EEG by independent component analysis

Abstract 

Objective: We propose a method that allows the separation of epileptiform discharges (EDs) from the EEG background, including the ED's waveform and spatial distribution. The method even allows to separate a spike in two components occurring at approximately the same time but having different waveforms and spatial distributions.

Methods: The separation employs independent component analysis (ICA) and is not based on any assumption regarding generator model. A simulation study was performed by generating ten EEG data matrices by computer: each matrix included real background activity from a normal subject to which was added an array of simulated unaveraged EDs. Each discharge was a summation of two transients having slightly different potential field distributions and small jitters in time and amplitude. Real EEG data were also obtained from three epileptic patients.

Results: Through ICA, we could isolate the two epileptiform transients in every simulation matrix, and the retrieved transients were almost identical as the originals, especially in their spatial distributions. Two epileptic components were isolated by ICA in all patients. Each estimated epileptic component had a consistent time course.

Conclusion: ICA appears promising for the separation of unaveraged spikes from the EEG background and their decomposition in independent spatio-temporal components.

Keywords:  EEG, Epileptiform discharge, Independent component analysis, Principal component analysis