ICA with a reference: extracting desired electromagnetic brain signals

James, C.J. and Gibson, O. (2002) ICA with a reference: extracting desired electromagnetic brain signals. IEE Seminar: Medical Applications of Signal Processing, London, UK. 07 Oct 2002. p. 4 . (doi:10.1049/ic:20020282).

Abstract

ICA is a technique for the extraction of statistically independent components from a set of measured signals. This technique enjoys numerous applications in biomedical signal analysis in the literature, especially in the analysis of electromagnetic (EM) brain signals (measured via the EEG and MEG). Standard implementations of ICA are limiting mainly due to the assumptions of equal numbers of sources and measured signals inherent with methods that assume square mixing. For EM brain signal recordings which have large numbers of channels (e.g. MEG), the large number of resulting extracted sources makes the subsequent analysis laborious and highly subjective. However, there are many instances in neurophysiological analysis where there is strong a priori information about the signals being sought, such signals generally include artifacts, as well as specific rhythms or even seizure or spike waveforms in the ictal and interictal EEG. Constrained ICA, as introduced by Lu and Rajapakse (2001, Proceedings of Third International Conference on Independent Component Analysis and Blind Signal Separation), consists of a variation of standard ICA that can extract signals that are statistically independent, yet which are constrained to be similar to some reference signal-the measure of similarity could be, for example, mean-squared error, correlation, or any other suitable a priori information that can be included. The algorithm is fast and suitable for online analysis and here we demonstrate this method on a number of artifactual and clinically significant waveforms identified in recordings of EEG and MEG, where constrained ICA was applied to each in turn using a crude reference waveform, derived from the raw recordings in each case. The algorithm repeatedly converged to the desired component within a few iterations and subjective analysis indicated waveforms of the expected morphologies and with realistic spatial distributions. Here we show that constrained ICA can be applied with great success to EM brain signal analysis, automating artifact extraction in MEG and EEG, as well as on seizure extraction in the EEG.

This record has no associated files available for download.

Contributors

Download statistics