An ICA Approach for Extracting Task-related Components from EEG signals

Independent component analysis (ICA) is suggested to be useful in the analysis of signals (electroencephalography (EEG) and event-related potential (ERP)) recorded from the scalp electrodes to examine brain activity. But ICA analysis of ERP has the problem that it lacks validity across subjects. Therefore, we examined a new approach to give validity across subjects to ICA results obtained from single subjects. To do this, we combined two criteria of how much a single subject´s signals contain task-related components and the similarity of the spatial patterns of those task-related components across subjects. We applied our approach to EEG data recorded in an experiment. As a result, analysis with a combination of ICA and our approach extracted significantly more task-related components from the observed signals than analysis without the combination did, although both analyses suggested that the occipital area activity was most relevant to performing the task across subjects. These results suggest that our approach is useful for extracting task-related components from the observed EEG signals and for giving validity across subjects in ICA analysis of EEG/ERP