To be presented at Human Brain Mapping Conference, Toronto, 2005

Spatial dependence of EEG independent component source distributions on task demands

Julie Onton & Scott Makeig

Swartz Center for Computational Neuroscience

Institute for Neural Computation

University of California San Diego

Independent Component Analysis (ICA) is a powerful method for blind separation of EEG or MEG source signals recorded from scalp electrodes. In applying ICA, we conceive of EEG ‘sources’ as spatially-fixed patches of cortical neuropile within which local field activity is continuously or intermittently partially synchronized. As in all linear data decompositions, each ICA component consists of an activity time course plus a map giving its projection strengths to each of the scalp sensors. In our hands, stable ICA decompositions of high density EEG recordings yield on the order of twenty-five maximally independent components whose scalp maps strongly resemble the projection of one or occasionally two symmetric equivalent current dipoles. Here, for the first time we assess the degree to which the three-dimensional spatial distributions of independent source dipoles (and thus, presumably, the corresponding cortical source patches) are similar across subjects participating in two dissimilar tasks.

We compared the spatial distributions of independent component equivalent-dipole source models across two separate subject groups participating in a modified Sternberg visual working memory task and an emotion imagination task. In the working memory task, twenty-three subjects attended and memorized a series of black letters presented at fixation, then indicated with a button press, after a few-second delay, whether or not a subsequently presented probe letter was or was not in the memorized letter set. Thirty-three subjects performed the imagination task with eyes closed, listening to short verbal narratives interspersed with silent periods of several minutes in which subjects imagined circumstances and events associated with suggested emotional states, while concentrating on evoking and sensing physiological changes associated with those emotions, only occasionally pressing a finger button to indicate changes in their perceived task engagement. Working memory task data were recorded from 71 electrodes, and emotion imagination data from 253 electrodes distributed over the head surface.

Statistical comparison of the resulting three-dimensional group source dipole distributions demonstrated that the locations of dipolar ICA components tended to be concentrated in several characteristic areas, but that the locations of some group EEG information ‘hot spots’ identified by ICA also depended in part on the task being performed. In both tasks, component dipole locations were broadly distributed, with highest densities over bilateral hand somatomotor cortex (mu rhythm activity) and inferior lateral and central medial occipital cortex (alpha band and visual evoked response activity). The working memory task was associated with higher component density in the dorsal anterior cingulate cortex (theta and low-beta band activities), while in data recorded during the emotional imagination task ICA identified more independent EEG sources in bilateral orbitofrontal and inferior parietal cortex. These results show, first, that macroscopic synchronization of field activity, producing EEG scalp signals, is unevenly distributed in cortex and, second, that the locations of the synchronizing cortical areas depend in part on mental activity. These results illustrate the feasibility of developing new modes of functional brain imaging based on high-density EEG recording.