RUMBA Workshop on Concepts and Methods in Functional Imaging
NIPS2001 Workshops
Whistler, BC
December. 7-8, 2001
Scott Makeig, Institute for Neural Computation, UCSD & The Salk Institute, La Jolla CA
Top-down Brain Dynamics in Functional Imaging Data
An essential part of our experience of self is our "top-down" experience of "myself" as interacting with events and/or with past impressions, experiences we might recount, for example, by saying, "All of a sudden, ...," "I searched my mind for ...," "It occurred to me that ...," "I wondered if...," "I decided that ...," "I imagined that ...," etc. We tend to associate such experiences with moments in time. Together, they form our personal "stream of consciousness" -- equally as, or even more than, events in the external world. They are the stuff "interesting" first-person accounts are largely made of, and form a large part of our waking (and dreaming) experience of "self." Beginning in the early 1960s, cognitive neuroscience researchers believed that neural processes accompanying at least some such events could be captured by averaged event-related scalp potentials (ERPs) -- for example, the P300 ("Aha!"), CNV ("Rrrrready-Go!"), N400 ("???") and ERN ("Oops!"). In the last few years, similar hopes have arisen in relation to averaged event-related BOLD responses. Unfortunately, the average of a set of time series time locked to some class of events may not adequately capture the the event-related dynamics expressed in the data.
Historically, scientific neuroscience began with observing activation of hierarchically structured peripheral sensorimotor pathways. A similar model of (passive) perception in cortex -- in which external stimuli successively activated pools of cortical neurons tuned, first, to specific and, next, to broader features of a presented stimulus -- was used at first to model ERP dynamics. The most obvious possible ERP correlates of brief periods of information processing within functionally distinct cortical areas were the successive positive and negative peaks contained in single-channel averaged ERP waveforms. It is now clear, however, that this model is insufficient for characterizing event-related EEG dynamics and for modeling their relationship to neural processes supporting cognitive behavior and experience (Makeig, in press).
Currently, more sufficient models of EEG dynamics are beginning to emerge, at the same time as new neurophysiologic evidence linking distributed oscillatory burst events in extra-cellular fields to top-down behavior and experience. I will show evidence from event-related EEG experiments of the insufficiency of ERP averaging, and will give an example of more adequate modeling of event-related EEG dynamic events during self-error recognition. I will then discuss the prevalence of top-down events in BOLD data, and prospects for concurrent EEG/fMRI analysis of top-down brain dynamics.