Society for Neuroscience Abstract 2007

Single-trial EEG changes associated with specific behavioral contexts in a two-back task

Julie Onton* and Scott Makeig

Instead of ascribing trial-to-trial variability to brain or recording 'noise,' we seek to understand its behavioral relevance and functional significance. Here we present a novel approach to trial-by-trial data analysis that represents each trial at a brain location of interest by (a) its event-related log spectral power (ERSP) image and by (b) a 'context vector' coding 'yes' or 'no' answers to multiple questions about trial events or trial context. During a 'two-back' continuous performance task, subjects responded with a button press to each presented letter to indicate whether the current letter was or was not the same as that shown two letters earlier. The 28 questions used to create the context vector included, 'Was the current letter a match?', 'Was the subject response correct?', 'Did the subject press the same button following the previous letter?', "Will the subject answer correctly in the following trial?', etc. We decomposed the concatenated log ERSP and context vectors using independent component analysis (ICA). Result context factors included many context-vector templates with intuitively meaningful profiles, for example letters presented during a string of correct responses (or not), trials containing a matching letter (or not), and whether (or not) the subject would respond correctly in the following trial. Many of these context factors were associated with modulations of midline theta or parietal/occipital alpha power. Average measures of EEG activity across supposedly comparable event-related trials typically reveal only a small remnant of the rich variability exhibited by EEG activity across outwardly similar single trials. Context ICA (xICA) avoids the need to 'blindly' average measures over subsets of event-related trial epochs assumed by the experimenter to elicit difference types of EEG activity. Instead, xICA reveals the most distinct patterns of EEG dynamic changes associated with particular trial contexts, modeling the EEG data at a recording channel or in a spatially-filtered source location as a product of modulatory spectral influences operating multiplicatively on a base broadband EEG spectrum.