Society for Psychophysiological Research (SPR)

Lisbon, Portugal

Sept. 22-25, 2005

Advances in information-based EEG signal processing


Scott Makeig - Advances in information-based EEG signal processing

Stefan Debener - Independent component analysis of EEG recorded simultaneously with fMRI identifies medial frontal cortex in performance monitoring

Peter Ullsperger - Beyond averaging: Exploring N400 by independent component analysis

Julie Onton - Frontal midline EEG dynamics during working memory

Martin McKeown - Frontal theta rhythms in normal subjects and in subjects with Parkinson's disease

Scott Makeig*

Swartz Center, Institute for Neural Computation, University of California San Diego, La Jolla CA

Advances in information-based EEG signal processing

For nearly two centuries, statistical methods development focused mainly on mean and second-order (Gaussian) statistics because of their analytic and computational simplicity. The last decade has seen ever accelerating application of more general Bayesian statistics and information theory in signal processing, including the development of independent component analysis (ICA) and related methods for performing blind source separation from multidimensional data. These methods attempt to model the information content of the whole data, rather than only its mean and variance. For EEG analysis, this involves eliminating redundant information projecting to many scalp channels by volume conduction. An area of emerging interest is frontal activities in the theta band, whose complex dynamics and relationship to experimental events are not well characterized by standard time- and frequency-domain averaging.

Stefan Debener(1)*, Markus Ullsperger(2), Markus Siegel(1), Katja Fiehler(2), D. Yves von Cramon(2), Andreas K. Engel(1)

(1) Institute of Neurophysiology and Pathophysiology, University Hospital Hamburg-Eppendorf, Hamburg, Germany (2) Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany

Independent component analysis of EEG recorded simultaneously with fMRI identifies medial frontal cortex in performance monitoring

Little is yet known about the relation between scalp-recorded EEG signals and the functional blood oxygen level dependent (BOLD) response measured with magnetic resonance imaging (fMRI). To address this issue, we investigated the error-related negativity (ERN) using simultaneous 3 Tesla fMRI and 32 channel EEG recordings. Data were collected while participants performed a speeded flanker task. Following gradient and pulse artifact correction, single-subject EEGs were submitted to extended infomax independent component analysis (ICA). In all 13 participants a performance-monitoring independent component (PMIC) could be identified. The PMIC was characterized by a radial central dipolar map and reflected the event-related potential differences between error and correct trials, i.e., a response-locked ERN. Moreover, a stimulus-evoked PMIC power increase in the theta frequency range was observed, which was larger for error trials as compared to correct trials. Single-trial PMIC amplitudes were determined by 2-10 Hz filtering the back-projected PMIC and averaging signals across electrodes of interest (Fc1, Fc2, Cz). These single-trial amplitudes were then used as parametric regressor for fMRI analysis. This analysis revealed a significant negative correlation between the single-trial PMIC amplitude and the BOLD response in the rostral cingulate zone (RCZ). A larger single-trial amplitude was associated with a larger RCZ BOLD response. Our approach confirms a trial-by-trial coupling of scalp-recorded EEG and fMRI in an event-related design.

Peter Ullsperger(1)*, Hilit Serby(2), Scott Makeig(2)

(1) Federal Institute for Occupational Safety and Health, Berlin, Germany (2) Swartz Center, Institute for Neural Computation, University of California San Diego, La Jolla CA

Beyond averaging: exploring N400 by independent component analysis

Event-related brain dynamics of 20 younger and 8 older participants (mean ages, 24 and 48 years) were recorded during a semantic decision task involving auditorily presented German noun pairs. Each trial began with a prime word, followed after 1100 ms by a target word. The inter- trial asynchrony was 2400 ms. Participants indicated by choice button press whether each target word was semantically related or unrelated to the previous prime word. A total of 240 word pairs (120 related and 120 unrelated) were presented to each subject twice. EEG was recorded from 60 scalp channels. N400 deflections in average event-related potentials (ERPs) time locked to target word onsets were larger following unrelated targets. ICA, performed on the continuous data after artifact rejection, produced several clusters of independent components (ICs). Most clusters had similar ERPs to prime and target word presentations. One IC cluster accounted for much of the N400 response difference. ERP-image plots of sorted single-trial epochs of channel data or IC activations, smoothed across neighboring trials, suggest several insights into the semantic processing of the target words that could not be gained from conventional averages. While further systematic investigations are needed to understand the interplay between independent brain activities during cognitive information processing, the results demonstrate the value for ERP analysis of disentangling the activities of spatiotemporally overlapping brain processes in single trials.

Julie Onton*, Arnaud Delorme, Scott Makeig

Swartz Center, Institute for Neural Computation, University of California San Diego, La Jolla CA

Frontal midline EEG dynamics during working memory

During a working memory task, an electroencephalographic (EEG) process originating in or near dorsal anterior cingulate cortex exhibited several characteristic modes of oscillatory activity in at least three frequency bands whose amplitudes varied widely across trials. During presentations of letter series to be memorized or ignored, this process, identified by independent component analysis (ICA) of the unaveraged data, expressed distinct frontal midline theta (FMT, 5-7 Hz) and low-beta (12-15 Hz) activities that increased weakly, on average, with memory load. Low-beta activity, but not FMT, was larger during memorize letters than during ignore letter presentations. Following onset of a probe letter after each letter sequence, the same FMT processes emitted a brief burst of 3-Hz activity whose amplitude was unaffected by memory load or by the latency of the succeeding subject button press. The weak mean FMT increase with increasing memory load was produced by progressively larger theta power in a relatively small fraction of FMT component trials, this increase accounting for the entire mean theta increase with memory load on the frontal midline scalp. Trial-to-trial variations in theta power co-varied moderately with theta power in other frontal and left temporal processes. Some of this variability may be linked to trial-to-trial differences in task demands and behavioral context.

Martin J. McKeown*, Wing Lok Au

Parkinson's Research Centre, University of British Columbia, Vancouver, Canada

Frontal theta rhythms in normal subjects and subjects with Parkinson's disease

Theta oscillations in the frontal cortex are a marker for many neural processes including acute changes in attention and protracted drifts in arousal. However isolating the contributions of these different neural functions to midline theta oscillations in human EEG is challenging due to their spatiotemporal overlap. We describe results from two studies incorporating joystick tracking tasks of varying difficulty in normal subjects and subjects with Parkinson's disease (PD) before and after L-dopa medication. Independent Component Analysis successfully isolated two temporally distinct, but spatially overlapping patterns of theta oscillations over the frontal midline EEG. In most subjects we isolated a 'diffuse theta' component which was characterized by temporally sparse, high amplitude bursts of theta oscillations over the experiment and a 'focal midline theta' (FMT) component characterized by a continual, lower amplitude theta oscillations that had a more slowly varying modulation. The power and bursting frequency of the FMT component was modulated by changes in task complexity independent from tracking error. L-dopa medication surprisingly worsened performance in some PD subjects, but had unreliable influence on FMT bursting. These findings suggest that theta oscillations in the raw frontal EEG may be due to two functionally distinct neural processes separable by their unique spatio-temporal profiles, and L-dopa medication may be a less effective treatment for deficits which involve phasic as opposed to tonic changes of dopamine.

Return to Current Abstracts