New Insights into Human Brain Dynamics

Scott Makeig
University of California San Diego
Institute for Neural Computation
The Salk Institute, La Jolla CA 92037
>smakeigucsd.edu

Rapid advances in available technology for recording high-density EEG, MEG and fMRI signals from the human brain afford an unprecedented opportunity to observe and model human brain dynamics during a very wide range of human experience and task performance. Interpreting the mass of derived data requires new computational tools. An important emerging area in signal processing of biomedical signals is independent component analysis (ICA), a major approach to blind signal separation that explicitly attempts to separate the total information content of the input data into informationally distinct components. Work at Salk Institute, UCSD, and elsewhere over the last five years has demonstrated the utility of ICA for studying event-related human brain dynamics using EEG, MEG, fMRI and other psychophysiological signals. The symposium will introduce the use of ICA for biomedical signal processing and give more detailed presentations a number of approaches to analysis of EEG and MEG brain dynamics during perception and processing of visual and auditory events in task paradigms requiring subject attention and behavioral responses.

Presenters:

Terrence J. Sejnowski - Independent components of psychophysiological time series
Tzyy-Ping Jung, Analyzing single-trial event-related potentials
Marissa Westerfield, Applying single-trial analysis to clinical event-related potential data.
Akayha Tang, Measuring populational response timing in humans using magnetoencephalography and blind source separation
Scott Makeig, Transient event-related coherence between independent EEG components

Independent components of psychophysiological time series

T.J. Sejnowski
Howard Hughes Medical Institute
Computational Neurobiology Lab, Salk Institute, La Jolla, CA
Department of Biology, University of California San Diego, La Jolla CA
terry at salk.edu

Biomedical signals from many sources including hearts and brains pose a challenge to researchers who may have to separate weak signals arriving from multiple sources contaminated with artifacts and noise. The analysis of these signals is important both for research and for medical diagnosis and treatment. The applications of Independent Component Analysis (ICA) to biomedical signals is a rapidly expanding area of research and many groups are now actively engaged in exploring the potential of blind signal separation and signal deconvolution for revealing new information about the brain and body. I will survey some recent applications of ICA to a variety of biomedical signals including functional magnetic resonance imaging (fMRI) and the electrocardiogram (ECG), and will discuss contributions made using ICA to the study of sparse neural coding in the visual system.

Analyzing single-trial event-related potentials

T-P Jung, S. Makeig and T.J. Sejnowski
Institute for Neural Computation, UCSD, La Jolla, CA
Computational Neurobiology Lab, Salk Institute, La Jolla, CA
jung at salk.edu

Event-related potentials (ERPs), portions of EEG signals that are both time- and phase-locked to some experimental events, are usually averaged to increase their signal/noise ratio relative to non-phase locked EEG activity regardless of the fact that in single stimulus epochs response activity may vary widely in both time course and scalp distribution. This study applies a linear decomposition method, Independent Component Analysis (ICA), to single-trial ERPs recorded at multiple scalp electrodes to derive spatial filters that decompose complex EEG data into a sum of temporally independent and spatially fixed components. We have explored applications of ERP-image and ICA decomposition to single-trial ERPs in a visual selective attention task involving, (1) removing eye and muscle artifacts that interfere with EEG analysis, while preserving the underlying brain activity in the EEG; (2) extracting event-related responses from spontaneous EEG; (3) identifying spatially-overlapping patterns of coherent activity rather than focusing on single scalp channels or channel pairs as in all current analysis methods; (4) separating oscillatory EEG activity into several components with distinct frequency contents.

Applying single-trial analysis to clinical event-related potential data.

Marissa Westerfield, Jeanne Townsend, Eric Courchesne, Tzyy-Ping Jung
Institute for Neural Computation, UCSD, La Jolla, CA
Laboratory on Research on the Origins of Autism, Childrens Hospital Research Center, San Diego
marissa at salk.edu

Because the scalp recorded event-related potential (ERP) is a sum of brain electrical activities from different brain generators, it is difficult to separate the contributions from separate brain networks when analyses are limited to fairly gross measurements such as peak latency, area or amplitude. Analytic techniques including Independent Component Analysis (ICA) show promise in reducing ERPs waveforms to their component parts, each of which possibly reflect activity in a specific brain region. We have first used traditional ERP analysis techniques to anlayzre response averages preprocessed using ICA to remove gross artifacts and have demonstrated abnormalities (specifically in N1 and P300) in ERPs elicited by a visual spatial task in cerebellar lesion and cortical lesion subjects. This talk will summarize those results, describe the additional information revealed by Independent Component Analysis of the single trials from the same experiments, and discuss implications for possible brain areas involved in the generation of the scalp-recorded signals.

Measuring populational response timing in humans using magnetoencephalography and blind source separation

Akaysha C. Tang, Barak A. Pearlmutter
Department of Psychology
Department of Neurosciences,
Department of Computer Science
University of New Mexico, NM 87131
akaysha at kongzi.unm.edu

Magnetoencephalography (MEG) has millisecond temporal resolution, allowing for the observation of rapidly changing neuronal activity. Compared to EEG, which has similar temporal resolution, MEG has a superior spatial sensitivity (a few millimeters). In the past, due to the relative small signal amplitudes compared to various noise sources, heavy signal averaging over many trials of repeated measures was required to obtain the spatial localization of the neuronal populations under study. Therefore, it was difficult to measure trial-to-trial changes in the precise timing of populational neuronal responses. Combining the Second Order Blind Identification (SOBI), one of the blind source separation methods, with standard Neuromag source localization tools, and co-registering MEG and MRI, we were able to localize the SOBI separated components to anatomical locations consistent with known neuroanatomy and neurophysiology, therefore establishing a match between a BSS separated component and a neuronal population at a given anatomical location. Because SOBI, like ICA, simultaneously separate neuronal components from various noise components, the separated neuronal components are much less contaminated by various noise sources, making single-trial measurement of populational response time possible. We will survey recent results brought about by the above described enabling techniques, in measuring inter-hemispherical transfer time (IHTT), variability in populational neuronal response timing, and changes in populational response timing as a result of learning. (Supported by grants from the National Foundation for Functional Brain Imaging and National Science Foundation and by a gift from George Cowan).

Transient event-related coherence between independent EEG components

Scott Makeig
University of California San Diego, La Jolla CA
Institute for Neural Computation
UCSD, La Jolla CA 92037
smakeig at ucsd.edu

Event-related coherence in the human EEG, computed using a short-time Fourier wavelet decomposition, shows multiple synchronization frequencies with abrupt event-related jumps in coherence frequency and, less frequently, in synchronization lag. Since projections from different brain generators typically overlap, single scalp electrode channels typically pick up and sum activity from a relatively large number of discrete EEG sources. Changes in coherence between two scalp electrode channels thus may reflect any of several types of changes in the summed mixtures of activities they record. Computing coherence between independent components (ICs) of the EEG, derived using the infomax ICA algorithm, increases the functional specificity of the derived results. Study of event-related coherence between ICs suggests that cooperativity between brain networks may be accomplished through joint oscillatory processes operating at multiple discrete (theta to gamma) frequencies, only some of which may appear as peaks in the EEG power spectra, and these oscillatory linkages may be transiently created, broken or transformed by experimental events.