Human Brain Mapping 2010 Barcelona, Spain
June 6-10, 2010
Tim Mullen1, Julie Onton2, Arnaud Delorme1, Scott Makeig1 , 1Swartz Center for Computational Neuroscience, Institute for Neural Computation, University of California San Diego, La Jolla, CA
2Naval Health Research Center, San Diego, CA
Analysis and Visualization of Theta-band Information Flow Dynamics in an ERN-producing task
Introduction:
A significant challenge in contemporary neuroscience lies in estimating and visualizing the time- and frequency-dependent dynamics of information flow within distributed anatomical networks and relating these dynamics to cognitive phenomena. While several very useful toolboxes exist for analyzing causality and information flow in electrophysiological datasets [1-3], they are often specific to one estimation approach (e.g., bivariate granger causality) and may lack sophisticated visualization routines and integration into comprehensive analysis packages. We are developing a new toolbox geared towards EEG analysis which affords a variety of causality/information flow estimation approaches and visualization routines and is integrated into EEGLAB as a plugin. Here we examine multivariate granger-causal interactions across time, frequency, anatomical location, and subject population in an ERN-producing two-back task.
Methods:
128-channel (256 Hz) EEG data were collected from 24 subjects performing a visual letter two-back task with auditory feedback. Trials were segregated based on response type (Incorrect vs. Correct). Following zero-phase FIR high-pass filtering, response-locked datasets were subjected to Infomax Independent Component Analysis (ICA). ICA is effective at separating source components that are maximally independent, which can be further analyzed for transient dependencies [4]. A single (or dual symmetric) equivalent dipole model was then fit to each independent component (IC). We rejected ICs corresponding to artifacts such as eye blinks and muscle activity, and those with a poor dipole fit (> 15% r.v., or lying outside brain volume). Following standard normalization and detrending preprocessing steps and model order selection, an adaptive multivariate autoregressive model (AMVAR) was fit to the remaining IC activations and the time-varying mutivariate partial coherence (pCoh), directed transfer function (DTF), and partial directed coherence (PDC) were estimated from the MVAR coefficients [5]. The null hypothesis of no information flow was tested using phase randomization [6]. Bootstrap estimates of the distributions of the causal estimators were used to establish statistical significance of deviation from baseline as well as between-condition differences. Finally, original visualization techniques were used to interactively explore individual and group-level information flow dynamics between estimated IC sources across time, frequency and spatial location (in MRI-coregistered 2-D and 3-D spaces).
Results:
Our analysis revealed statistically significant increases in theta-band directed information flow at and following the button press between IC sources localized in or near anterior and posterior cingulate cortex together with sources distributed in medial frontopolar, somatomotor, posterior parietal, and visual cortex. Moreover, there was significantly more information flow following incorrect responses than correct responses. The theta peak (frequently near 5-6 Hz) showed an early maximum immediately following the response, near the latency of the error-related negativity (ERN) potential. Prominent sources of theta outflow included dorsal medial somatomotor, anterior cingulate, posterior cingulate, and posterior parietal cortex.
Conclusions:
We demonstrated several routines from a new toolbox for analysis and visualization of information flow dynamics in electrophysiological data. While the toolbox is geared towards source separated/localized EEG, it can also be applied to MEG or intracranial EEG datasets. Analysis of a two-back task revealed theta-band granger-causal influences between multiple IC sources in or near anterior and posterior cingulate cortex together with sources distributed in medial frontopolar, somatomotor, posterior parietal, and visual cortex. These theta bursts immediately preceded and followed speeded button presses made in error. These findings are consistent with previously published spectral/coherence analysis of other ERN-producing tasks [7] and support the hypothesis that a top-down updating of attention or motor planning following recognition of unintended or previously unanticipated event consequences is indexed by transient theta communication through a distributed corticolimbic network.
References:
[1] Seth, A. (2009), 'A MATLAB toolbox for Granger causal connectivity analysis', Journal of Neuroscience Methods.
[2] Schlögl, A. (1996-2002), 'Time Series Analysis - A toolbox for the use with Matlab'.
[3] Cui, J. (2008), 'BSMART: a Matlab/C toolbox for analysis of multichannel neural time series', Neural Networks, Special Issue on Neuroinformatics, vol. 21, pp. 1094 - 1104.
[4] Makeig, S. (2002), 'Dynamic brain sources of visual evoked responses', Science, vol. 295, pp. 690-694.
[5] Kaminski, M. (2007), 'Multichannel Data Analysis in Biomedical Research', Understanding Complex Systems, Handbook of Brain Connectivity series, pp. 327-355.
[6] Theiler, J. (1992), 'Testing for nonlinearity in time series: the method of surrogate data', Physica D, vol. 58, pp. 77-94.
[7] Luu, P. (2004), 'Frontal-midline theta and the ERN', Clinical Neurophysiology, vol. 115, pp. 1821-35.