|
Roberto D. Pascual-Marqui1, Rolando J. Biscay-Lirio2, Michaela Esslen3, Martin Meyer3, Lorena R.R. Gianotti1, Pascal Faber1, Kieko Kochi1, Dietrich Lehmann1
1The KEY Institute for Brain-Mind Research, University Hospital of Psychiatry, Zurich 2Institute for Cybernetics, Mathematics, and Physics, Havana 3Institute of Psychology, University of Zurich The emergence of coherent and phase-synchronized oscillatory electric activity between different brain areas is thought to be the basis of human cognition. It is therefore of great interest to compute, non-invasively, signals of cortical electric activity, and to have an adequate quantification of cortical connectivity for the discovery of oscillatory cortical networks. A major part of published research in this field makes use of coherence and phase synchronization measures between non-invasive signals of scalp electric potentials (EEG). There are two arguments that partly invalidate this procedure: 1. A signal at a scalp electrode does not reflect exclusively the activity of the underlying cortex. 2. Due to volume conduction, incorrect high coherence and phase synchronization values are obtained even when there are non-coherent and desynchronized sources. One proposed methodology for solving these problems consists of estimating cortical electric signals non-invasively with EEG-based neuroimaging techniques. The particular method proposed is exact low resolution brain electromagnetic tomography (eLORETA), which belongs to the family of linear, discrete, 3D distributed tomographies. It is a genuine inverse solution with exact, zero error localization bias even in the presence of measurement noise and structured biological noise (http://arxiv.org/abs/0710.3341). However, due to the low spatial resolution of the method, the cortical electric signals will appear to be instantaneously coherent and synchronized. This problem can be partly solved by decomposing the classical measures of coherence and phase synchronization into instantaneous and lagged (causal) measures of connectivity. The lagged component has a pure physiological origin, and is not biased by volume conduction and low spatial resolution. The new methods proposed here (http://arxiv.org/abs/0706.1776 and http://arxiv.org/abs/0711.1455) differ from previous work (Nolte et al 2004, Clin Neurophysiol 115: 2292-2307; Stam et al 2007, Hum Brain Mapp 28: 1178-1193) in that the instantaneous common sources are explicitly modeled and eliminated. Simulation results show that the new measures are much less affected by common sources. This methodology has two straightforward extensions: it can be applied to time-varying oscillatory activity, and it can be applied to pairs or groups of multivariate time series. The methods are illustrated in the analysis of EEG recordings during a verbal fluency task, compared to the resting state (default mode). |
|