Methods to estimate functional and effective brain connectivity from MEG data robust to artifacts of volume conduction

Abstract

Due to the high temporal resolution of MEG data, they are well suited to study brain dynamics, while the limited spatial resolution constitutes a major confounder when one wants to estimate brain connectivity. To a very large extent, functional relationships between MEG sensors and estimated sources are caused by incomplete demixing of the brain sources. Many measures of functional and effective connectivity are highly sensitive to such mixing artifacts. In this chapter, we review methods that address this problem. They are all based on the insight that the imaginary part of the cross-spectra cannot be explained as a mixing artifact. Several variants of this idea will be presented. We will present three different methods adapted to localize source interactions: (a) minimum overlap component analysis (MOCA) decomposes linear estimates of the P most relevant singular vectors of the imaginary parts of the cross-spectra, (b) the MUSIC algorithm can be applied to this same subspace, and (c) the estimated sources can be analyzed further using multivariate generalizations of the imaginary part of coherency. Finally, a causal relation between these sources can be estimated using the phase slope index (PSI). The methods will be illustrated for empirical MEG data of a single subject under resting state condition.