The more, the merrier: multivariate phase synchronization methods excel pairwise ones in estimating brain connectivity from reconstructed neural sources

Abstract

The estimation of functional connectivity (FC) from electro-(EEG) or magnetoencephalographic (MEG) recordings suffers from low spatial resolution, being one of the reasons for the reduced number of sensors compared to the number of reconstructed sources of activity. This problem can be avoided by estimating FC between larger regions instead of individual sources. However, combining all the sources in each area to produce a single time series per region is far from trivial. We have used simultaneous EEG/MEG data from 11 participants and compared the FC estimates from both techniques by using different multivariate approaches. Since the underlying generators are identical for EEG and MEG, the more similar the FC estimation from both techniques is, the more accurate it is likely to be. The results show that using either the average or the root-mean-square of the bivariate source-to-source FC estimates consistently outperforms the use of a representative time series from each area. We concluded that the reconstructed activity in each brain region is too complex to be reduced to a single representative time series and that full multivariate approaches must be used to describe between-area FC from electrophysiological recordings accurately. Moreover, the high correlation between the FC values estimated from EEG and MEG suggests that the results found in the high-sensitivity, low-noise MEG can be transferable to the more affordable EEG, at least when high-quality source reconstruction is used.