Disclosing large-scale directed functional connections in MEG with the multivariate phase slope index

Abstract

The phase slope index (PSI) is a method to disclose the direction of frequency-specific neural interactions from magnetoencephalographic (MEG) time series. A fundamental property of PSI is that of vanishing for linear mixing of independent neural sources. This property allows PSI to cope with the artificial instantaneous connectivity among MEG sensors or brain sources induced by the field spread. Nevertheless, PSI is limited by being a bivariate estimator of directionality as opposite to the multidimensional nature of brain activity as revealed by MEG. The purpose of this work is to provide a multivariate generalization of PSI. We termed this measure as the multivariate phase slope index (MPSI). In order to test the ability of MPSI in estimating the directionality, and to compare the MPSI results to those obtained by bivariate PSI approaches based on maximizing imaginary part of coherency and on canonical correlation analysis, we used extensive simulations. We proved that MPSI achieves the highest performance and that in a large number of simulated cases, the bivariate methods, as opposed to MPSI, do not detect a statistically significant directionality. Finally, we applied MPSI to assess seed-based directed functional connectivity in the alpha band from resting state MEG data of 61 subjects from the Human Connectome Project. The obtained results highlight a directed functional coupling in the alpha band between the primary visual cortex and several key regions of well-known resting state networks, e.g. dorsal attention network and fronto-parietal network.

Keywords: Directed functional connectivity; Human connectome project (HCP); Magnetoencephalography (MEG); Multivariate method; Phase slope index (PSI); Resting state networks (RSN).