Functional connectivity eigennetworks reveal different brain dynamics in multiple sclerosis patients

Abstract

Resting state functional connectivity is defined as correlations in brain activity measured by functional magnetic resonance imaging without any stimulation paradigm. Such connectivity is dynamic, even over the course of minutes, and the development of tools for its analysis is an important challenge in neuroscience. We propose a novel data-driven technique to extract connectivity patterns from dynamic whole-brain networks of multiple subjects. Our technique is based on singular value decomposition and decomposes a collection of networks into linearly independent “eigennetworks” and associated time courses. To deal with the temporal redundancy of networks, we propose a novel subsampling method based on the standard deviation of the connectivity strength. We apply the proposed technique to dynamic resting-state networks of healthy subjects and multiple sclerosis patients, and show its potential to detect aberrant connectivity patterns in patients.