Identifying components that vary in space and time from resting-state functional MRI

Abstract

A widespread assumption of fMRI-derived large-scale intrinsic connectivity networks (ICNs) is that they are spatially static over time. However, the assumption of spatial stationarity of ICNs has been challenged by a range of techniques that allow for time-varying connectivity between brain regions and demonstration that canonical networks like the default model network (DMN) can be fractionated according to time-varying connectivity relationships of their subcomponents. Previously, we developed a simple spatiotemporal ICA (stICA) technique to allow the discovery of patterns of spatiotemporal evolution in task fMRI data in a way that avoided the traditional constraint of spatial stationarity on brain networks, and we validated the approach in fMRI of task-to-rest transitions. Here, we apply our stICA technique to resting-state fMRI datasets to explore whether spatiotemporally evolving components of brain activity can be identified in the absence of an overt behavioural task. We found that stICA components could generally be described in terms of graded onsets and offsets of ICNs that had been calculated based on techniques that assumed spatial stationarity. Our results suggest that, to a reasonable approximation, stable ICNs can be taken to be building blocks of the spatiotemporal patterns measured with resting-state fMRI.