Validating dynamicity in resting state fMRI with activation‐informed temporal segmentation

Confirming the presence (or absence) of dynamic functional connectivity (dFC) states during rest is an important open question in the field of cognitive neuroscience. The prevailing dFC framework aims to identify dynamics directly from connectivity estimates with a sliding window approach, however this method suffers from several drawbacks including sensitivity to window size and poor test–retest reliability. We hypothesize that time‐varying changes in functional connectivity are mirrored by significant temporal changes in functional activation, and that this coupling can be leveraged to study dFC without the need for a predefined sliding window. Here, we introduce a data‐driven dFC framework, which involves informed segmentation of fMRI time series at candidate FC state transition points estimated from changes in whole‐brain functional activation, rather than a fixed‐length sliding window. We show our approach reliably identifies true cognitive state change points when applied on block‐design working memory task data and outperforms the standard sliding window approach in both accuracy and computational efficiency in this context. When applied to data from four resting state fMRI scanning sessions, our method consistently recovers five reliable FC states, and subject‐specific features derived from these states show significant correlation with behavioral phenotypes of interest (cognitive ability, personality). Overall, these results suggest abrupt whole‐brain changes in activation can be used as a marker for changes in connectivity states and provides new evidence for the existence of time‐varying FC in rest.We present a new approach for assessing dynamic functional connectivity that leverages instantaneous changes in functional activations to detect transition points between brain states and derives an informed segmentation of the fMRI time series. Our approach accurately recovers known state transitions in the context of a structured working memory task, outperforming the popular sliding window approach. In resting data, our method identifies five dynamic states that are highly test‐retest reliable, exhibit complex transition dynamics, and are correlated with multiple behavioral phenotypes.Peer Reviewedhttp://deepblue.lib.umich.edu/bitstream/2027.42/170902/1/hbm25649_am.pdfhttp://deepblue.lib.umich.edu/bitstream/2027.42/170902/2/hbm25649.pd