Test-Retest Reliability of Dynamic Functional Connectivity Parameters for a Two-State Model

Abstract

Dynamic functional connectivity (DFC) metrics derived from resting-state functional magnetic resonance imaging (rs-fMRI) that measure dynamic features of brain communications that change over time are gaining more and more attention. A key concern is the reliability of DFC parameters, especially given the many methodological options. The DFC parameters mean dwell time (MDT), prevalence (Prev.), inter-transition interval (ITI) and state variability (Var.) were computed in 23 participants using a two-state model, sliding window analysis, and two imaging sessions. Reliability of the connectivity states, test-retest reliability of the parameters as well as correlations between DFC parameters were investigated for different scan lengths (i.e., 16 vs 8 min.), atlases (i.e., 116 vs. 442 regions of interest) and with/without within-subject centering of DFC matrices. The results showed an integrated (I) and a segregated (S) brain state with high intra-class correlation (ICC) values of the states between sessions (0.67 [≥] ICC [≥] 0.99). Reliability of the DFC parameters is, at best, intermediate with an ICC = 0.51 for prevalence for not-centered, 16 min. data and the coarse atlas. Moreover, MDTs of both states were positively correlated for centered data. Prevalence of one state was positively correlated with MDT of the same state and frequently negatively correlated with MDT of the other state. Var was positively correlated with MDT and Prev for the same state and negatively correlated for the other state for not-centered data. This study suggests that prevalence is the most reliable DFC parameter with lower reliability for shorter scans, more atlas regions and centering. Thus, it is advantageous to formulate hypotheses related to brain dynamics in terms of prevalence, especially in small-scale studies.