Dynamic changes of large-scale resting-state functional networks in major depressive disorder

Abstract

Sliding window method is widely used to study the functional connectivity dynamics in brain networks. A key issue of this method is how to choose the window length and number of clusters across different window length. Here, we introduced a universal method to determine the optimal window length and number of clusters and applied it to study the dynamic functional network connectivity (FNC) in major depressive disorder (MDD). Specifically, we first extracted the resting-state networks (RSNs) in 27 medication-free MDD patients and 54 healthy controls using group independent component analysis (ICA), and constructed the dynamic FNC patterns for each subject in the window range of 10–80 repetition times (TRs) using sliding window method. Then, litekmeans algorithm was utilized to cluster the FNC patterns corresponding to each window length into 2–20 clusters. The optimal number of clusters was determined by voting method and the optimal window length was determined by identifying the most representative window length. Finally, 8 recurring FNC patterns regarded as FNC states were captured for further analyzing the dynamic attributes. Our results revealed that MDD patients showed increased mean dwell time and fraction of time spent in state #5, and the mean dwell time is correlated with depression symptom load. Additionally, compared with healthy controls, MDD patients had significantly reduced FNC within FPN in state #7. Our study reported a new approach to determine the optimal window length and number of clusters, which may facilitate the future study of the functional dynamics. These findings about MDD using dynamic FNC analyses provide new evidence to better understand the neuropathology of MDD.