Adaptive window selection in estimating dynamic functional connectivity of resting-state fMRI

Investigation of the intrinsic brain networks using the resting-state functional magnetic resonance imaging (rs-fMRI) is generally based on the assumption that the functional organization is stationary across the duration of the scan. Hence, the presence and potential of temporal and spatial dynamics of the functional connectivity (FC), which is usually measured by the correlation coefficients between rs-fMRI signals of two regions, are not taken into account in most of the research. Recent studies have shown that the resting-state brain activities are time-varying in nature with substantial dynamic characteristics. However, an effective method for estimating the time-varying FC is lacking, which is mainly due to the difficulty in choosing an appropriate window to localize the time-varying correlation coefficients (TVCC). In this paper, we introduce a novel method for adaptively estimating the TVCC of non-stationary signals and study its application to infer dynamic FC of rs-fMRI data. The proposed method employs a sliding window with adaptive size, which is selected by a local plug-in rule to minimize the mean square error, to estimate the TVCC locally. Simulation results on synthetic data show that the proposed method outperforms the conventional TVCC estimators with a fixed window. Furthermore, the proposed method is used on real rs-fMRI signals, and the results demonstrate that the FC in the resting state are transient and the variability of FCs between different brain regions differs substantially.