Complexity organization of resting‐state functional‐MRI networks

Abstract

AbstractEntropy measures are increasingly being used to analyze the structure of neural activity observed by functional magnetic resonance imaging (fMRI), with resting‐state networks (RSNs) being of interest for their reproducible descriptions of the brain's functional architecture. Temporal correlations have shown a dichotomy among these networks: those that engage with the environment, known as extrinsic, which include the visual and sensorimotor networks; and those associated with executive control and self‐referencing, known as intrinsic, which include the default mode network and the frontoparietal control network. While these inter‐voxel temporal correlations enable the assessment of synchrony among the components of individual networks, entropic measures introduce an intra‐voxel assessment that quantifies signal features encoded within each blood oxygen level‐dependent (BOLD) time series. As a result, this framework offers insights into comprehending the representation and processing of information within fMRI signals. Multiscale entropy (MSE) has been proposed as a useful measure for characterizing the entropy of neural activity across different temporal scales. This measure of temporal entropy in BOLD data is dependent on the length of the time series; thus, high‐quality data with fine‐grained temporal resolution and a sufficient number of time frames is needed to improve entropy precision. We apply MSE to the Midnight Scan Club, a highly sampled and well‐characterized publicly available dataset, to analyze the entropy distribution of RSNs and evaluate its ability to distinguish between different functional networks. Entropy profiles are compared across temporal scales and RSNs. Our results have shown that the spatial distribution of entropy at infra‐slow frequencies (0.005–0.1 Hz) reproduces known parcellations of RSNs. We found a complexity hierarchy between intrinsic and extrinsic RSNs, with intrinsic networks robustly exhibiting higher entropy than extrinsic networks. Finally, we found new evidence that the topography of entropy in the posterior cerebellum exhibits high levels of entropy comparable to that of intrinsic RSNs.