Frequency-specific brain network architecture in resting-state fMRI

The analysis of brain function in resting-state network (RSN) models, which has been found in the functional connectivity pattern of resting-state functional magnetic resonance imaging (rs-fMRI), is sufficiently powerful for studying large-scale functional integration of the brain. Although there has been an increasing interest in the relatively higher frequency of rs-fMRI data, the network architecture has been regarded as the same through different frequency bands in RSN-based research. This study examined whether the network architecture changes with frequency. The blood-oxygen-level-dependent (BOLD) signal was decomposed into four frequency bands (ranging from 0.007 Hz to 0.438 Hz), for each of which the clustering algorithm was applied. The best clustering number was selected for each frequency-band based on the overlap ratio with task activation maps provided by Neurosynth. The results demonstrate that (1) resting-state BOLD signals have frequency-specific network architecture, that is, the networks finely subdivided in the lower frequency bands are integrated into fewer networks in higher frequency bands rather than reconfigured, and (2) the default mode network(DMN) is the only associative network that has a strong enough architecture to survive the increasing noise in higher frequency bands. These findings provide a novel framework that enables a better understanding of brain function through the multiband frequency analysis of ultra-slow rs-fMRI data.