Recently, functional connectome studies based on resting-state functional magnetic resonance imaging (R-fMRI) and graph theory have greatly advanced our understanding of the topological principles of healthy and diseased brains. However, how different strategies for R-fMRI data preprocessing and for connectome analyses jointly affect topological characterization and contrastive research of brain networks remains to be elucidated. Here, we used two R-fMRI data sets, a healthy young adult data set and an Alzheimer's disease (AD) patient data set, and up to 42 analysis strategies to comprehensively investigate the joint influence of three key factors (global signal regression, regional parcellation schemes, and null network models) on the topological analysis and contrastive research of whole-brain functional networks. At the global level, we first found that these three factors affected not only the quantitative values but also the individual variability profile in small-world related metrics and modularity, wherein global signal regression exhibited the predominant influence. Moreover, strategies without global signal regression and with topological randomization null model enhanced the sensitivity of the detection of differences between AD and control groups in small-worldness and modularity. At the nodal level, strategies of global signal regression dominantly influenced the spatial distribution of both hubs and between-group differences in terms of nodal degree centrality. Together, we highlight the remarkable joint influence of global signal regression, regional parcellation schemes and null network models on functional connectome analyses in both health and diseases, which may provide guidance for the choice of analysis strategies in future functional network studies.
Keywords: connectomics; graph theory; hub; modularity; small-worldness.
© 2018 Wiley Periodicals, Inc.