Abstract
While graph theoretical modeling has dramatically advanced our understanding of complex brain systems, the feasibility of aggregating brain graphic data in large imaging consortia remains unclear. Here, using a battery of cognitive, emotional and resting fMRI paradigms, we investigated the reproducibility of functional connectomic measures across multiple sites and sessions. Our results revealed overall fair to excellent reliability for a majority of measures during both rest and tasks, in particular for those quantifying connectivity strength, network segregation and network integration. Higher reliabilities were detected for cognitive tasks (vs rest) and for weighted networks (vs binary networks). While network diagnostics for several primary functional systems were consistently reliable independently of paradigm, those for cognitive-emotional systems were reliable predominantly when challenged by task. Different data aggregation approaches yielded significantly different reliability. In addition, we showed that after accounting for observed reliability, satisfactory statistical power can be achieved in the multisite context with a total sample size of approximately 250 when the effect size is at least moderate. Our findings provide direct evidence for the generalizability of brain graphs for both resting and task paradigms in large consortia and encourage the use of multisite, multisession scans to enhance power for human functional connectomic studies.