Reproducibility of resting-state functional connectivity: a mini-multiverse analysis

Resting-state functional magnetic resonance imaging (rsfMRI) is one of the most widely applied tools in the network neurosciences, but the reliability of the metrics derived from its signal remains unclear. We use back-to-back resting-state scans in a healthy (N = 41) and clinical sample (traumatic brain injury; TBI, N = 45) to assess the replicability of resting-state functional connectivity (RSFC) using a “mini” multiverse approach. The goal was to establish the most reliable graph metrics and determine if moderate-severe TBI has an influence on RSFC reliability using intraclass correlation coefficients (ICCs). Findings revealed that graph metrics were reliable in both healthy adults and individuals with TBI. Metrics such as within-network connectivity and characteristic path length were most reliable whereas other whole-brain connectivity estimates (e.g., clustering coefficient, eigenvector centrality) were least reliable. The default mode and salience networks were the most reliable resting-state canonical networks. Choice of brain atlas had a modest effect on findings. There was a notable influence of motion scrubbing on ICCs, with diminished reliability proportional to the number of volumes removed. Overall, RSFC reproducibility is preserved after significant neurological compromise, and we identify a subset of graph metrics and canonical networks with promising reliability. Future investigations should continue to implement multiverse approaches and examine factors that contribute to variations in RSFC reliability.