When most fMRI connectivity cannot be detected: Insights from time course reliability

The level of correlation between two phenomena is limited by the accuracy at which these phenomena are measured. Despite numerous group reliability studies, the strength of the fMRI connectivity that can be detected given the within-subject time course reliability remains elusive. Moreover, it is unclear how within-subject time course reliability limits the robust detection of connectivity on the group level. We estimated connectivity from a working memory task. The grand mean connectivity of the connectome equaled r = 0.41 (95% CI 0.31-0.50) for the test run and r = 0.40 (95% CI 0.29-0.49) for the retest run. The mean connectivity decreased to r = 0.09 (95% CI 0.03-0.16) when test-retest reliability and auto-correlations of single time courses were considered, indicating that less than a quarter of connectivity is detectable. The square root of the detectable connectivity r = 0.09 suggests that only 0.81% of the connectivity is explained by working memory-related time course fluctuations. Null hypothesis significance testing (NHST)-based analysis revealed that within-subject time course reliability markedly affects the significance levels at which paths can be detected at the group level. This was in particular the case when samples were small or connectome coordinates were randomly selected. With a sample of 50 individuals, the connectome of a test session was completely reproduced in a retest session at P < 2.54e-6 despite the fact that almost no connectivity was explained by the cognitive experiment. Within-subject time course reliability can offer valuable insights on the detectable connectivity and should be assessed more frequently.