Short-term test-retest reliability of the human intrinsic functional connectome

Abstract

Functional connectivity is frequently used to quantify the complex synchronous distributed fluctuations in neuronal activity derived from functional Magnetic Resonance Imaging and to generate network representations of human brain function. Such \"functional connectomes\" have great promise for mechanistic studies and for clinical translation. However, we do not know to what extent a functional connectome is stable over time for an individual. In the present work, we evaluate the short-term test-retest reliability of functional connectomes in a large publicly available sample of healthy participants (N=833) scanned on two consecutive days. We also assess the consequences on reliability of three methodological procedures for which a clear guideline in the community is lacking: atlas choice, global signal regression and thresholding. By adopting the intraclass correlation coefficient as a reliability metric, we demonstrate that a relatively small portion of the intrinsic functional connectome is characterized by good (4-6%) to excellent (0.08-1%) stability over time. In particular, connectivity between prefrontal, parietal and temporal areas appears to be especially stable over short timescales. Also, while unreliable edges of the functional connectome are generally weak in terms of average functional connectivity, reliable edges are not necessarily strong. Methodologically, we demonstrate that multimodal parcellation and averaging of connections within known networks are practices that improve reliability. Harnessing this knowledge, for example by honing in on the reliable portion of the connectome, offers one way forward for studies of trait-like features within the normative connectome and for discovery of biomarkers in clinical cohorts.