Stability of the functional connectome across childhood: a five-year dense-sampling case study with multi-subject validation

Functional connectomics has advanced our understanding of child development, neurodiversity, and brain organization. Most studies assume that collected fMRI data approximate the stable, trait-like connectome of interest. Here, we asked how stable a young child's functional connectome is, how stability is affected by scan paradigm (rest vs. passive viewing) and age, and how much data are needed to estimate a "stable connectome." We addressed these questions using a densely sampled single-subject dataset and a multi-subject pediatric cohort. We first collected 25 h of resting-state and passive-viewing data from a single child, ages 5 to 9. Split-half connectome stability was near perfect (r = 0.98), even when data were split chronologically (r = 0.97). Connectome similarity differed across scan paradigms and collection years. Consistent patterns were observed in the multi-subject cohort, where individual-specific effects substantially outweighed movie-specific effects. A linear mixed-effects model showed that greater head motion and longer scan intervals were associated with lower session-to-session stability, whereas older age was associated with greater stability. A whole-connectome similarity threshold of r ≥ 0.8 was reached with 13.5 min of raw passive-viewing data, although many high-motion recordings failed to reach this threshold within the available recording duration. Similarly, ∼15.4 min of post-censor data were required for 70% of participants in the multi-subject cohort to reach r ≥ 0.7, while most high-motion scans failed to reach this threshold. Allowing for typical motion-related data loss (∼22% of volumes censored), these findings support approximately 20-minute acquisitions to obtain about 15 min of usable data, with longer recordings or multiple sessions for children with elevated head motion.