Subject-level time series processing

All functional time series data were preprocessed using a procedure that has been previously validated in large-scale developmental datasets and has been shown to be highly effective at reducing the influence of motion artifact on connectivity data [40–42]. A recent benchmarking paper comparing more than a dozen preprocessing pipelines for resting-state fMRI data demonstrated that the 36-parameter models are considered an optimal approach for pediatric group comparisons compared to the most common non-GSR pipeline (24-parameter) [43]. Preprocessing included removal of the first four volumes to allow for signal stabilization, slice time correction, realignment to the median volume, brain extraction, spatial smoothing (7 mm FWHM), and grand mean scaling. Mean white matter (WM) and cerebrospinal fluid (CSF) signals were extracted from the filtered time series data using tissue segments generated for each subject. Improved confound regression included nine standard confound signals (six motion parameters + global/WM/CSF) as well as the temporal derivative, quadratic term, and the temporal derivative of the quadratic term (36 parameters in total). We band-pass filtered the functional time series and the confound regressors simultaneously to retain frequencies between 0.01 and 0.08 Hz; identical temporal filtering prevented frequency mismatch between the confound parameters and the time series data [44].