Functional data postprocessing

Outputs of fMRIPrep were post-processed by XCP-D 0.0.8 (NKI), 0.3.0 (HCP-D), and 0.3.2 (PNC and HBN). XCP-D⁸³ is an extension of the eXtensible Connectivity Pipeline Engine (XCP)^31,32 and was developed to mitigate motion-related artifacts and noise in functional MRI data from developmental cohorts. First, outlier detection was performed. In order to identify high-motion outlier volumes, framewise displacement was calculated⁹² with a head radius of 50 mm. Then, the BOLD data were despiked, mean-centered, and linearly detrended. Despiking is a temporal censoring operation that performs similarly to scrubbing in prior benchmarking studies⁷⁵. Thirty-six confounds were estimated based from the preprocessed timeseries in fMRIPrep: six motion parameters, mean global signal, mean white matter signal, mean CSF signal with their temporal derivatives, and the quadratic expansion of six motion parameters, tissues signals and their temporal derivatives^31,75. The 36 nuisance regressors were regressed from the BOLD data using linear regression as implemented in Scikit-Learn 0.24.2 (NKI), Scikit-Learn 1.1.3 (HCP-D), or nilearn 0.9.2 (PNC and HBN). Processed functional timeseries were extracted from residual BOLD using Connectome Workbench⁹¹ for the following atlases: the Schaefer 17-network 200 and 400 parcel atlas⁴⁰, the HCP-MMP atlas⁹³, and the Gordon atlas⁹⁴. The Schaefer 200 atlas was used as the primary atlas and Schaefer 400, HCP-MMP, and Gordon atlases were used in sensitivity analyses. Lastly, parcellated rest and task fMRI timeseries were concatenated and the Pearson correlation between concatenated timeseries was computed for every pair of cortical regions.