Spatial ICA was performed using GIFT28 on preprocessed data for all participants. As implemented in GIFT, ICA identifies networks of brain regions with temporally correlated oscillations. The joint fMRI data from the participants is structured by a temporal concatenation approach that assumes a common aggregate spatial map (ie, common funcitonal anatomy) but allows for participant-specific differences in the time-dependent networks signals. Initial data reduction was achieved using a two-step standard principal components analysis process (participant- and group-level, as described by Erhardt et al.)29 with the full covariance matrix computed from stacked data sets in conjunction with packed storage, single precision, and selective eigen computation options available in GIFT.28 Resting state time series data were normalized by removing the image mean per timepoint.29 ICA was conducted using the FastICA algorithm,30 selecting the symmetrical approach, tanh nonlinearity, and stabilization parameters in GIFT. Algorithmic reliability was also evaluated because the methods used here are stochastic and iterative, such that model solutions vary across runs. Using ICASSO, each ICA was repeated 20 times to evaluate the reliability and stability of resultant components through bootstrapping
