Functional images were generated and processed using a mixture of freeware and commercial packages including the Analysis of Functional NeuroImages (AFNI) [Cox, 1996 (link)], GIFT [Calhoun, 2004 ], MATLAB (Mathworks, Sherborn, MA) and FSL [Smith et al., 2004 (link)] packages. Time series images were first spatially registered (to the third image from the first resting run) in both two- and three- dimensional space to minimize effects of head motion, temporally interpolated to correct for slice-time acquisition differences, de-spiked, linearly detrended and spatially blurred using a 10-mm Gaussian full-width half-maximum filter.
The GIFT software package was then used to calculate the individual components on a subject-by-subject basis. Minimum description length (MDL) was used to establish the number of components necessary to be generated [Calhoun et al., 2001 (link); Rissanen, 1983 ]. The ideal number of components ranged from 7 to 20 across the 42 subjects; therefore, 20 components were generated for all subjects in order to maintain consistency1. Components were calculated for each subject for 3 (corresponding to the first run), 6 (corresponding to the first two runs), and 9 (corresponding to all three runs) minutes of data collection using the Infomax algorithm [Bell and Sejnowski, 1995 (link)]. For the first 3 min of data collection, single-subject single-run ICA was performed. To calculate the components for 6 and 9 min, a group ICA [Calhoun et al., 2001 (link)] was implemented, as a simple concatenation of the time-courses of the individual runs cannot be performed since no baseline of resting state exists. The resulting components from individual and multirun ICAs were then converted to a 1 mm3 standard stereotaxic coordinate space [Talairach and Tournoux, 1988 ].