The ICA algorithm selected the optimal number of components for each subject, ranging from 32 to 57. The mean component numbers were 42.3, 44.8, and 42.9 for the control, bipolar disorder, and schizophrenia groups, respectively. Note that these component maps do not reflect areas of task-related blood flow change, but rather a collection of voxels with coherent activity. An automated two-step process was next used to select the component in each subject that most closely matched the DMN as previously described (Greicius et al., 2004). Briefly, a frequency filter was first applied to remove any components in which high-frequency signal (>0.1 Hz) constituted 50% or more of the total power in the Fourier spectrum. A template of the DMN (templates and software courtesy of Dr. Michael Greicius, Stanford University, Stanford, CA) was next used to select the “best-fit” component in each subject where the sum of z-scores inside the template minus the sum of z-scores outside it was maximal (Greicius et al., 2004). For each subject, we visually inspected all component spatial maps to ensure that the appropriate DMN was chosen by this procedure, and did not detect any discrepancies.
