Figure 1.

Estimation of the functional (fMRI) data feature space. Aggregate states were estimated by decomposing the windowed correlations by temporal ICA. Subject-specific states were next estimated through a spatio-temporal (dual) regression procedure wherein, for each subject, the aggregate states were regressed into the subject's windowed FNC data to estimate subject-specific component time-courses in the first regression step, and the estimated time-courses were regressed into the subject's windowed FNC to derive the subject-specific states in the second regression step; (B) Summary of the mCCA + jICA framework. For each subject, the functional data feature space as estimated in (A) was concatenated with the smoothed, modulated and warped gray matter maps (as the structural data feature space) and fused using the joint “mCCA+jICA” framework. This framework combines the mCCA and jICA algorithms to decompose the observed data into a linear combination of sources mixed through “effective” modality-specific mixing matrices as illustrated above.