A Multivariate Model for Comparison of Two Datasets and its Application to FMRI Analysis

In this work, we propose a structured approach to compare common and distinct features of two multidimensional datasets using a combination of canonical correlation analysis (CCA) and independent component analysis (ICA). We develop formulations of information theoretic criteria to determine the dimension of the subspaces for common and distinct features of the two datasets. We apply the proposed method to a simulated dataset to demonstrate that it improves the estimation of both common and distinct features when compared to performing ICA on the concatenation of two datasets. We also apply the method to compare brain activation in functional magnetic resonance imaging (fMRI) data acquired during a simulated driving experiment and observe distinctions between the driving and watching conditions revealed in relevant brain function studies.