A feature-based fusion method for making group inference in epileptic fMRI and DTI using canonical correlation analysis

In recent years, there have been a great interest for combined analysis of functional magnetic resonance imaging (fMRI) and structural MRI (sMRI) data, because they present complementary information of different tissue types. Canonical correlation analysis (CCA) is a simple data fusion scheme to evaluate brain connectivity. We specify the versatility of CCA to extract features of resting state fMRI and Diffusion Tensor Imaging (DTI). The most informative features, ALFF and FA, are extracted from datasets of epilepsy and healthy subjects. CCA has been successfully utilized for joint data analysis such as combined analysis of EEG and fMRI of a single subject. In the current work, we present a new technique for combination of two modalities across subjects and back reconstruction of components for each group and each subject. Our results indicate that temporal gyrus, cuneus, posterior cingulate cortex and cingulate gyrus are highly correlated with white matter integrity between two hemispheres (corpus callosum) and cerebro-spinal fluid. In addition, there are significant changes in the thalamus that shows extensive damages in this brain structure.