Characterization of schizophrenia by linear kernel canonical correlation analysis of resting-state functional MRI and structural MRI

In almost every mental disorder, there are deficiencies in both structure and function of the brain. So the need for analyzing complementary modalities that project all aspects of the brain is rising. The most severe kind of these disorders is schizophrenia. The main cause of schizophrenia is still unknown. Therefore, analyzing resting-state fMRI (rs-fMRI) and structural MRI (sMRI) to investigate the differences between schizophrenia and healthy control subjects is going to be helpful. For this aim, we used linear kernel canonical correlation analysis (L-kCCA). We extracted gray matter volume and amplitude of low frequency fluctuation (ALFF) as features for sMRI and rs-fMRI respectively. In this method we applied CCA to much lower dimension data compared to real one. In other words, we applied CCA to similarity matrices which were representative of the correlation of voxel values between subjects. So, the time and the need for memory are reduced. In addition to inter-subject variations, this method allows us to extract the regions which are associated to the subjects' variation in the two modalities. The method was applied to the images of 11 schizophrenia and 11 matched healthy control subjects which were acquired in Imam Khomeini hospital, Tehran, Iran. Based on the results, we can observe gray matter volume reduction in schizophrenia in precuneus, temporal and frontal regions. In frontal, temporal and occipital regions the ALFF is higher in healthy control subjects than schizophrenia and in precentral and right and left insula regions brain activity at rest is lower than patients.