Exploratory factor analysis of MRI brain structure measures in schizophrenia

Much of the literature shows various regional structural brain abnormalities in schizophrenia, but the complexity and variability of brain makes it difficult to determine how these regions are related. Statistical methods which estimate factors underlying patterns of covariance have not been widely used, but could be useful for analyzing such complex data. We applied exploratory and confirmatory factor analysis procedures to specific cortical and subcortical regional brain volume measures from MRI data in 60 normal and 44 schizophrenic subjects. Basal ganglia, heteromodal cortical gray, and medial temporal lobe factors were present in both the normal and the schizophrenia groups. The factor structure observed in the normal group showed a high degree of bilateral symmetry which is present but disrupted in the schizophrenia group. In the bilateral data, the disruption is most pronounced with medial and lateral temporal lobe structures including entorhinal cortex and anterior and posterior superior temporal gyri. There was a significant correlation between the basal ganglia factor and the heteromodal cortical gray factor in the normal group that was not present in the schizophrenia group. In the unilateral data, left posterior superior temporal gyrus did not load onto any factor in the schizophrenia group. Confirmatory factor analyses showed significant differences between the two groups in factor structure. A number of specific brain regions are affected in schizophrenia, and structural relationships between groups of regions also are abnormal. The results suggest that heteromodal dorsolateral prefrontal and superior temporal cortical gray regions are structurally related, whereas inferior parietal cortical gray is less so. These results should be viewed as preliminary as the ratio of parameters to subjects was relatively low, and replication is needed. However, the results demonstrate the potential utility of latent structure methods such as factor analysis in study of complex relationships in neuropsychiatric data.