MULTIMODAL IMAGING STUDIES OF EARLY STAGES AND PERSISTENT SCHIZOPHRENIA

40 subjects. The method consists of four stages: (1) a support vector machine (SVM) based classifier ensemble for SNPs, (2) a SVM classifier ensemble forvoxels in fMRI, (3) anSVMclassifier for independent factors of fMRI activation, and (4) an integrated SNP-fMRI classifier. Results: The classification accuracy obtained was: 0.74 for the SNP only classifier; 0.82 for the fMRI voxel only classifier, 0.85 for the fMRI factor classifier and 0.90 for the combined SNP-fMRI classifier. The most discriminating SNP loci identified by classifiers locate in genes of SELP, COMT, GAD2, HTR3B, DISC1, CYP2C19, and etc. The brain regions contributing most to classification consist of inferior, middle andmedial frontal gyri, cingulate gyrus, superior temporal gyrus and precuneus. Discussion: Some SNPs selected in the study have previously been shown to be schizophrenia related, such as COMT, DISC1, and HTR3B. The brain regions identified including large portions of frontal lobe and superior temporal lobe, are also consistent with previous literature showing functional differences between schizophrenia and healthy subjects. In summary, experimental results show that the proposed method achieves better classification accuracy by combining genetic data and fMRI data than using either of them. Results suggest an effective way in identifying schizophrenic individuals from healthy control groups which may prove useful for assisting in the diagnosis and treatment of schizophrenia.