Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia

• DOI: 10.1038/srep38897
• Publication Date: 2016-12-12T11:00:45Z
• Authors: Walter H. L. Pinaya; Ary Gadelha; Orla M. Doyle; Cristiano Noto; André Zugman; Quirino Cordeiro; Andrea P. Jackowski; Rodrigo A. Bressan; João R. Sato
• Keywords: Putamen; Deep belief network; Brain morphometry

Neuroimaging-based models contribute to increasing our understanding of schizophrenia pathophysiology and can reveal the underlying characteristics this other clinical conditions. However, considerable variability in reported neuroimaging results mirrors heterogeneity disorder. Machine learning methods capable representing invariant features could circumvent problem. In structural MRI study, we trained a deep model known as belief network (DBN) extract from brain morphometry data investigated its performance discriminating between healthy controls (N = 83) patients with 143). We further analysed classifying first-episode psychosis 32). The DBN highlighted differences classes, especially frontal, temporal, parietal, insular cortices, some subcortical regions, including corpus callosum, putamen, cerebellum. was slightly more accurate classifier (accuracy 73.6%) than support vector machine 68.1%). Finally, error rate 56.3%, indicating that representations learned were not suitable define these patients. Our suggest improve psychiatric disorders such by improving neuromorphometric analyses.