A03Development of computational assay for psychiatric disorders using deep learning
In this research project, we investigate a novel technique for analyzing resting state functional magnetic resonance imaging (rsfMRI) data using deep neural networks. rsfMRI is a functional brain imaging method for evaluating regional interactions (functional connectivity: FC) that occur when a subject is not performing an explicit task. Resting state FC analysis provides useful information corresponding to the functional organization of individuals’ brain in healthy subjects, neurological or psychiatric disorder patients. However, even though rsfMRI includes a lot of temporal information, analyzing methods for temporal dynamics of rsfMRI remains unestablished yet. In order to address this issue, in the current project, we investigate a novel technique for extracting self-organized features with spatio-temporal hierarchy from rsfMRI data using deep neural networks. We expect that these extracted features could reflect individual’s brain states which are useful for the evaluation and treatment of psychiatric disorders. Particularly, we focus on the dimensional approach for psychiatric disorders, which attempts to find biological basis corresponding to observable behavioral and symptom measures, regardless of conventional categories of psychiatric disorder. The dimensional approach for psychiatric disorder is consistent with the idea that individual brain states can be evaluated as continuous feature-spaces from normal states to abnormal states. The proposed technique could be one of the base technologies of examination and evaluation for psychiatric disorders based on computational psychiatry (computational assay).