A multi-modal fusion of features method based on deep belief networks to diagnosis schizophrenia disease

Abstract

Computer-based diagnostics of schizophrenia (SZ) have been recently the subject of intense interest. Some recent studies have shown promising results in the diagnosis of this disorder by the process and analysis of neuroimaging data, such as functional magnetic resonance imaging (fMRI), structural magnetic resonance imaging (sMRI) and diffusion tensor imaging (DTI). Each of these neuroimaging data provides different brain structural and functional features that integrating them can help accurately diagnose brain disorders. In this study, we propose a deep learning architecture to combine functional brain data with two different types of structural brain data to accurately diagnose SZ. Our two-step model consists of deep belief networks (DBN) and a fusing rule to extract and learn hidden features via the combining resting-state fMRI with both sMRI and DTI data. We examine this architecture on all combinations of features extracted from fMRI, MRI and DTI. The experimental results indicate that multi-modal fusion improved classification accuracy between the SZ and the healthy cases. By finding a suitable configuration for algorithm parameters, the proposed method showed an accuracy level of 99.36% for the COBRE dataset and outperformed similar state-of-the-art works.