Functional brain network analysis based on unsupervised deep learning

In the neuroimaging and brain mapping communities, researchers have proposed a variety of computational methods and tools to learn functional brain network (FBN), such as general linear models (GLM), independent component analysis (ICA) and sparse dictionary learning (SDL). Recently, deep learning has attracted much attention in the fields of machine learning and data mining, and it has been proven that deep learning approach has superb representation power over traditional shallow models. In this research, three deep models, which are volumetric sparse deep belief networks (VS-DBN), neural architecture search based DBN (NAS-DBN) and recurrent autoencoder (RAE), were designed to explore representations of fMRI volumes. The quantitative analysis showed that these deep models have promising capability in learning meaningful FBNs and revealed novel insights into the organizational architecture of human brain.