Learning from Brain Data for Neurological Disorder Analysis

In recent years, the advancement in neuroimaging technology has given rise to various modalities of brain imaging data, which provides us with unprecedented opportunities for investigating the inner organization and activity of human brain for neurological disorder analysis. These brain data can be acquired in different forms, such as the spatio-temporal tensor data (e.g., fMRI 4D tensor image), graph data (e.g., fMRI brain connectivity networks) and multi-view graph data (e.g., fMRI and DTI brain networks). Learning from these brain data and leveraging the information for neurological disorder analysis can potentially facilitate the clinical investigation and therapeutic intervention of many brain diseases. In this dissertation, I introduce our recent works on modeling and learning from brain data in multiple perspectives for neurological disorder analysis. In the first part, I focus on the spatio-temporal tensor modeling of fMRI image data for whole-brain classification. Then I present an approach based on interior-node topology of graphs for the clustering of brain networks. Furthermore, a multi-view clustering framework is proposed with graph embedding for the clustering of multi-view brain networks. Finally, I introduce a multi-view graph embedding approach for jointly learning the multi-view representation and detecting hubs from multi-view brain networks.