Double Collaborative Learning on Functional Brain Networks for Brain Disease Classification

Stationary and dynamic functional brain networks (FBNs) characterize the complex interactions of the human brain from different insights, which may offer complementary information for brain disease analysis. However, most studies only focus on one type of FBNs, thus limiting the performance of characterization. Moreover, multi-view (or multi-modal) collaborative contrastive learning has shown great success in many fields, but few work try to apply it to FBN analysis. To this end, we propose a Double Collaborative Learning Network (DCLNet), which leverages both collaborative encoder and collaborative contrastive learning to jointly learn the representations and the inter- and intra-class relationships of both stationary and dynamic FBNs, for brain disease classification. We evaluate the DCLNet on the ADNI and ADHD-200 datasets, the experimental results demonstrate the superiority of our method.