Multi-Source Multi-View Graph Domain Adaptation with Hyperbolic Residual Encoding for Cross-Site MDD Identification from rs-fMRI

Researchers proposed a multi-source, multi-view graph domain adaptation framework for rs-fMRI-based major depressive disorder (MDD) identification.

The method combines Pearson correlation, sparse representation, and Granger causality graphs with graph attention networks and adaptive fusion.

It also uses hyperbolic residual encoding, alignment losses, and pseudo-labeling to reduce site-to-site distribution shifts.

Evaluation on seven unlabeled target domains showed improved cross-site classification performance.