Reproduce “Learning Dynamic Graph Representation of Brain Connectome with Spatio-Temporal Attention” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Byung-Hoon Kim, Jong Chul Ye, Jae-Jin Kim
Researchers from Yonsei University and KAIST developed the Spatio-Temporal Attention Graph Isomorphism Network (STAGIN), a method for learning dynamic graph representations of brain connectomes that integrates spatio-temporal attention. STAGIN achieved 88.20% accuracy for gender classification and 99.19% for task decoding on HCP fMRI datasets, outperforming prior GNN methods while providing interpretable attention maps that align with neuroscientific findings.
Key insights
- Effectively integrating dynamic functional connectivity with spatio-temporal attention in GNNs significantly improves predictive performance and offers interpretable insights into brain activity.
- The model's attention mechanisms provide neuroscientifically consistent explanations, such as identifying gender-specific dynamic FC patterns (e.g., DMN/SMN differences) and revealing unsupervised task-specific temporal dynamics.
- Spatial attention across GNN layers can expose a cognitive processing hierarchy during tasks, with lower layers attending to sensorimotor functions and higher layers engaging cognitive control and default mode networks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.