multimodal-brain-connectivity-gnn

Multimodal brain data integration uses modality-specific encoders and graph attention layers to fuse fMRI, DTI, and sMRI into graph-level cognitive predictions. The model aligns these modalities using Glasser atlas parcellation across 360 cortical regions.

Explainability for brain connectivity uses learnable edge masks to weight connections and quantify modality importance. This reveals key structural and functional biomarkers by highlighting important graph connections and modality contributions.

Multimodal GNNs support disease profiling and biomarker discovery by mapping brain connectivity to cognitive outcomes. The explainability module identifies important connections and modality weights, providing clinical insights for neurological research.

Yes, fMRI and DTI data need alignment using the Glasser atlas for 360 cortical regions. This atlas-based parcellation ensures structural and functional connectivity matrices match across modalities for graph neural network processing.

The best way to weight multimodal connectivity is using learnable edge masks that adjust connection importance based on modality differences. This produces interpretable graph attention network results for biomarker discovery.

Global pooling aggregates region-level features from GAT and GCN layers into a single graph-level representation. This allows the multimodal graph neural network to output cognitive score estimates and disease profiles.