Deep Representation Learning For Multimodal Brain Networks.

Area of Science:

• Neuroscience
• Medical Imaging Analysis
• Network Science

Background:

• Investigating human brain function and anatomy using network science is common in medical imaging.
• Analyzing multimodal brain networks presents challenges due to complex topology and non-linear cross-modality relationships.
• Existing deep learning methods may overlook graph topology or require group-shared network bases.

Purpose of the Study:

• To propose a novel end-to-end deep graph representation learning framework, Deep Multimodal Brain Networks (DMBN).
• To fuse multimodal brain networks by deciphering cross-modality relationships through graph encoding and decoding.
• To learn higher-order network mappings from structural to functional brain networks.

Main Methods:

• Developed a novel deep graph representation learning framework (DMBN).
• Employed a graph encoding and decoding process to model cross-modality relationships.
• Learned node-level mappings from structural to functional brain networks, generating informative node features for saliency maps.

Main Results:

• The DMBN framework successfully fuses multimodal brain networks.
• The method effectively learns higher-order network mappings in the node domain.
• Experimental results on synthetic and real data demonstrate superior performance compared to state-of-the-art models.

Conclusions:

• DMBN offers a powerful approach for integrating multimodal brain network data.
• The learned node features are informative for inducing brain saliency maps.
• This framework advances deep learning applications in brain network analysis.