DEEP MULTIMODAL BRAIN NETWORK LEARNING FOR JOINT ANALYSIS OF STRUCTURAL MORPHOMETRY AND FUNCTIONAL CONNECTIVITY.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Multimodal brain imaging data offers complementary insights beyond single modalities.
• Integrating diverse data types (e.g., morphometry, functional connectivity) presents significant challenges due to variations in signals and structure.
• Advanced analytical methods are needed to effectively unify knowledge from heterogeneous brain imaging sources.

Purpose of the Study:

• To develop a supervised deep learning model for joint analysis of brain morphometry and functional connectivity.
• To address the challenge of unifying image-based knowledge from multimodal brain imaging data.
• To create a robust framework for brain network analysis and classification tasks.

Main Methods:

• Proposed a deep multimodal brain network learning (DMBNL) model.
• Introduced two graph-based kernels: geometry-aware surface kernel (GSK) for morphometry and topology-aware network kernel (TNK) for functional connectivity.
• Integrated vertex features from GSK into TNK and computed graph-level features for classification.

Main Results:

• The DMBNL model successfully integrated multimodal brain imaging data.
• The proposed GSK and TNK effectively processed cortical surface morphometry and functional networks.
• The model achieved effective classification performance on a large autism imaging dataset.

Conclusions:

• The developed DMBNL model demonstrates effectiveness in jointly analyzing multimodal brain imaging data.
• The approach provides a powerful tool for understanding complex brain networks and their alterations in conditions like autism.
• This study highlights the potential of deep learning for advancing multimodal brain network analysis.