Interpretable modality-specific and interactive graph convolutional network on brain functional and structural...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Image Analysis

Background:

• Brain functional connectivity (FC) and structural connectivity (SC) are crucial for cognition and neurological diseases.
• Interactions between SC and FC in association regions are linked to cognitive alterations and diseases.
• Existing methods lack the ability to fully leverage both modality-specific features and high-order interactions between SC and FC.

Purpose of the Study:

• To propose an interpretable graph convolutional network (MS-Inter-GCN) that integrates modality-specific information and structure-function interactions.
• To develop a novel framework capable of leveraging both unique neural mechanisms of each modality and the underlying structural basis of brain function.
• To enhance regression and classification tasks by effectively modeling the interplay between structural and functional brain connectivity.

Main Methods:

• Developed a modality-specific and interactive graph convolutional network (MS-Inter-GCN) using graph convolutional encoder-decoder modules.
• Generated modality-specific, task-relevant embeddings and learned interactive weights between corresponding regions of FC and SC.
• Constructed a novel graph structure incorporating embeddings and interactive weights, utilizing GNNExplainer for post-hoc analysis and identification of salient regions and interactions.

Main Results:

• The MS-Inter-GCN framework outperformed ten state-of-the-art methods on multi-modal brain features for fluid cognition prediction and PD, AD, and SZ classification.
• GNNExplainer successfully identified salient structural and functional regions and their interactions related to fluid cognition and the studied diseases.
• Demonstrated that strong structure-function coupling in executive/control networks and weak coupling in motor networks are associated with fluid cognition.

Conclusions:

• Structure-function decoupling in specific brain regions serves as a potential biomarker for Parkinson's disease, Alzheimer's disease, and schizophrenia.
• MS-Inter-GCN provides an effective and interpretable approach for analyzing multi-modal brain connectivity data.
• The findings highlight the importance of integrating structural and functional brain information for understanding cognition and neurological disorders.