Identifying influential nodes in brain networks via self-supervised graph-transformer.

Area of Science:

• Neuroscience
• Brain Imaging
• Graph Theory
• Deep Learning

Background:

• Identifying influential nodes (I-nodes) in brain networks is crucial for understanding brain function.
• Traditional methods rely on graph theory, potentially missing intrinsic network characteristics.
• Self-supervised deep learning offers a data-driven approach to explore I-nodes without manual feature engineering.

Purpose of the Study:

• To propose a novel framework for identifying influential nodes (I-nodes) in brain networks.
• To leverage self-supervised learning and Graph-Transformer for robust I-node detection.
• To explore the functional and structural significance of identified I-nodes.

Main Methods:

• Developed a Self-Supervised Graph Reconstruction framework based on Graph-Transformer (SSGR-GT).
• Utilized self-supervised learning to extract node importance for reconstruction.
• Employed Graph-Transformer for capturing local and global brain graph features.
• Integrated multimodal analysis using graph-based fusion of functional and structural brain data.

Main Results:

• Identified 56 influential nodes (I-nodes) in critical brain areas like the superior frontal and lateral parietal lobes.
• These I-nodes demonstrate greater involvement in brain networks, longer fiber connections, and central structural positions.
• Observed strong functional and structural connectivity, high node efficiency, and overlap with rich-club regions.

Conclusions:

• The proposed SSGR-GT method effectively identifies influential nodes in brain networks.
• The findings provide new insights into the role and characteristics of I-nodes.
• This research enhances the understanding of brain network mechanisms and offers avenues for future studies.