Signed graph representation learning for functional-to-structural brain network mapping.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Graph Theory

Background:

• Magnetic resonance imaging (MRI)-derived brain networks reveal functional and structural interactions.
• Graph mining of brain networks aids in discovering biomarkers for clinical phenotypes and neurodegenerative diseases.
• Integrating cross-modality brain networks (functional and structural) has significant clinical implications.

Purpose of the Study:

• To propose a novel graph learning framework, Deep Signed Brain Graph Mining (DSBGM).
• To learn cross-modality representations by projecting functional networks to structural counterparts.
• To address limitations of existing methods that map static structural networks to dynamic functional networks.

Main Methods:

• Developed a novel graph learning framework, DSBGM.
• Employed a signed graph encoder for projecting functional networks to structural counterparts.
• Validated the framework on clinical phenotype and neurodegenerative disease prediction tasks.

Main Results:

• DSBGM demonstrated superior performance in clinical phenotype and neurodegenerative disease prediction.
• Experimental results showed clear advantages over several state-of-the-art methods.
• The framework effectively learns cross-modality representations from functional and structural brain networks.

Conclusions:

• DSBGM offers a novel approach for integrating functional and structural brain networks.
• The proposed method shows promise for advancing biomarker discovery in neurodegenerative diseases.
• This framework provides a new perspective for analyzing signed graphs in brain network research.