A graph transformer-based foundation model for brain functional connectivity network.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Foundation models have advanced medical imaging, but their application in neuroimage analysis remains limited, hindering progress in neuroscience and clinical practice.
• Brain functional connectivity (FC) analysis is crucial for understanding brain function and is widely employed in neuroscience research.

Purpose of the Study:

• To propose a novel foundation model specifically designed for brain functional connectivity networks (FCN).
• To develop a versatile model capable of handling various neuroimaging analysis tasks, including classification, regression, and clustering.

Main Methods:

• A graph transformer model integrating node and edge embeddings for FCN feature extraction.
• Task-specific adapters enabling flexible adaptation to different analytical objectives.
• Validation on functional MRI (fMRI) data from 10,718 subjects across multiple tasks.

Main Results:

• The proposed foundation model consistently outperformed 14 competing methods in gender classification, mental disorder classification, brain age prediction, and disorder biotyping.
• The model demonstrated effectiveness in identifying task-specific FC patterns, facilitating biomarker discovery.
• Achieved superior performance across diverse neuroimaging analysis tasks.

Conclusions:

• The presented foundation model offers a novel and versatile solution for FCN analysis in neuroimaging.
• This model advances neuroimaging research by enabling scalable, interpretable, and high-performance analysis of brain connectivity.
• The findings highlight the potential of foundation models to drive innovation in understanding brain function and neurological disorders.