Graph Neural Networks in Brain Connectivity Studies: Methods, Challenges, and Future Directions.

Abstract

Brain connectivity analysis plays a crucial role in unraveling the complex network dynamics of the human brain, providing insights into cognitive functions, behaviors, and neurological disorders. Traditional graph-theoretical methods, while foundational, often fall short in capturing the high-dimensional and dynamic nature of brain connectivity. Graph Neural Networks (GNNs) have recently emerged as a powerful approach for this purpose, with the potential to improve diagnostics, prognostics, and personalized interventions. This review examines recent studies leveraging GNNs in brain connectivity analysis, focusing on key methodological advancements in multimodal data integration, dynamic connectivity, and interpretability across various imaging modalities, including fMRI, MRI, DTI, PET, and EEG. Findings reveal that GNNs excel in modeling complex, non-linear connectivity patterns and enable the integration of multiple neuroimaging modalities to provide richer insights into both healthy and pathological brain networks. However, challenges remain, particularly in interpretability, data scarcity, and multimodal integration, limiting the full clinical utility of GNNs. Addressing these limitations through enhanced interpretability, optimized multimodal techniques, and expanded labeled datasets is crucial to fully harness the potential of GNNs for neuroscience research and clinical applications.