The Landscape of Brain Artery Network: Database Foundation, Graph Analysis, and Biomarker Application

Abstract

The cerebrovascular network plays a critical role in brain health, with its structural and functional characteristics linked to various neurological and cardiovascular conditions. Despite this importance, current approaches to cerebrovascular analysis face significant limitations: they often rely on isolated morphological features that inadequately capture complex network properties, utilize tools with limited 3D manipulation capabilities, lack large-scale harmonized datasets, and employ analytical methods not optimized for vascular networks’ graph-like nature. These limitations hinder our understanding of how cerebrovascular architecture relates to neurological health and disease. This thesis establishes a comprehensive, reproducible framework for analyzing Brain Artery Networks (BANs) through graph-based approaches, integrating multimodal data to advance neurovascular health understanding across diverse populations. The framework consists of four major contributions. First, we created a harmonized, multi-site BAN dataset accompanied by demographic and clinical metadata, addressing data scarcity and heterogeneity through the development of VesselVoyager—an advanced 3D vessel annotation tool—and ComBat statistical harmonization. Second, we developed and improved advanced imaging sequences (iSNAP and CineMerge) that enable multi-contrast visualization and dynamic assessment of arterial pulsatility within a single acquisition, providing richer vascular characterization beyond static morphology. Third, we implemented specialized graph neu-ral networks tailored for vascular analysis, including a novel hierarchical graph transformer with edge-aware structural encoding that captures the intrinsic hierarchical organization of arterial networks. Fourth, we identified and validated clinically relevant biomarkers through systematic ablation studies. The impact of this work includes establishing a foundation for collaborative research through the Brain Artery Visualization & Analysis platform, enabling sophisticated quantitative analyses with lower barriers to entry. The long-term implications extend to precision medicine, where graph-derived vascular biomarkers may facilitate earlier disease detection, more accurate risk stratification, and personalized therapeutic monitoring in cerebrovascular diseases. Methodologically, our framework bridges imaging science and network theory, advancing analytical approaches that naturally represent the complex, interconnected nature of cerebrovascular systems. While limitations exist in demographic representativeness, imaging resolution, and causal inference capabilities, this thesis represents a significant step toward decoding the rich information contained in brain arterial networks, with the ultimate goal of improving outcomes for individuals affected by cerebrovascular conditions worldwide.