Work

A unified graph neural network framework for artery labelling across multiple vascular territories

Year: 2026
Type: article
Abstract: Vascular disease is a major public health challenge, with coronary artery disease the leading cause of death worldwide. Improved diagnostics through vascular imaging, including computed tomography coronary angiography, depend on accurate anatomical labelling of arteries. However, manual labelling is labour-intensive because arterial trees are complex and variable. Traditional rule-based methods lack adaptability across protocols, while convolutional neural networks often fail to capture the spatial and topological relationships required for accurate labelling. We propose an adaptable graph neural network (GNN) framework for automated artery labelling across coronary, intracranial, and abdominal anatomies. Artery centrelines are represented as graphs, with anatomically informed node features, including hierarchical depth and segment orientation, capturing structural patterns shared across vascular regions. By learning from anatomical connectivity rather than fixed geometric templates, the framework reduces reliance on manual feature engineering and supports cross-domain application. To our knowledge, this is the first study to evaluate GNN-based artery labelling across multiple anatomical domains. Graph Convolutional Networks achieved F1-scores of 0.895 and 0.870 on two independent coronary datasets, while GraphSAGE performed best for intracranial and abdominal arteries, achieving F1-scores of 0.967 and 0.992. These results demonstrate a practical, adaptable, and generalisable approach to automated artery labelling in medical image analysis.
Institution: UNSW Sydney
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236743067
FWCI:
Cites: 45
Cited by:
Related to: 10