Graph Network Labels Arteries Across Three Anatomies
A new framework automates artery labelling in coronary, intracranial and abdominal scans, with F1-scores of up to 0.992.
Updated on Nov. 19, 2026 in Heart Disease

Researchers have proposed a graph neural network framework for automated artery labelling across coronary, intracranial and abdominal anatomies. Graph Convolutional Networks reached F1-scores of 0.895 and 0.870 on two independent coronary datasets.
Why it matters
Coronary artery disease is the leading cause of death worldwide, and manual artery labelling is labour-intensive. An automated method that adapts across anatomical domains could reduce that workload for clinicians and researchers.
Peer-reviewed F1-scores show Graph Convolutional Networks reaching 0.895 and 0.870 on two coronary datasets, and GraphSAGE reaching 0.967 and 0.992 on intracranial and abdominal arteries. The source does not establish clinical performance beyond these evaluations.
The details
The framework represents artery centrelines as graphs with anatomically informed node features, including hierarchical depth and segment orientation. It learns from anatomical connectivity rather than fixed geometric templates, which the researchers say makes it more adaptable than rule-based methods across imaging protocols.
Timeline
• Article published early as accepted peer-reviewed research on 19 November 2026.
Big Picture
The study extends manual artery labelling, the conventional method for mapping arterial trees, by automating it across coronary, intracranial and abdominal anatomies. It is described as the first to evaluate graph neural network artery labelling across multiple anatomical domains.
For patients, the practical effect of this research is not yet defined, since the study reports laboratory evaluation rather than a clinical tool. Any change to imaging analysis or diagnosis would depend on further validation before it reaches routine care in New South Wales.
The takeaway
Automated artery labelling that works across several body regions could shorten the time clinicians spend on manual mapping. Readers should note that these figures come from research evaluations and not yet from routine clinical use.
Further reading
For related coverage, see Heart Disease reporting across New South Wales.
Source note: This article includes information reported by Nature.






