Coronary Artery Segmentation

Coronary artery segmentation identifies vessel lumens and branches in X-ray angiography, CT angiography, and intravascular images, enabling measurements and downstream tasks such as stenosis assessment and procedural planning. Common approaches build on U-Net-style CNNs, often adding attention, transformers, multi-scale features, or temporal information for thin, low-contrast vessels and angiography videos. Researchers also use topology-aware losses, centerline and surface metrics, synthetic data, and self-supervised pretraining to preserve small branches and improve performance when labels are scarce. A central challenge is that strong pixel-overlap scores can still hide broken or missing branches, while methods must also generalize across scanners, institutions, and imaging protocols.