Automatic Coronary Artery Segmentation in X-ray Angiograms

Coronary artery disease is one of the leading causes of morbidity and mortality worldwide, with X-ray coronary angiography serving as the clinical gold standard for diagnosis and intervention planning. Accurate segmentation of coronary arteries is essential for quantitative analysis, assessment of stenosis severity, and reliable clinical decision-making. In current clinical practice, coronary artery evaluation largely depends on manual or visual interpretation by cardiologists. This process is time-consuming, subjective, and affected by inter-observer variability, particularly in angiograms characterized by low contrast, image noise, and complex vascular structures. To overcome these challenges, automated segmentation methods based on deep learning have emerged as a promising solution. In this study, an automated pipeline for coronary artery segmentation in X-ray angiograms is proposed and evaluated using four deep learning architectures: U-Net, U-Net++, U-Net3+, and nnU-Net. A unified experimental framework incorporating standardized data organization, image preprocessing, and k-fold cross-validation is employed. Model performance is assessed using Dice coefficient, Intersection over Union, precision, and recall. The results show that U-Net++ achieved the highest segmentation accuracy with a Dice score of 0.724, compared to 0.671 for U-Net, 0.716 for U-Net3+, and 0.712 for nnU-Net. These findings confirm that modern U-Net-based architectures can provide reliable and reproducible coronary artery segmentation under consistent evaluation conditions.