Accurate delineation of coronary arteries from X-ray angiography supports early identification of narrowing or blockages, but is complicated by low contrast and thin, branching vessel structures. This study compares three U-Net-based segmentation architectures — U-Net, U-Net++ and U-Net 3+ — trained on 1,000 images from the ARCADE dataset and evaluated on a held-out set of 300 images. All three architectures achieved comparable performance (Dice scores of 0.748-0.752), with U-Net 3+ performing best overall, U-Net++ showing higher sensitivity to smaller vessels, and U-Net showing higher precision, demonstrating the potential of U-Net-based models for coronary artery segmentation. This work was presented at the 5th Serbian International Conference on Applied Artificial Intelligence (SICAAI 2026), Kragujevac, Serbia, and was carried out within the STRATIFYHF project.