Abstract Background The segmentation of contrast‐filled vessels from X‐ray coronary angiography (XCA) is a crucial step in the diagnosis and treatment of coronary artery disease. However, the accurate and complete extraction of vessels is particularly challenging due to XCA's image noise, motion artifacts, and complex vascular structures. Purpose To improve the accuracy and completeness of vessel segmentation, we propose an iterative joint learning network that integrates temporal and geometric information (ITG‐Net). Method The network leverages both temporal and spatial information from coronary artery image sequences to enhance vessel representation. Meanwhile, we combine the auxiliary task of centerline‐aware distance transform (CDT) with the primary task of vessel segmentation to guide the network in capturing the geometric connectivity of the vessels. Furthermore, the iterative learning strategy progressively refines the segmentation results through repeated bottom‐up and top‐down reasoning. Results We validated our method on the XCA dataset, achieving Dice of 85.67 ± 0.06%, centerline Dice (clDice) of 85.24 ± 0.19%, and average path length similarity (APLS) of 77.03 ± 0.48%. Compared with other advanced methods, our approach outperforms existing techniques in reducing vessel fragmentation and enhancing vessel continuity while better preserving vascular topology. Conclusion The results indicate that integrating temporal and geometric information into the iterative learning network can effectively enhance vessel segmentation in coronary angiography images.
Gao et al. (Sun,) studied this question.