SACH-Net: Shape-Adaptive Convolution and Hierarchical Topology Constraints Framework for Coronary Artery Segmentation

Automatic segmentation of coronary artery is a crucial step in computer-aided diagnosis and treatment planning of coronary artery disease (CAD). A precise coronary artery mask aids clinicians in identifying potential stenosis and determining appropriate interventional treatment, signifying crucial medical importance in the efficient management of CAD. However, existing coronary segmentation methods encounter challenges, manifesting in complications like discontinuity of vessel mask and the mis-segmentation of small branches attributed to the intricate tree-like tubular structure of coronary artery. In this paper, we propose a novel coronary artery segmentation framework (called SACH-Net), which enhances segmentation effect by introducing shape-adaptive convolution (SA-Conv) and hierarchical topology constraints (HTC). Specifically, SA-Conv adjusts the convolution kernel adaptively based on the vessel shape to effectively learn the tree-like tubular feature representation, overcoming challenges posed by the intricate vascular structure. In addition, HTC module is introduced to supervise the feature expression of the network in three dimensions of continuity, overlap, and topological correctness, to alleviate the situation of segmentation fracture and discontinuity. The experimental results on the public dataset ARCADE show that SACH-Net significantly outperforms the state-of-the-art methods in coronary artery segmentation. The code is available at https://github.com/shbc2001/SACH-Net.Clinical Relevance—This research improves the accuracy of coronary segmentation and provides a more comprehensive evaluation, holding promising clinical implications for medical image analysis and healthcare applications.