<p>Accurate segmentation of coronary arteries is crucial for the diagnosis and monitoring of cardiovascular diseases. To effectively segment coronary arteries and leverage their inherent properties, we propose tubular-aware network (TANet), a network specifically designed for coronary artery segmentation. TANet is designed to leverage the inherent properties of coronary arteries: their macro-level single-connectivity as a tree structure, and the tubular shape of both thick and thin vessels. TANet incorporates a position information propagation (PIP) mechanism, aiming to selectively propagate positional information. In conjunction with a single-connectivity selection (SCS) module, it selects the vascular region from multiple large, vessel-like areas, thereby reducing the impact of substantial non-vascular elements. Finally, a tubular alignment (TA) module employs multi-thickness features fusion to thickness-aligned vascular features, attempting to maintain vascular morphological accuracy while accommodating richer vascular morphology, which further guides the model’s final output. Experimental results on three public, de-identified datasets demonstrate that TANet outperforms most existing methods in improving segmentation accuracy.</p>