Morphology-Integrated Segmentation for Coronary Interventional Imaging

Robotic-assisted percutaneous coronary intervention (R-PCI) represents a significant technological advancement in the field of cardiovascular intervention. The combined segmentation of guidewires and vessels serves as the perceptual foundation for R-PCI systems, providing critical anatomical and instrumental positioning data to the robotic platform. Coronary interventional imaging relies on two core modalities: (1)2D DSA (Digital Subtraction Angiography) acquired during contrast opacification for precise vessel mapping, and (2)2D X-ray fluoroscopy captured after contrast clearance for real-time instrument tracking. However, both coronary vessels and guidewires typically exhibit slender, elongated linear structures that are susceptible to imaging challenges such as noise, low contrast, motion artifacts, and complex anatomical backgrounds, severely hindering segmentation accuracy and structural continuity. To address these challenges, we propose a coronary intervention image segmentation method with integrated morphological features, specifically two task-specific deep neural network architectures. For guidewire segmentation, we introduce GS-UNet (Guidewire Segmentation U-Net), an enhanced U-Net-based model tailored for detecting fine linear structures. GS-UNet incorporates a line-sensitive enhancement module and a guided feature augmentation mechanism, significantly improving the continuity and completeness of guidewire segmentation. The model is extensively validated on a self-constructed guidewire dataset, achieving superior performance in maintaining connectivity and accurately identifying guidewire endpoints. Building upon the architectural insights from GS-UNet, we further propose MSFNet (Multi-Scale Feature Fusion Network) for coronary vessel segmentation. MSFNet integrates multi-scale feature fusion modules and a composite attention mechanism to capture local fine details and global vascular topology jointly. Moreover, it enhances robustness against modality variation and complex backgrounds. Extensive experiments conducted on the authoritative ARCADE dataset demonstrate that MSFNet consistently outperforms state-of-the-art methods in terms of vessel continuity, boundary precision, and detection of fine vascular branches, highlighting its strong potential for real-world clinical applications. The morphology-integrated segmentation framework proposed in this study effectively enhances guidewire-vessel co-segmentation precision, enabling more precise anatomical localization for R-PCI navigation in complex coronary interventions.