Total: 1
Accurate and complete segmentation of airways in chest CT images is essential for the quantitative assessment of lung diseases and the facilitation of pulmonary interventional procedures. Although deep learning has led to significant advancements in medical image segmentation, maintaining airway continuity remains particularly challenging. This difficulty arises primarily from the small and dispersed nature of airway structures, as well as class imbalance in CT scans. To address these challenges, we designed a Multi-scale Nested Residual U-Net (MNR-UNet), incorporating multi-scale inputs and Residual Multi-scale Modules (RMM) into a nested residual framework to enhance information flow, effectively capturing the intricate details of small airways and mitigating gradient vanishing. Building on this, we developed a three-stage segmentation pipeline to optimize the training of the MNR-UNet. The first two stages prioritize high accuracy and sensitivity, while the third stage focuses on repairing airway breakages to balance topological completeness and correctness. To further address class imbalance, we introduced a weighted Breakage-Aware Loss (wBAL) to heighten focus on challenging samples, penalizing breakages and thereby extending the length of the airway tree. Additionally, we proposed a hierarchical evaluation framework to offer more clinically meaningful analysis. Validation on both in-house and public datasets demonstrates that our approach achieves superior performance in detecting more accurate airway voxels and identifying additional branches, significantly improving airway topological completeness. The code will be released publicly following the publication of the paper.