Boundary-Emphasized Weight Maps for Distal Airway Segmentation
Ali Keshavarzi, Elsa Angelini
TL;DR
The paper tackles automated airway segmentation from chest CT by addressing leakage, breakage, and extreme class imbalance that impedes distal airway recovery. It introduces Boundary-Emphasized Loss (BEL), a boundary-aware loss with a distance-to-boundary weight and a differentiable soft-skeleton breakage map to emphasize edge voxels and preserve connectivity, formalized as $BELoss = 1 - \frac{\sum_i w_i p_i^r g_i}{\sum_i w_i (\alpha p_i + \beta g_i)}$ with $w_i = (1 - μ (d_i/d_{\max})^{γ})(1 + θ B_i)$, where $d_i$ is the distance to the boundary and $B_i$ derives from soft skeletons $S_{GT}$ and $S_P$. The approach is evaluated on ATM22 and AIIB23, showing superior topology-related metrics (e.g., Detected Length Rate, Detected Branch Rate) and improved small-branch segmentation, while maintaining competitive overlap metrics. BEL achieves more stable performance than centerline-based methods, demonstrating improved boundary preservation and reduced breakages, with qualitative results revealing better distal airway detail. The work offers a practical pathway to topology-enhancing airway segmentation, with future external validation and benchmarking against synthetic degradations suggested.
Abstract
Automated airway segmentation from lung CT scans is vital for diagnosing and monitoring pulmonary diseases. Despite advancements, challenges like leakage, breakage, and class imbalance persist, particularly in capturing small airways and preserving topology. We propose the Boundary-Emphasized Loss (BEL), which enhances boundary preservation using a boundary-based weight map and an adaptive weight refinement strategy. Unlike centerline-based approaches, BEL prioritizes boundary voxels to reduce misclassification, improve topology, and enhance structural consistency, especially on distal airway branches. Evaluated on ATM22 and AIIB23, BEL outperforms baseline loss functions, achieving higher topology-related metrics and comparable overall-based measures. Qualitative results further highlight BEL's ability to capture fine anatomical details and reduce segmentation errors, particularly in small airways. These findings establish BEL as a promising solution for accurate and topology-enhancing airway segmentation in medical imaging.