Automated Topology-Aware Lung Airway Labeling
Abstract
Purpose
A key barrier to clinical efficacy of functional lung sparing radiation therapy (RT) is that current functional-sparing strategies fail to account for airway paths and connected lung parenchymal sub-volumes in conjunction – an approach that, we previously showed, spared over 50% more post-RT lung function than the usual strategy of mapping and sparing high-functioning lung volumes only. We report on an important step toward the clinical implementation of our approach - automated individual airway labeling using CT-derived virtual bronchoscopy images.
Methods
An automated graph-based topological labeling algorithm was developed through bifurcation detections and traversing for individual airway segmentation. In-house modules for estimating airway diameters and tube length were also developed. Both labeling and meta-information estimation were validated retrospectively against commercial virtual bronchoscopy in 20 lung cancer patients. Labeling metrics were segmentation surface scores: average symmetric surface distance (ASSD), normalized surface distance (NSD), and 95th percentile Hausdorff (HD95surf). Meta-information metrics were absolute errors (AEs) of branch major diameter, minor diameter, and tube length. Outliers and airway-generation-based errors were analyzed on all branches from the dataset. Final root-cause analysis was implemented to identify sources of inaccuracy in our results.
Results
Means (standard deviations) were 0.56(0.21) mm for ASSD, 87.66(3.20) % for NSD, and 2.51(0.68) mm for HD95surf. MAEs/RMSEs for meta-information metrics in labeling validation were: 0.33(0.09) / 0.67(0.16) mm for major diameter, 0.24(0.07) / 0.53(0.15) mm for minor diameter, and 3.27(0.82) / 5.53(1.53) mm for tube length. MAEs/RMSEs in meta-information validation were: 1.81(0.30) / 2.31(0.41) mm for major diameter, 1.25 (0.27) / 1.14 (0.26) mm for minor diameter, and 2.20(0.32) / 3.24(0.49) mm for tube length.
Conclusion
The proposed open-source framework provides stand-alone, fully automated airway labeling that can be used with any commercial or open-source virtual bronchoscopy software to import the airway tree as an organ-at-risk into a planning system.