363 - A CT-Based Quantitative Study of Airway Dose-Toxicity Correlations after SABR for Central Lung Tumors

Purpose/Objective(s): Airway toxicity after stereotactic ablative radiotherapy (SABR) is related to the tumor proximity to proximal bronchial tree (PBT). Late onset of toxicity can lead to an under-reporting of events, making determination of tolerance doses difficult. For example, a systematic review identified only 37 cases of grades 3–5 toxicity from 9 studies on central tumors after SABR [1]. Tolerance doses for bronchial segments (BS) may vary per location after SABR and require branch-specific dosimetric analyses. Quantitative analysis of CT airway features (QCAF) allow for serial comparisons of BS anatomy and for earlier detection of damage. We employed an Artificial Intelligence (AI)-assisted airway segmentation approach to compare baseline and post-SABR QCAF, and combined this with a BS dosimetric analysis.

Materials/Methods: An ethics-approved database was used to identify patients with lesions located =2 cm from PBT. Breath-hold gated SABR was delivered using either 54Gy (in 3 fractions), 55Gy (5 fractions), or 60Gy (8 fractions). Study eligibility required the availability of pre- and post-SABR CT scans with slice thickness of =1 mm. BS diameters were analyzed using an AI-based airway segmentation method, and individual BS < 1.5cm of planning target volume were labeled (Boyden nomenclature). Stenosis was defined as a =50% diameter reduction from baseline; occlusion as a complete loss of visible lumen. Deformable image registration was used to warp planned doses to pre-SABR thin-slice CT scan. Primary study endpoints included dosimetric analysis of each bronchial structure, using EQD2 (a/ß=3) to assess the relationship between DVH parameters and stenosis or occlusion.

Results: In 29 patients analyzed (178 BS), stenosis or occlusion were detected in 93 branches representing Boyden airway generations 2–6. Median time to a full stenosis was 9.9 months (IQR: 7.1–14.6 months). Compared to open airways, stenotic segments had received higher doses. Dmean and D0.03cc showed the strongest association with stenosis: median Dmean (129.4 Gy, interquartile ranges [IQR]: 45.7 vs. open airways 50.7 Gy, IQR: 50.0; p < 0.001) and median D0.03cc (149.6 Gy, IQR: 42.3 vs. open airways 83.7 Gy, IQR: 72.8; p < 0.001). Statistically significant differences were observed between Dmean (p<0.001) and D0.03cc (p<0.001) for airway generations 2–3 and airway generations 4–6, respective groups, with smaller airways more likely to develop a bronchial occlusion.

Conclusion: AI-assisted airway segmentation combined with branch-specific dosimetric analysis identified sub-clinical bronchial dose-response relationships. Dmean and D0.03cc may serve as additional toxicity end-points but require confirmation in QCAF studies in larger patient cohorts.