Session Item

Stereotactic ablative radiotherapy (SABR) for central lung tumors is associated with increased toxicity, including hemorrhage, airway stenosis and atelectasis. As toxicity can manifest months to years after SABR, serial assessment of follow-up CT scans may identify subclinical toxicity. As manual segmentation and evaluation of thoracic airways is time-consuming, we investigated the use of semi‑automatic algorithms (SAA) to quantify post SABR changes in airways.

An Ethics-approved imaging database of patients who completed central lung SABR on a CT-linac was accessed. For this exploratory study, 10 patients with available pre- and post SABR high-resolution CT scans (defined as HR-CT <1mm resolution) were identified. Images from 1 patient were used to develop the SAA workflow, which was then applied in 9 other patients. A publicly available SAA implemented in the 3D Slicer Chest Imaging Platform (CIP) was used to contour tracheal and bronchial lumens on each HR-CT (Fig.1 left panels). A manually placed seed point in the airway is used as the starting point, and the threshold (HU) is iteratively increased until leakage occurs. A centerline curve through the airway lumen, including any branch terminate point (Fig.1 middle panels), is then extracted using the Vascular Modeling Toolkit (VMTK) in 3D Slicer. Cross-sectional metrics are calculated along the centerline curve (Fig.1 right panels). Airway segments were classified using the Boyden nomenclature. End-results were reviewed and verified by a clinician. Pre- and post SABR comparisons of cross-sectional airway metrics were performed.

In the pilot case, the SAA was able to contour the airway structure from the trachea to the segmental bronchi. A total of 34 HR-CT scans were evaluated in this study. Minor manual interaction of post SABR scans was required in 12 HR-CT after use of SAA to include missed segmental bronchi. The time required to complete airway lumen contours on a single HR scan averaged 15 min (range: 7-36 min). Estimated manual segmentation times by a clinician for a typical case was estimated to be 50 min. Cross-sectional centerline analysis detected significant narrowing of a lobar bronchus in 5 patients, and a complete obstruction of ≧1 segmental bronchi in 7 cases (Tab. 1). In the latter group, SAA detected airway narrowing preceding complete obstruction in 5 cases.

This semi-automated segmentation approach enabled fast quantification of SABR-induced airway stenosis, with limited manual interaction required typically for changes in segmental bronchi. These findings suggest that early airway stenosis may serve as an additional toxicity endpoint in prospective trials, and that this approach may minimize interobserver variations in scoring stenosis.