Abstract Rationale Malignancy risk stratification of pulmonary nodules is limited by interobserver variability in subjective image assessment. Bronchosolve is a closed-loop, fully automated, DICOM-only tool that risk-classifies pulmonary nodules without manual image manipulation. The tool executes a fully automated pipeline: optimal image selection, normalization/preprocessing, 3D nodule detection, risk scoring, and report generation. Its efficacy in distinguishing malignant from benign nodules has been previously validated in lung cancer screening cohorts and shown to be comparable to Lung-RADS and the Brock model. In this study, we evaluated Bronchosolve’s performance for incidentally detected nodules on CT scans performed for indications other than lung cancer screening. Methods We applied Bronchosolve to a series of 187 cases with ≥1 pulmonary nodule from two US sites. Inputs were DICOM images only (no demographic or clinical variables). Nodules were classified as malignant or benign based on biopsy or extended clinical follow-up. The model outputs a binary label (likely malignant vs likely benign) and a continuous score (0-100). Discrimination was assessed using the area under the receiver operating characteristic curve (AUC) and compared with the VA Lung Cancer Risk Model (VA model). Additional analysis was performed in cases with nodule diameter between 8 and 15 mm, when clinical decision-making is often challenging. Results Among 187 patients with incidentally detected nodules, median age was 67 years. The median nodule count was 1 with a median diameter of 15 mm. 88.2% were current or former smokers. The malignancy prevalence was 70% (131/187). Overall, Bronchosolve achieved an AUC of 0.895, exceeding the VA model’s AUC of 0.870. Performance differences were more pronounced for nodules 8-15 mm: Bronchosolve AUC 0.808 (95% CI 0.705-0.902) vs VA model AUC 0.701 (95% CI 0.578-0.813), as shown in Graph 1. Conclusions In a non-screening clinical setting, Bronchosolve demonstrated robust discrimination for lung nodule malignancy risk using imaging alone and outperformed a widely used clinical risk model, with the even greater gains in intermediate-sized nodules (8-15 mm). These findings support Bronchosolve’s potential as a clinical decision-support tool for evaluating incidentally detected pulmonary nodules beyond screening populations. This abstract is funded by: None
Liu et al. (Fri,) studied this question.