Risk Stratification of Incidental Pulmonary Nodules: A Fully Automated Approach to CT Imaging Analysis.

Background

Incidental detection is the most common pathway through which pulmonary nodules are identified. With the advancement of navigational and robotic-assisted bronchoscopy, existing risk stratification models often lack sufficient discrimination power, particularly for intermediate-risk nodules. Bronchosolve is a fully automated, imaging-only risk stratification tool that was previously validated in lung cancer screening cohorts.

Methods

We applied Bronchosolve to a cohort of adults with incidentally detected pulmonary nodules identified from registries at 2 large health care systems between 2006 and 2019. Nodules were classified as malignant or benign based on biopsy or extended clinical follow-up. Discrimination was assessed and compared with the VA Model. Sensitivity and specificity were reported at the Youden and high-sensitivity operating points. Additional analysis was performed in cases with nodule diameter 8 to 15 mm.

Results

Among 187 patients, malignancy prevalence was 70%. Bronchosolve achieved an AUC of 0.895 versus 0.870 for the VA model. Among nodules 8 to 15 mm, discrimination differences were larger (AUC 0.808 vs. 0.701). In nodules of 8 to 15 mm in diameter with an intermediate risk for malignancy by the VA model, Bronchosolve reclassified 51 patients to high risk and 23 to low risk; 33/37 (89%) malignancies were up-classified and 19/37 (51%) benign nodules were down-classified.

Conclusion

In a nonscreening setting, Bronchosolve demonstrated robust discrimination for lung nodule malignancy risk using imaging alone compared with a widely used clinical risk model, with more pronounced separation in intermediate-sized nodules. In this subgroup, Bronchosolve also yielded clinically meaningful risk reclassification, with potential implications for downstream management.