Abstract Rationale Indeterminate pulmonary nodules remain one of the greatest diagnostic challenges in lung cancer evaluation, accounting for false positives and unnecessary invasive procedures. Artificial intelligence (AI) models may help address this uncertainty by refining human-based risk assessments. Bronchosolve, a fully automated closed-loop AI platform for chest computed tomography (CT) interpretation, has shown strong standalone performance in prior screening studies. However, its sequential integration with clinician assessment—particularly in moderate-risk cases—has not been evaluated. This study examined whether applying Bronchosolve to the ACCP-defined moderate-risk category (5-65% estimated malignancy probability) could enhance clinicians’ diagnostic accuracy by reclassifying these indeterminate cases into low or high risk. Methods We retrospectively analyzed 296 chest CT scans from 13 U.S. institutions (mean age 62.5 ± 7.1 years; 49.7% male), encompassing both screening and incidental nodules across five CT scanner manufacturers. All cases had specific benign or malignant diagnosis or clinically documented stability over a median of 6.3 years. Ten clinicians (5 radiologists and 5 pulmonologists) independently classified each case as Low, Moderate, or High risk following ACCP guidelines (5%, 5-65%, and 65%, respectively). In the subset of Moderate-risk nodules (n = 200; 67.6%), Bronchosolve was applied to reclassify these nodules as Low or High risk. Diagnostic performance (sensitivity, specificity, and area under the receiver operating characteristic curve AUC) was measured for Clinicians alone and Sequential (Clinicians → Bronchosolve) workflows. Results Across the full dataset, AUCs were 0.84 (CI: 0.83-0.86) for Clinicians, 0.87 (CI: 0.86-0.88) for Bronchosolve, and 0.88 (CI: 0.86-0.89) for the Combined setting (p 0.0001 for Clinicians vs Bronchosolve; p = 0.653 for Bronchosolve vs Combined). Following the sequential model of re-classifying Moderate Risk cases, applying Bronchosolve improved clinicians’ sensitivity from 72.1% (CI: 69.8-74.3%) to 82.4% (CI: 80.4-84.2%) and specificity from 61.9% (CI: 59.3-64.5%) to 72.4% (CI: 70.0-74.8%). This stepwise integration yielded a net diagnostic gain of + 10.3% in sensitivity and +10.5% in specificity, with overall accuracy increasing from 67.2% to 74.3% (Figure 1). Conclusions Sequential integration of Bronchosolve into clinical workflows significantly improved the sensitivity, specificity, and overall diagnostic accuracy of indeterminate moderate-risk pulmonary nodules. These findings support Bronchosolve’s role as a pragmatic, decision-augmenting tool in routine practice, streamlining management of moderate-risk pulmonary nodules where diagnostic uncertainty and variability are greatest. This abstract is funded by: Imavaria, Inc
Taha et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: