Background: The impact of artificial intelligence (AI) tools for lung nodule evaluation on low-dose CT (LDCT) have been evaluated primarily using experimental retrospective reader designs, outside of clinical practice. Objective: To compare interpretation times and lung nodule detection rates between real-world interpretations conducted with versus without an AI-based lung nodule evaluation tool for LDCT examinations in asymptomatic individuals. Methods: This prospective single-center, parallel, open-label clinical trial included consecutive individuals who underwent LDCT of the chest during a self-initiated general health checkup from May 19, 2025 to September 4, 2025. Individuals underwent 1:1 random allocation to an intervention (AI-assisted interpretation) or control (interpretation without AI) group. In the intervention group, a commercial AI tool automatically detected, classified, and measured nodules, and the results were displayed within PACS. Examinations were interpreted by one of ten thoracic radiologists, who reported nodules only when measuring ≥4 mm. The primary outcome was interpretation time per examination. Secondary outcomes included the detection rates (proportion of examinations reporting a finding) of Lung-RADS-positive (category 3 or 4) nodules and of all nodules and the frequency of recommendations for follow-up LDCT. Subsequent lung cancer diagnoses were recorded. Results: The final analysis included 911 individuals (517 men, 394 women; mean age, 62 years; 447 and 464 in the intervention and control groups, respectively). The intervention group, in comparison with the control group, showed no significant difference in interpretation time per examination (187 vs 172 seconds, respectively; P=.23), but a significantly higher detection rate of Lung-RADS-positive nodules (16.9% vs 10.3%, respectively; P=.03), detection rate of all nodules (52.9% vs 32.6%, respectively; P=.002), and frequency of follow-up LDCT recommendations (15.3% vs 7.4%, respectively; P=.04). No individual in either group was diagnosed with lung cancer (median follow-up of 215 and 216 days in intervention and control groups, respectively). Conclusion: Use of an AI-based nodule evaluation tool integrated into PACS during real-world clinical workflows was not associated with a significant difference in interpretation times. However, the tool was associated with significantly greater detection of clinically actionable nodules. Clinical Impact: This randomized clinical trial provides pragmatic evidence regarding the implications of AI-assisted LDCT interpretation.
Hwang et al. (2026) studied this question.