Use of artificial intelligence during low-dose chest computed tomography interpretation was associated with increased detection of lung nodules but no significant reduction in interpretation time in a randomized clinical trial.
In the prospective trial, researchers evaluated 911 asymptomatic individuals undergoing low-dose chest computed tomography (LDCT) as part of self-initiated routine health checkups in South Korea. Patients were randomly assigned to interpretation with an artificial intelligence (AI)–based lung nodule evaluation tool integrated into the picture archiving and communication system or to standard interpretation without AI. Ten thoracic radiologists interpreted the scans and reported noncalcified nodules measuring at least 4 mm.
The primary outcome was interpretation time per examination, defined as the interval from image opening to report finalization. Secondary outcomes included detection rates of Lung Imaging Reporting and Data System (Lung-RADS)–positive nodules, detection of all nodules, number of nodules detected per scan, and recommendations for follow-up imaging.
Interpretation time was similar between groups (187 vs 172 seconds), with no significant differences across baseline or follow-up examinations.
Detection outcomes differed. Lung-RADS–positive nodules were identified in 17% of scans interpreted with AI compared with 10% without AI. Detection of all nodules was 53% with AI vs 33% without AI. The number of nodules detected per examination was also higher with AI, particularly for nodules measuring 4 to 8 mm.
Follow-up imaging recommendations were more frequent with AI assistance, occurring in 15% of cases compared with 7% without AI. Subgroup analyses showed that increased detection was more pronounced in follow-up examinations, while interpretation time remained unchanged across subgroups. One radiologist demonstrated longer interpretation time with AI, while the remaining radiologists showed no meaningful differences.
Despite higher detection and more follow-up recommendations, no patients in either group were diagnosed with lung cancer during a median follow-up of about seven months. Among patients who underwent follow-up imaging, nodules were stable or resolved, suggesting limited short-term clinical impact of increased detection.
The study was limited by its single-center design, inclusion of asymptomatic individuals undergoing screening, including many at low risk for lung cancer, use of a dedicated reporting interface that may not fully reflect routine workflows, and short follow-up duration. The findings also reflect a specific implementation of AI within a picture archiving and communication system, which may not generalize to other clinical environments.
“Use of the AI tool was associated with significantly greater detection of clinically actionable nodules” but “was not associated with a significant difference in interpretation times,” wrote lead study author Eui Jin Hwang, MD, PhD, of Seoul National University Hospital, and colleagues.
Disclosures: The study was funded by Coreline Soft, which developed the AI tool. The funder had no role in study design, data collection, analysis, or manuscript preparation. Several researchers reported research funding, consulting, or equity relationships with imaging and AI companies outside the submitted work.
