Retrospective Evaluation of an AI-Based Computer-Aided Detection Algorithm for Lung Cancer Detection on Cardiac CT: A Multicenter Study.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology and Research Institute of Radiological Science, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea.
- Department of Radiology, Dongsan Medical Center, Keimyung University College of Medicine, Daegu, Korea.
- Department of Radiology and Imaging Sciences, Emory University, Atlanta, GA, USA.
- Department of Radiology, Seoul St. Mary's Hospital, College of Medicine, The Catholic University of Korea, Seoul, Korea. [email protected].
- Department of Radiology and Research Institute of Radiological Science, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea. [email protected].
Abstract
To investigate the effectiveness of an artificial intelligence (AI)-based computer-aided detection (CAD) system in identifying incidental lung cancer on cardiac computed tomography (CT) scans and to compare its performance with that of radiologists. In this retrospective, multicenter study, 652 cardiac CT scans from 581 patients subsequently diagnosed with lung cancer were analyzed. A commercial AI-CAD system was employed to detect pulmonary lesions on cardiac CT. The detection rate of AI-CAD was compared to that of the radiologist, based on the radiology report, as well as to the detection rate when combining AI-CAD and the radiologist. The characteristics of the lesions detected and missed by the radiologist and AI-CAD were compared. Radiologists and AI-CAD demonstrated similar detection rates for lung cancer (76.2% vs. 77.4%, <i>P</i> = 0.551). However, combining radiologists and AI-CAD significantly improved the detection rate to 90.4% (<i>P</i> < 0.001) compared to that of the radiologist alone. AI-CAD showed a higher detection rate in identifying small, peripheral, and part-solid lesions (all <i>P</i> < 0.001). Furthermore, AI-CAD outperformed radiologists in detecting limited-stage lung cancer (80.3% vs. 74.7%, <i>P</i> = 0.006). Among lung cancer cases missed by radiologists, 94.2% experienced diagnostic delays of > 100 days, with 78.2% leading to stage progression. AI-CAD identified 58.5% of these diagnostic delays. AI-CAD demonstrated the potential to improve the detection rate of incidental lung cancer by identifying a subset of lesions that were initially overlooked by radiologists on cardiac CT. It exhibited particular strength in identifying early-stage cancers and small, subsolid lesions.