학회지검색
대한유방검진의학회 학회지
Original Article
1Department of Radiology, Yongin Severance Hospital, Yonsei University College of Medicine, Yongin, Gyeonggi-do, Korea 2Department of Radiology, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea 3Department of Statistics, Keimyung University, Daegu, Korea
Background: Artificial intelligence-based computer assisted diagnosis (AI-CAD) has been shown to improve the diagnostic performance of breast cancer diagnosis on mammography. We evaluated the diagnostic performances of AI-CAD and the conventional CAD through a one-on-one comparison.
Materials and Methods: From January to December 2017, of 997 women who visited a health examination center to undergo screening mammography, 978 had normal or benign results with stable follow-up for two years and 19 had cancer diagnosed within the two years of follow-up. Conventional CAD was applied when performing mammography and AI-CAD was retrospectively applied. We compared the diagnostic performances of the two CADs used and did a case-level comparison for immediately and delayed diagnosed cancers.
Results: Standalone AI-CAD presented significantly higher specificity (92.7% vs. 48.7%, P <0.001), PPV (14.5%, 2.3%, P <0.001), and accuracy (92.2% vs. 48.9%, P <0.001) than conventional CAD. For 978 women without breast cancers, conventional CAD presented at least one mark for 502 women (51.3%), which was significantly higher than AI-CAD for 71 women (7.2%). AI-CAD correctly localized three cancers among eight delayed diagnosed cancers, and conventional CAD detected two of them.
Conclusion: AI-CAD showed better diagnostic performance than conventional CAD, by lowering the number of false-positive results with higher specificity.
Key Words: Digital Mammography; Breast cancer; Diagnosis, Computer-Assisted; Artificial Intelligence