Bolstering the Performance of Breast Radiologists with AI-CAD in Mammography: A Multireader Study.
Authors
Affiliations (11)
Affiliations (11)
- Department of Oncologic Imaging, National Cancer Center, Singapore (W.L.L.O.).
- Department of Breast Imaging and Intervention, Singapore General Hospital, Singapore (W.M.I.T.).
- Healthway Cancer Care Hospital, Philippines (M.T.B.).
- Department of Radiology, Mansoura University, Egypt (M.E.).
- Department of Breast Imaging, Lisie Hospital, India (J.P.D.).
- Cardinal Santos Medical Center, Philippines (N.J.A.L.).
- Department of Breast Imaging, Prajnam Complete Breast Care, India (S.K.).
- Department of Breast Imaging, Davao Doctors Hospital, Philippines (R.L.A.).
- Department of Medical Imaging, Tung Wah College, Hong Kong, China (C.L.).
- BIEDX Pte Ltd., Singapore (R.K.S.).
- Department of Women's Imaging, RadLink Diagnostic and Imaging Center, Singapore (N.C.). Electronic address: [email protected].
Abstract
Breast cancer is the most common malignancy among females globally and across most Asian countries. In 2022, Asia's age-standardized incidence rate (ASIR) was 34.3/100,000, with age-standardized mortality rate (ASMR) of 10.5.1000,000. Many Asian countries experience high incidence-to-mortality ratio due to limited organized screening programs, resource constraints and manpower limitations. The application of artificial intelligence (AI) in mammography interpretation may help address these challenges. This multinational retrospective study evaluated the performance of AI-based computer-aided diagnosis (AI-CAD) in mammographic interpretation. A total of 302 digital mammograms, including 89 biopsy-proven breast cancers, were interpreted by nine experienced breast radiologists from multiple Asian institutions. Each radiologist participated in two reading sessions-one unaided and one with AI-CAD assistance. Diagnostic performance and reading time were compared between sessions. AI-CAD assistance significantly improved diagnostic performance, with the average area under the receiver operating characteristic curve (area under the curve [AUC]) increasing from 0.799 to 0.851 (p = 0.0151). Specificity improved from 77.0-88.4% (p = 0.03), while sensitivity showed no statistically significant difference. AI assistance also led to a significant reduction in average interpretation time, from 121.5 to 83.2 s per case (p < 0.001). AI-CAD significantly enhances specificity and reduces reading time in mammographic interpretation without compromising sensitivity. These findings support the integration of AI in breast cancer screening workflows, to improve diagnostic efficiency and optimize clinical outcomes. Importantly, they also underscore the importance of maintaining human oversight and critical judgment when using AI in clinical practice.