Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide. Therefore, population-based mammographic screening programs play a crucial role in early detection. Despite reported reductions in mortality associated with screening, conventional mammography has limitations, with studies suggesting that approximately 20–40% of cancers may be missed. The introduction of digital breast tomosynthesis (DBT) has improved cancer detection rates and reduced recall rates. However, the volumetric nature of DBT substantially increases the number of images that radiologists must interpret, leading to a 50–200% increase in reading time and raising concerns regarding the operational sustainability of screening programs.

First-generation computer-aided detection (CADe) systems failed to demonstrate clear diagnostic benefit in clinical practice and produced high false-positive rates that reduced radiologist confidence. When combined with the large number of image slices in DBT examinations, these limitations have further increased the workload for radiologists.

Recent advances in artificial intelligence, particularly deep learning–based systems, have demonstrated promising results in breast cancer detection on mammography. In several studies, AI systems have achieved performance comparable to or even exceeding that of average radiologists. When used as concurrent decision support tools, AI algorithms can improve diagnostic accuracy without increasing reading time. Evidence from DBT screening studies suggests that AI-assisted workflows may reduce radiologist workload by approximately 30–70% without clinically meaningful loss of sensitivity. The MASAI randomized trial demonstrated that AI-supported single reading detected more cancers than conventional double reading while significantly reducing radiologist workload.

Despite these promising results, several challenges remain, including model generalizability across populations, suboptimal detection of certain lesion types, potential algorithmic bias, and unresolved medicolegal considerations in clinical implementation.