A multi-reader multi-case study comparing Breast Cancer Reading Efficiency and Accuracy in Screening with Tomosynthesis versus AI-assisted mammography.
Authors
Affiliations (6)
Affiliations (6)
- Saw Swee Hock School of Public Health, National University Health System, National University of Singapore.
- National University Health System, Department of Breast and General Surgery, Singapore.
- Department of Surgery, Yong Loo Lin School of Medicine. National University of Singapore.
- Department of Diagnostic Imaging, National University Hospital, Singapore.
- Department of Diagnostic Radiology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore.
- Department of Radiology and Imaging Sciences, Emory University School of Medicine, Atlanta, USA.
Abstract
Full-field digital mammography (FFDM) is the standard for breast cancer screening but has reduced sensitivity in dense breasts. Digital breast tomosynthesis (DBT) improves cancer detection compared to 2D mammography but is limited by cost and longer reading times. Artificial intelligence (AI)-enhanced mammography has demonstrated improved sensitivity compared to 2D mammography. This study compares AI-enhanced 2D mammography (DM+AI) with DBT in recalling interval breast cancers (IBCs). : A multi-reader, multi-case crossover study was conducted using the Emory BrEast Imaging Dataset, which includes paired FFDM and DBT images with biopsy-proven outcomes. Six radiologists reviewed 53 IBC and 53 screen-detected cancer (SDC) cases, with random assignment to DM+AI or DBT, followed by crossover after a washout period. Primary outcomes included recall rate of IBCs. Secondary outcomes included recall rate of SDCs, interpretation time, and subgroup analysis by breast density. : For IBCs, DM+AI and DBT demonstrated similar recall rates (41.5% vs 40.9%) and detection rates (14.8% vs 11.9%). DM+AI identified more non-invasive (31.5% vs 13.0%) and mass-forming (61.0% vs 21.8%) cancers. For SDCs, recall and detection rates were comparable across modalities. In fatty breasts, recall was higher with DM+AI (91.7% vs 50.0%), while DBT performed better in extremely dense breasts. DM+AI enabled faster interpretation, averaging 24.1 seconds less per case. : AI-enhanced 2D mammography demonstrated comparable performance to DBT with shorter interpretation times, supporting a stratified screening approach. This study is the first direct comparison of AI-enhanced mammography and digital breast tomosynthesis for IBC detection.