ORIGINAL RESEARCH article
Front. Oncol.
Sec. Cancer Imaging and Image-directed Interventions
This article is part of the Research TopicHarnessing Machine Learning for Enhanced Biomedical Diagnosis and Early Disease Detection: Bridging Data Science and HealthcareView all 14 articles
Dual-modality CAD for breast cancer screening: Dealing with discordant diagnosis between mammography and tomography
Provisionally accepted
- Therapixel, Paris, France
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Background Full Field Digital mammography (FFDM) is the standard for breast cancer screening. Digital breast tomosynthesis (DBT), compared to FFDM, enhances cancer detection, and reduces unnecessary biopsies. Despite DBT's adoption, critical questions remain - higher radiation, time, cost, and clinical benefits, particularly for systematic breast screening. In the era of AI CADs (Computer-Aided Detection/Diagnosis) for breast screening, one unresolved question is the role of bimodal algorithms in predicting cancer and offering guidance when opinions differ, and we aim to understand this. Methods We retrospectively assembled an enriched screening cohort of 1,816 women who underwent both FFDM and DBT at two Hologic sites. Analyses requiring paired CAD scores were performed on a lesion-level subset for which both FFDM and DBT CAD scores were available (low suspicion = 1; high suspicion = 10) and reference standard outcomes were known, comprising 1,071 lesions from 657 examinations. From the joint distribution, we defined areas of "perpendicular scoring" (PS) as the areas of highly discordant scoring. We estimated the inter-modality agreement using three classes (Low, indeterminate and high suspicious) Kappa Cohen index. We evaluated the potential of systematic, lossless and AI powered reclassifications of PS both for tumoral masses and calcifications and in considering breast density as a risk factor for PS. Results We observed a moderate inter-modality agreement, indicated by a kappa of 0.49 (95% CI: 0.46; 0.52). PS scoring was present in 32.7% (95% CI: 29.7; 35.8) of tumoral masses (soft tissue lesion) cases and 38.6% (95% CI: 30.1; 47.6) of calcification cases. Breast density was a risk factor of PS for masses (Odd, 0.66 [95% CI: 0.48; 0.91]). AI-powered and lossless models were found effective for reclassifying 82.2% and 67.3% of PS of respectively masses and calcification. Conclusions When processed by CAD, FFDM and DBT provided complementary information at the expense of unavoidable discordant diagnosis. Post-processing has the potential of reclassifying part of discordant diagnosis in improving the overall performance of the CAD, therefore, exploring alternative reclassification methods is essential.
Keywords: Breast Neoplasms, Computer-Assisted, Mammography, Mass Screening, Reproducibility ofResults
Received: 02 Nov 2025; Accepted: 05 Feb 2026.
Copyright: © 2026 Beaumont, IANNESSI, Louis, Pacile and Fillard. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Hubert Beaumont
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.