AI Mammography Triage Captures 50% of Screen-Detected Cancers in Top 2% of Scans — JACR Study

A study published in the Journal of the American College of Radiology (JACR), conducted by researchers at UC Davis Health (lead author Diana Miglioretti), evaluated two FDA-cleared AI mammography triage algorithms using the ClinValAI framework on 3,468 digital breast tomosynthesis (DBT) screening exams that included 25 confirmed breast cancers. Key finding: flagging only the top 2% of scans by AI risk score identified approximately 50% of all screen-detected cancers, with area under the curve (AUC) values of 0.952 and 0.908 for the two algorithms respectively. This performance suggests AI could enable same-day diagnostic mammography by prioritizing the highest-risk examinations for urgent radiologist review, potentially reducing diagnostic delays that disproportionately affect under-resourced and rural populations lacking specialist radiologists. The study is one of the first head-to-head comparisons of multiple FDA-cleared AI mammography products on an identical prospective real-world DBT dataset. Researchers emphasized AI operates as a decision support tool — both algorithms evaluated require radiologist oversight under current FDA clearances — and noted false-positive rate management and cross-population generalizability remain important considerations for clinical deployment.