Prospective evaluation of artificial intelligence integration into breast cancer screening in multiple workflow settings: the GEMINI study

Clarisse Florence de Vries, Gerald Lip, Roger Todd Staff, Jaroslaw Artur Dymiter, Benjamin Tse, Annie Ng, Georgia Fox, Cary Oberije, Lesley Ann Anderson
Nat Cancer. 2026 Mar;7(3):484-493. doi: 10.1038/s43018-026-01126-1. Epub 2026 Mar 10.
Abstract
Artificial intelligence (AI) tools can improve breast screening performance but different screening sites have varying needs. Here the GEMINI prospective evaluation of 10,889 women, within one UK region, used both live AI integration and simulations to model 17 different ways AI could be used in breast screening. All women received routine care. One AI tool was assessed. When the AI tool recommended recall but routine double reading did not, cases underwent additional human review, detecting 11 additional cancers. The primary AI workflow could improve cancer detection by 10.4% (1 per 1,000), maintain the recall rate (0.8% reduction) and reduce workload by up to 31%. Other workflow variations significantly improved all measured metrics (superiority in cancer detection rate, recall rate, positive predictive value (PPV), sensitivity and specificity) with up to 36% workload savings. Different AI integrations in breast screening could offer various clinical and operational gains, allowing for adaptation to local healthcare needs.