A prospective evaluation of 10,889 women in the UK has found that adding AI to breast cancer screening detected 10.4% more cancers while reducing radiologist workload by up to 31%. The GEMINI study, published in Nature Cancer, represents the UK’s largest real-world test of AI-assisted mammography.
What They Did
Researchers from the University of Glasgow, University of Aberdeen, NHS Grampian, and Kheiron Medical Technologies (now part of DeepHealth Inc.) tested how an AI system called Mia v.3 could fit into routine breast screening workflows.
All 10,889 participants received standard care with double-reading by radiologists. But researchers also ran Mia on every scan and tracked where the AI disagreed with human readers. When AI flagged something the radiologists missed, an additional human reviewer examined those cases.
The team didn’t just test one approach. They simulated 17 different workflow configurations to find optimal ways to integrate AI into existing screening programs.
The Numbers
The headline findings:
Detection improvement: The AI-assisted workflow caught 10.4% more cancers — one additional case per 1,000 women screened. When AI disagreed with routine double-reading and triggered extra review, radiologists found 11 additional cancers that would have been missed.
Workload reduction: Radiologist reading volume dropped by up to 31%. In optimized workflows, this could reach 36% — significant relief for chronically understaffed screening programs.
Faster notification: Women with detected cancers were notified in 3 days instead of 14. For high-grade cancers that benefit from early treatment, this matters.
Recall rates: The AI workflow actually reduced recalls by 0.8%, meaning fewer women called back for additional testing that turns out negative.
Different From MASAI
This isn’t the Swedish MASAI trial published last month. That randomized study showed AI-supported screening caught 9% more cancers with 27% fewer aggressive interval cancers. GEMINI is a prospective feasibility study focused on implementation — how to actually integrate AI into NHS workflows rather than just proving AI can detect cancer.
Both studies used different AI systems (MASAI used Transpara, GEMINI used Mia), different study designs, and different populations. But they’re converging on similar conclusions: AI catches cancers that skilled radiologists miss.
The Fine Print
A few caveats worth noting.
The study focused on one UK region (NHS Grampian in Scotland). Results may vary in populations with different breast density distributions or screening practices.
Mia v.3 has been trained for over a decade on international screening data. Newer, less mature AI systems might not match these results.
The 10.4% improvement sounds dramatic, but breast cancer screening is already highly effective. We’re talking about the difference between catching 6.3 versus 7.3 cancers per 1,000 women. Important for those women and their families, but not a transformation of screening outcomes.
What Comes Next
The research team is now running the EDITH trial, a larger evaluation across multiple UK screening programs. If results hold, AI assistance in mammography could become standard NHS practice.
For the radiologists drowning in screening volumes, this might be the rare AI application that genuinely helps without threatening their jobs. The AI isn’t replacing readers — it’s acting as a highly attentive second opinion that never gets tired.