AI-Supported Mammogram Reading Caught More Cancers Without More False Alarms

CATEGORY: WOMEN’S HEALTH

Mammogram screening in which artificial intelligence helped radiologists read the scans caught 80.5% of breast cancers, compared with 73.8% for the standard approach of two radiologists reading every scan—with no increase in false alarms—in a randomized trial of 105,934 Swedish women, the first trial of its kind to report complete results.¹

AI Routing Cut Screen Reading Workload by 44%

Screening mammograms in much of Europe are read by two radiologists, a safeguard that catches more cancers but doubles the reading work. The MASAI trial asked whether an AI system could do better: it screened each scan, flagged suspicious areas for the radiologists, and routed the clearly low-risk majority to a single reader instead of two. Earlier results from the same trial showed this cut the radiologists’ screen-reading workload by 44%.² What remained unknown—and what this final report answers—is whether the AI-supported approach actually caught cancers as reliably over time.

Randomized at Four Swedish Sites, Cancers Tracked by Registries

Between April 2021 and December 2022, women arriving for routine screening at four Swedish sites were randomly assigned, half and half, to AI-supported reading or standard double reading. Then, over the roughly two years until each woman’s next scheduled visit, Sweden’s cancer registries recorded every breast cancer diagnosed in both groups—including interval cancers, meaning cancers that surface between screening visits after a normal result. Counting those is what let researchers measure the share of cancers each method truly caught, not just how many scans it flagged.

Interval Cancers 1.55 Versus 1.76 per 1,000 Women

The AI-supported group’s screening caught 80.5% of the breast cancers that emerged, versus 73.8% with standard double reading, and the advantage held across age groups and breast densities. False alarms did not rise: in both groups, about 98.5% of women without cancer were correctly left unrecalled. Interval cancers were slightly less common in the AI group—1.55 versus 1.76 per 1,000 women—a difference small enough that it could be due to chance, though fewer of those missed cancers were invasive, large, or aggressive subtypes.¹ The AI-supported group had 75 invasive interval cancers, compared with 89 in the standard double-reading group.¹

Humans Still Read Every Scan, Deaths Not Measured

The AI assisted radiologists rather than replacing them—every scan still had at least one human reader, and a callback after screening is not a diagnosis, since most recalls turn out to be benign findings such as cysts or overlapping tissue. The trial measured cancer detection, not deaths; longer follow-up would be needed to show whether earlier detection translates into fewer breast cancer deaths. The results also come from Sweden’s organized screening program, and countries that already use single reading, like the United States, may see different effects.

Germany’s 463,094 Women, 18% More Cancers Detected

A separate study of Germany’s national screening program pointed the same way. Among 463,094 women, radiologists using AI support detected about 18% more breast cancers than those reading without it—6.7 per 1,000 women screened versus 5.7—while callbacks for further tests did not increase.³ That was real-world practice rather than a randomized trial, so it shows an association rather than proof that the AI itself caused the difference.

Key Takeaways

• In a randomized trial of 105,934 women, AI-supported mammogram reading caught 80.5% of breast cancers versus 73.8% with standard double reading by two radiologists, with no rise in false alarms.
• Cancers surfacing between screening visits were no more common with AI support, and fewer of them were invasive or aggressive types—though that difference was small enough to be due to chance.
• The AI assisted radiologists rather than replacing them, and earlier results from the same trial showed it cut screen-reading workload by 44%.
• The trial measured detection, not mortality—whether AI-supported screening leads to fewer breast cancer deaths will take longer follow-up to know.

References

1. Gommers J, Hernström V, Josefsson V, et al. Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial. The Lancet. 2026;407(10527):505–514. link
2. Lång K, Josefsson V, Larsson AM, et al. Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study. The Lancet Oncology. 2023;24(8):936–944. link
3. Eisemann N, Bunk S, Mukama T, et al. Nationwide real-world implementation of AI for cancer detection in population-based mammography screening. Nature Medicine. 2025;31(3):917–924. link.