Breast cancer screening is the field in which AI has so far been tested most thoroughly. Two of the largest studies on medical AI ever conducted come from Europe – and their results are encouraging.
How screening works today
In Germany, women aged between 50 and 75 are invited for a mammogram every two years. Each image is assessed independently by two specialists (“double reading”). This is thorough but labour-intensive – and even so, some tumours are missed or develop between two examinations (interval cancers).
The Swedish MASAI trial
In the randomised MASAI trial, 105,934 women in Sweden were either double-read in the conventional way or screened with an AI-supported workflow: the AI rated the cancer risk of each image. Unremarkable images were read by only one person, suspicious ones by two. In addition, the AI marked suspicious areas.
- Sensitivity was 80.5 per cent with AI and 73.8 per cent without AI. Specificity was equally high in both groups (98.5 per cent).
- Interval cancers occurred slightly less often with AI (1.55 versus 1.76 per 1,000 women). The difference was not statistically significant, but the trial was designed to show that AI is not inferior.
- Notably, the AI group had fewer interval cancers with unfavourable characteristics, such as larger or more aggressive tumours.
- According to an earlier analysis, the workload for readers fell by 44 per cent.
The final results were published in The Lancet in February 2026.
The German PRAIM study
The PRAIM study, led by the University of Lübeck and the University Hospital Schleswig-Holstein, investigated use in real-world screening: 463,094 women at twelve screening sites. The doctors were free to decide whether to use the AI.
- With AI, 6.7 breast cancer cases were detected per 1,000 women, compared with 5.7 without AI – an increase of 17.6 per cent.
- The recall rate fell slightly from 38.3 to 37.4 per 1,000 women.
- The proportion of recalls that actually turned out to be cancer rose from 14.9 to 17.9 per cent.
The results were published in Nature Medicine in January 2025.
What this means for women
Both studies suggest that AI helps to find more tumours without more women being recalled because of a false alarm. At the same time, it could help to cushion the shortage of staff in radiology. Importantly, in both approaches the doctor’s assessment remains central. The AI does not replace the readers but directs their attention.
Open questions
- Overdiagnosis: are the additional findings mainly small, slow-growing tumours that would never have caused symptoms? Early data on less aggressive interval cancers point rather to a real benefit, but this can only be clarified definitively over a longer period.
- Mortality: it has not yet been demonstrated that AI-supported screening reduces breast cancer mortality.
- Transferability: the results apply to the programs and workflows used. Other software or other devices must be tested separately.
- Funding and rules: before AI can be used routinely in the German screening programme, the regulations will have to be adapted.
Conclusion
Mammography screening is a showcase example of how medical AI should be properly tested: in large, partly randomised studies under real-world conditions. Regardless of the technology, taking part in screening remains an individual decision in which the benefits and possible disadvantages should be weighed up.
More on this topic: AI in Medicine 2026 – an overview.
Note: This article is for general information only and does not replace medical advice. As of: October 2026.
Video on this article
Watch on YouTube · Subscribe to the channel
Sources
- ASCO Post: Randomized Trial Shows AI-Supported Mammography Improves Sensitivity and Lowers Interval Cancer Rate (The Lancet, 2026)
- Lund University: AI support in breast cancer screening – fewer missed cancer cases
- Pharmazeutische Zeitung: KI verbessert Brustkrebsscreening signifikant (2025)
- Universität zu Lübeck: Projekt PRAIM
- Eisemann N et al.: Nationwide real-world implementation of AI for cancer detection in population-based mammography screening. Nature Medicine 2025, DOI 10.1038/s41591-024-03408-6
Leave a Reply