Healthcare & Bio AI
AI-supported double reading detected more breast cancers than standard reading in German mammography screening
In a large German mammography screening program, women in the AI-supported group had a breast cancer detection rate of 6.7 per 1,000, compared with 5.7 per 1,000 for standard double reading without AI. The study covered 463,094 women aged 50 to 69 at 12 screening sites from July 2021 to February 2023. Radiologists voluntarily chose whether to use the AI system, and 260,739 women were screened with AI support by 119 radiologists. The AI-supported group's detection rate was 17.6% higher, with a 95% confidence interval of +5.7% to +30.8%, which the authors report as statistically superior.
What did the study compare?
The study, called PRAIM, compared AI-supported double reading with standard double reading without AI. Double reading means a second radiologist also reviews each mammogram. In the AI-supported group, radiologists used the AI system. In the control group, they did not.
What were the main results?
The AI-supported group found breast cancer at a rate of 6.7 per 1,000 women screened, while the control group found 5.7 per 1,000. The recall rate, the share of women called back for further tests, was 37.4 per 1,000 in the AI group and 38.3 per 1,000 in the control group. The percentage difference was -2.5%, with a confidence interval of -6.5% to +1.7%. The authors describe this recall rate as lower than and noninferior to the control rate, meaning it did not fall meaningfully below the standard approach.
The positive predictive value of recall, the share of recalls that found cancer, was 17.9% in the AI group and 14.9% in the control group. The positive predictive value of biopsy, the share of biopsies that found cancer, was 64.5% in the AI group and 59.2% in the control group.
Frequently asked questions
Was the AI system required for radiologists?
No, it was not required. Radiologists in the study voluntarily chose whether to use the AI system. The abstract does not say how many radiologists used AI support or how that choice was made.
Does this prove AI caused the higher detection rate?
No. The study is observational, not randomized. Because radiologists chose whether to use AI, the abstract reports an association between AI-supported double reading and a higher breast cancer detection rate, not a proven cause.
Did AI support clearly reduce recall rates?
The recall rate was slightly lower in the AI group, but the abstract calls this noninferior rather than a clear reduction. The AI group had 37.4 recalls per 1,000 women, compared with 38.3 per 1,000 in the control group.
Limits and what's next
The abstract does not state limitations or next steps. It describes the study as observational and notes that radiologists voluntarily chose whether to use the AI system, so the finding is an association rather than a cause. The paper is peer-reviewed and was published in Nature Medicine on 2025-01-07.
Source: Eisemann et al., Nationwide real-world implementation of AI for cancer detection in population-based mammography screening, Nature Medicine, 2025-01-07.