Take-Away Points
■ Major Focus: To evaluate noninferiority of artificial intelligence (AI)–supported mammography interpretation compared to standard human double reading without AI in a population-based breast cancer screening program.
■ Key Result: The interval cancer rate was 1.55 per 1000 participants among the 53 043 female individuals randomized to AI-supported screening, which was noninferior (proportion ratio, 0.88; P = .41) to the interval cancer rate of 1.76 per 1000 participants among the 52 872 female individuals randomized to standard human double reading.
■ Impact: The Mammography Screening With Artificial Intelligence (MASAI) trial demonstrated favorable outcomes with AI-supported mammography screening, which streamlines and optimizes screening workflows.
For many years, AI-supported mammography interpretation followed the familiar trajectory of medical AI—early enthusiasm and ambitious predictions tempered by limited high-level evidence. Since 2022, however, several studies have demonstrated that AI tools can positively influence established screening end points, including cancer detection rate and interval cancer rate, the latter being a particularly meaningful surrogate of screening benefit in contemporary trials.
Gommers et al present the eagerly awaited data on interval cancer rates in the MASAI trial, a Swedish randomized controlled trial that introduced AI into the screening workflow to triage mammograms. The very few examinations deemed high risk underwent human double reading, whereas the large number of those flagged as low risk were interpreted by a single reader. A total of 105 934 female individuals were randomized 1:1 to standard human double reading or to AI-supported screening. Importantly, the trial was designed to test noninferiority of AI-supported screening rather than superiority. This objective was achieved: The interval cancer rate was 1.55 per 1000 participants (95% CI: 1.23, 1.92) in the AI-supported screening group and 1.76 per 1000 participants (95% CI: 1.42, 2.15) in the control group, corresponding to a noninferior proportion ratio of 0.88 (95% CI: 0.65, 1.18; P = .41). AI-supported interpretation also yielded favorable results across several secondary end points, including reductions in interval cancers with adverse prognostic features (eg, invasive cancers, T2+ cancers, non–luminal A cancers).
These findings provide a much needed foundation for integrating AI into breast cancer screening in a manner that preserves human resources, which is particularly relevant amid ongoing radiologist shortages, while maintaining key clinical performance benchmarks. This trial also paves the way for further evaluation of long-term effects and optimization of implementation strategies.
Highlighted Article
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. Lancet 2026;407(10527):505–514. doi: https://doi.org/10.1016/S0140-6736(25)02464-X
Highlighted Article
- 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 . Lancet 2026. ; 407 ( 10527 ): 505 – 514 . doi: 10.1016/S0140-6736(25)02464-X [DOI] [PubMed] [Google Scholar]