Lancet. 2026 Jan 31;407(10527):505-514. doi: 10.1016/S0140-6736(25)02464-X.
ABSTRACT
BACKGROUND: Evidence indicates that artificial intelligence (AI) can improve mammography screening by increasing cancer detection and reducing screen reading workload, but its effect on interval cancers (primary breast cancers diagnosed between two screening rounds or within 2 years after the last scheduled screening that were not detected at screening) is unknown. We aimed to compare the interval cancer rate in AI-supported mammography screening with standard double reading without AI.
METHODS: In this Swedish randomised, controlled, non-inferiority, single-blinded, population-based screening accuracy trial, participants were allocated in a 1:1 ratio to either AI-supported mammography screening (the intervention group) or standard double reading without AI (the control group). AI was used to triage examinations to single or double reading by radiologists and for detection support. This is a protocol-defined analysis of the primary outcome, interval cancer rate, with a 20% non-inferiority margin. Secondary outcomes reported in this analysis are interval cancer characteristics, sensitivity, specificity, and sensitivity by age, breast density, and cancer type (in-situ and invasive). Other secondary outcomes from the trial that have been previously reported are referenced in the Methods section of this Article. The trial is registered with ClinicalTrials.gov (NCT04838756) and is complete.
FINDINGS: Between April 12, 2021, and Dec 7, 2022, 105 934 women were randomly assigned to the intervention or control group, of whom 19 were excluded from the analysis. Median age was 53·8 years (IQR 46·5-63·3) in the intervention group and 53·7 years (46·5-63·2) in the control group. Interval cancer rates were 1·55 (95% CI 1·23-1·92) and 1·76 (1·42-2·15) per 1000 participants in the intervention and control group respectively, a non-inferior proportion ratio of 0·88 (95% CI 0·65-1·18; p=0·41). Descriptively, the intervention group had fewer interval cancers that were invasive (75 vs 89), T2+ (38 vs 48), or non-luminal A (43 vs 59) than the control group. Sensitivity was higher in the intervention group (80·5% [95% CI 76·4-84·2]) than the control group (73·8% [68·9-78·3]; p=0·031), an effect consistent across age and breast density, and for invasive cancer but not for in-situ cancer. Specificity was 98·5% (95% CI 98·4-98·6) for both groups (p=0·88).
INTERPRETATION: AI-supported mammography screening showed consistently favourable outcomes compared with standard double reading, with a non-inferior interval cancer rate, fewer interval cancers with unfavourable characteristics, higher sensitivity, and the same specificity, while also reducing screen reading workload. These findings imply that AI-supported mammography screening can efficiently improve screening performance compared with standard double reading and may be considered for implementation in clinical practice.
FUNDING: Swedish Cancer Society, Confederation of Regional Cancer Centres, Swedish governmental funding for clinical research.
PMID:41620232 | DOI:10.1016/S0140-6736(25)02464-X
