A large randomized, controlled clinical trial has found that artificial intelligence (AI)–supported mammography screening results in fewer interval breast cancer diagnoses compared with standard double readings by radiologists. The findings, from the Mammography Screening with Artificial Intelligence (MASAI) trial, were published in The Lancet and mark a significant milestone in the evaluation of AI in population-based cancer screening.
The MASAI trial is the first randomized controlled study to assess AI use in breast cancer screening and the largest to date examining AI applications in cancer screening overall. According to corresponding author Kristina Lång, PhD, of the Division of Diagnostic Radiology at Lund University in Sweden, the results show that AI-supported screening improves the early detection of clinically relevant breast cancers and reduces the number of aggressive or advanced cancers diagnosed between routine screening rounds.
Study Design and Methods
Conducted in Sweden, the MASAI trial was a randomized, controlled, non-inferiority, single-blinded, population-based study designed to evaluate whether AI-supported mammography readings could reduce interval cancer rates. Interval cancers are defined as primary breast cancers diagnosed between scheduled screening rounds.
Participants were randomly assigned in equal numbers to either an intervention group receiving AI-supported mammography screening or a control group undergoing standard double reading by radiologists without AI assistance. In the intervention arm, AI was used both to triage mammograms—determining whether scans required single or double radiologist review—and to support detection by highlighting suspicious findings.
The AI system used in the study was trained, validated, and tested on more than 200,000 mammography scans. The primary outcome measure was the interval cancer rate, while secondary endpoints included interval cancer characteristics, sensitivity, specificity, and subgroup analyses based on age, breast density, and cancer type.
Key Results
Of the 105,934 women initially randomized, 19 were excluded from the final analysis. The interval cancer rate was lower in the AI-supported group at 1.55 per 1,000 participants (95% CI, 1.23–1.92), compared with 1.76 per 1,000 participants (95% CI, 1.42–2.15) in the control group. The noninferiority proportion ratio was 0.88 (95% CI, 0.65–1.18; P = .41).
Notably, the intervention group had fewer interval cancers with unfavorable characteristics. These included fewer invasive cancers (75 cases versus 89 in the control group), fewer tumors classified as stage T2 or higher (38 versus 48), and fewer non–luminal A breast cancers (43 versus 59).
Screening sensitivity was significantly higher with AI-supported readings than with standard double reads (80.5% vs 73.8%; P = .031), while specificity remained equivalent between the two groups at 98.5% (P = .88). Higher sensitivity in the intervention group was consistent across subgroups defined by age, breast density, and invasive cancer type, though no difference was observed for in-situ cancers.
Broader Implications
Earlier interim safety analyses from the MASAI trial, published in The Lancet Oncology, showed that AI support reduced radiologists’ workload by 44%. Additional analyses also found that AI-supported screening detected 29% more cancers without increasing false-positive rates.
Despite these promising results, the researchers emphasized that AI is not intended to replace radiologists. “AI-supported mammography screening still requires at least one human radiologist,” said first author Jessie Gommers, MSc, a PhD student at Radboud University Medical Centre in the Netherlands. However, she noted that AI could help alleviate workforce pressures, allowing radiologists to focus on other clinical tasks and potentially reducing patient wait times.
Dr. Lång added that further follow-up studies and cost-effectiveness analyses are needed to assess the long-term benefits and risks of AI-supported mammography. If favorable outcomes persist in future screening rounds, she said, the evidence could support broader adoption of AI in population-wide breast cancer screening—particularly in health systems facing staffing shortages.
