AI-Supported Mammography Caught More Cancers During Screening

Favorable outcomes were found when artificial intelligence (AI) was used in mammography screenings compared with the standard double reading in a new study published in The Lancet.1 The use of AI was found to lead to a noninferior interval cancer rate, higher sensitivity, and fewer interval cancers with unfavorable characteristics.

AI has been utilized in medicine more frequently, especially in the reading of different tests and images. With promising results in diagnostic accuracy for AI, the use of AI in reading mammography screenings has been of interest in improving detection and reducing workload for radiologists. Interval cancers, which are diagnosed between screening rounds, were also of interest in terms of the ability for AI to screen for them. This analysis aimed to determine how well AI-supported screenings were able to detect interval cancer compared with standard double reading.

AI-supported breast cancer screening was able to reduce interval cancers by 12%. | Image credit: lordn - stock.adobe.com

The Mammography Screening with Artificial Intelligence (MASAI) trial (NCT04838756) was the basis of this analysis, where Swedish women aged 40 to 74 years were invited to screen every 1.5 to 2 years, and those with moderate risk due to family history were invited to screen annually. Women were included in MASAI if they had had their screening at 1 of the 4 screening sites in southwest Sweden. No women were excluded from the study.

All screening examinations were randomized 1:1 to either the AI-supported screening or the standard double reading group. The AI model used for the study had been trained and validated with more than 200,000 past examinations. The AI provided a risk score between 1 and 10 and a score of 1 to 98 for region marks and region-specific scores. Examinations of each mammography were conducted by 1 or 2 of the 16 radiologists; all but 2 of the radiologists had more than 5 years of experience. The screenings were randomized for assessment by the radiologists.

The primary outcome was the interval cancer rate, which was defined as primary breast cancer diagnosed between 2 screening rounds or within 2 yeas of the last scheduled screening that was not detected at a screening. The interval cancer characteristics, sensitivity, and specificity were the secondary outcomes of this study.

There were 105,915 women who were screened between April 12, 2021, and December 7, 2022, and included in this study. There were 53,052 assigned to the AI-supported screening group and 52,882 assigned to the double reading group. The median (IQR) age of the intervention group was 53.8 (46.5-63.3) years, and of the control group, 53.7 (46.5-63.2) years. All participants had at least a 2-year follow-up as of December 7, 2024.

There were 82 women in the interval group and 93 in the control group who had been diagnosed with interval cancer, amounting for 1.55 and 1.76 interval cancers diagnosed per 1000 participants, respectively. The intervention group had a higher sensitivity compared with the control group (80.5% vs 73.8%). Both the intervention and control groups had a specificity of 98.5%. Sensitivity in the AI-supported screening was higher compared with standard screening when it came to invasive cancer (78.3% vs 70.9%).

The median age of participants who were diagnosed with an interval cancer was 57.6 (49.0-64.6) years in the intervention group and 58.1 (48.5-66.3) years in the control group. AI-supported screening resulted in 12% fewer interval cancers and 16% fewer invasive interval cancers compared with standard screening. There were 27% fewer non–luminal A subtypes and 7% luminal A subtypes in the intervention group.

There were some limitations to this study. The study was conducted in Sweden, which could limit the generalizability. Outcomes could vary with radiologists with less experience or if different procedures were followed for screening. Race and ethnicity were not recorded. Settings with fewer resources may not be able to implement this AI screening. There was only 1 screening round during this trial. The way AI influences mammography screening over time would change, which could also have affected results.

“The MASAI trial showed consistently more favorable outcomes with AI-supported mammography screening compared with standard double reading without AI, including the primary outcome of interval cancer rate, showing noninferiority, and fewer interval cancers with unfavorable characteristics,” the authors concluded. Cost-effectiveness and how subsequent screening rounds affect the results should be further analyzed in future studies, they added.

"Our study does not support replacing health care professionals with AI, as the AI-supported mammography screening still requires at least 1 human radiologist to perform the screen reading, but with support from AI. However, our results potentially justify using AI to ease the substantial pressure on radiologists’ workloads, enabling these experts to focus on other clinical tasks, which might shorten the waiting times for patients,” said lead author Jessie Gommers, PhD student, Radboud University Medical Centre in the Netherlands.2

References

1. Gommers J, Hernström, 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;407:505-514.
2. AI-supported mammography screening results in fewer aggressive and advanced breast cancers, finds full results from first randomised controlled trial. News release. Lancet. January 29, 2026. Accessed January 29, 2026.