Using artificial intelligence (AI) for breast cancer screening reduces the incidence of malignant cancers by 12%, raising the chances of early detection, a study found. The results, the first from a large clinical trial with more than 100,000 participants, are expected to greatly increase the use of AI in breast cancer diagnosis.
A research team led by Professor Kristina Lång of Lund University Cancer Center in Sweden said, "In a clinical trial conducted within the national breast cancer screening program, AI-assisted image analysis proved more effective than the existing standard for breast cancer diagnosis," in a paper published on the 31st in the international journal The Lancet.
◇ Early diagnosis of aggressive cancers, 44% less workload for doctors
The team said this study is the largest to examine the effectiveness of AI in cancer screening. The clinical trial enrolled 105,934 women in Sweden who underwent mammography screening from April 2021 to December 2022. The average age of participants was 55. They were randomly assigned to a screening group in which AI analyzed images to support doctors, or to a European standard diagnosis group in which two radiologists read the scans.
There are two measures of diagnostic accuracy. Sensitivity is how accurately a test identifies someone with a disease as "having it," and specificity is how accurately it identifies someone without a disease as "not having it." In the trial, sensitivity was 80.5% in the AI-assisted group and 73.0% in the standard reading by two doctors. This means AI support improves the ability to detect cancers without missing them. Specificity was similar between the groups. The rate of mistakenly labeling someone without disease as a cancer patient was 1.5% in the AI-assisted group and 1.4% in the control group.
Notably, the group that received AI-assisted screening saw a 12% reduction in interval cancer diagnoses in the years after screening. Interval cancers are those found between screenings—cases in which the prior screening was normal but cancer is diagnosed before the next screening. In the AI-assisted group, 1.55 cancers were found per 1,000 women, compared with 1.76 in the control group.
Interval cancers are common in breast cancer, grow very quickly, and are aggressive. They are also more likely to metastasize to other parts of the body. A reduction in diagnoses of these dangerous breast cancers suggests that while doctors might miss small tumors that could progress to interval cancers, AI is able to capture them.
Early results from the same trial also showed the benefits of AI diagnostic support. In 2023, the Swedish team reported that AI reduced radiologists' image-reading workload by 44%. Last year, they reported that AI-assisted breast cancer screening increased cancer detection by 29% without a rise in false positives.
The AI used in this clinical trial was developed by ScreenPoint Medical, a Dutch biotech corporations. It marks areas in images that appear cancerous to assist radiologists. When risk is low, a single experienced radiologist reads the scan, while high-risk cases are routed to double reading by two radiologists. The AI was trained on more than 200,000 exam results collected from medical institutions in more than 10 countries.
◇ Additional clinical trials are needed in other ethnic groups
Breast cancer is a leading cause of death among women ages 35 to 50, and more than 2 million people worldwide are diagnosed with the disease each year. For early detection, recommendations are monthly self-exams after age 30, clinical breast exams every two years after 35, and mammography every one to two years after 40.
Mammography is an X-ray exam. Human tissues transmit X-rays differently depending on their density. Cancerous tissue with calcium deposits is denser, like bone, and transmits fewer X-rays. As a result, cancerous areas appear white on X-ray images. Many Korean women have dense breasts, so combining mammography with ultrasound is known to be more effective for early detection.
The results of this large clinical trial showed that when radiologists read mammography images with AI, the chances of early detection increase. Lång said, "Greater adoption of AI in breast cancer screening programs can not only reduce radiologists' workload but also help in the early diagnosis of dangerous breast cancers."
Simon Vincent, chief scientist at the U.K. breast cancer charity Breast Cancer Now, also said, "This first clinical trial shows the enormous potential of AI to support radiologists in breast cancer screening."
However, Lång said the purpose of this study was to determine whether AI is as effective as standard screening, not to check whether it is superior. Additional clinical trials are needed to see if AI is better. Moreover, the team did not evaluate whether AI is more effective for specific ethnic groups.
Fiona Gilbert of the University of Cambridge in the U.K. said, "These clinical trial results are striking," adding, "Additional studies, including trials under way in the U.K., will help address this question." The U.K. National Health Service (NHS) began clinical trials last year to assess the effectiveness of AI in breast cancer screening.
References
The Lancet (2026), DOI: https://doi.org/10.1016/S0140-6736(25)02464-X
The Lancet Digital Health (2025), DOI: https://doi.org/10.1016/S2589-7500(24)00267-X
The Lancet Oncology (2023), DOI: https://doi.org/10.1016/S1470-2045(23)00298-X