AARHUS, Denmark — Denmark will become the first country to use artificial intelligence as a standard reader in its national breast cancer screening program, the health ministry said on Wednesday, replacing one of the two radiologists who have examined every mammogram since screening began with software that a three-year trial found detected more cancers at a lower cost.
Under the new protocol, which takes effect across all five Danish regions in January, each of the roughly 420,000 mammograms taken each year will be read by an A.I. system and one radiologist, rather than two radiologists. Images the software flags as high-risk will be reviewed by two doctors, and those it rates as clearly normal will receive a single abbreviated review. The trial, conducted in the Central and North Jutland regions from 2023, found that the approach detected 11 percent more cancers than double human reading, reduced the rate of women recalled unnecessarily by 6 percent and cut radiologists’ screening workload by about 30 percent.
“We did not do this because the software is clever. We did it because we are short of 60 breast radiologists and will be short of 100 by 2030,” said Dr. Mette Lindqvist, who leads the screening program at Aarhus University Hospital and directed the trial. Denmark, like most of Europe, requires two independent readers for every screening mammogram, a safeguard that has become increasingly hard to staff. “The choice was never A.I. or two doctors. It was A.I. or longer intervals between screens.”
The additional cancers found were mostly small invasive tumors, the kind that are most treatable, and the trial’s interval cancer rate, tumors found between screens that the screening missed, fell by 9 percent. The system, developed by a Danish company and trained on more than 2 million European mammograms, assigns each image a score from 1 to 10; in the trial, no cancer was found among images it scored 1, which accounted for 28 percent of the total.
Prof. Erik Brandt, a radiologist at the University of Copenhagen who sat on the ministry’s advisory committee, supported the decision but warned against reading too much into it. “The trial was run by motivated people in two regions with excellent image quality,” he said. “The question now is whether the result holds in year five, when the machines have been updated three times and the radiologist who reads alongside it has learned to trust it a bit too much.” The protocol requires that 5 percent of images the software rates normal be audited by two doctors, and that the vendor cannot update the model without a new validation.
Patients’ groups broadly welcomed the change. Anne-Sofie Kjær, 54, a teacher in Randers whose cancer was found in the trial after, she later learned, the A.I. flagged an image one radiologist had cleared, said she had no objection to a computer reading her scan. “I never met the two doctors either,” she said. “What I care about is that somebody, or something, saw it.”
Sweden, Norway and the Netherlands are running similar trials, and Britain’s health service is evaluating A.I. readers at 30 sites. Dr. Lindqvist said she expected most of Europe to follow within five years. “Every country is asking the same question,” she said, “and they all have the same number of radiologists.”
A version of this article appears in print on Oct. 12, 2026, Section A, Page 11 of the European edition with the headline: Denmark Becomes the First Country to Make A.I. a Standard Reader in Breast Cancer Screening.
