AI detects early breast cancer signs up to six years before Dx

Clinical takeaway: No practice change yet, but applying AI detection approaches across serial mammograms may eventually help identify women who warrant closer surveillance.

Mammograms read by machine learning models already turn up cancers that radiologists miss. Less clear is whether the same tools, applied to images taken years earlier, can carry a usable early-warning signal. This study tested three commercial systems against that question across a decade of Swedish screening data.

Applied retrospectively to stored mammograms, all three systems produced scores that climbed steadily in the years before a cancer was diagnosed, while scores in women who remained cancer-free held flat. At a 90% specificity threshold, the systems flagged 19.0% to 19.7% of future cancers six years before diagnosis, 23.3% to 25.2% at four years, and 35.4% to 39.3% at two years. The steepest rise came in the final four years before diagnosis.

Across all early time points combined, the systems separated future cancers from cancer-free screens with an area under the curve of 0.63 to 0.67. That’s modestly better than mammographic density alone at 0.57. The signal was real but far from definitive. Individual score trajectories were noisy: some women crossed the threshold and dropped back below it across serial screens.

The analysis drew on the Validation of Artificial Intelligence for Breast Imaging (VAI-B) database, covering four Swedish regions from 2008 to 2019, with 88,963 mammograms from 31,394 women, each read by two radiologists under the national biennial program. The three systems were built for cancer detection on a single image, not risk prediction, and were applied here to earlier screens to test their early-warning potential.

The authors suggest AI scores might one day be combined with breast density to decide who is offered supplemental imaging, and that requiring a sustained score change across two screens could limit false alarms before any such approach reaches the clinic.

"Approximately 20% of breast cancer cases demonstrate mammographic signs that are already visible to AI around 6 years before diagnosis," said senior coauthor Fredrik Strand, MD, PhD, of Karolinska University Hospital in Stockholm. "Our study confirms the potential of AI to, in some cases, find signs of cancer in the mammograms much earlier than when radiologists detected it."

Source: Hickman SE. Radiology. 2026 Jun 9. Artificial Intelligence Detection Scores in Screening Mammography for Early Breast Cancer Alerts