An AI model assessing digital mammograms can accurately predict the risk of breast cancer in a large and diverse population of women, according to a new study by WashU Medicine researchers and published in JAMA Network Open.
The study—led by Shu (Joy) Jiang, PhD, an associate professor in the Division of Public Health Sciences in the Department of Surgery—included over 200,000 women between the ages of 40 to 70 years old who participated in the British Columbia Breast Screening Program and had at least one screening digital mammogram. The women were from a wide-range of racial and ethnic backgrounds, including East Asian, South Asian, Indigenous, and non-Hispanic White.
Following women for an average of five years, the study found that the AI model generated accurate risk scores based on women’s prior mammograms – and that performance was better when using up to four previous mammograms compared to a single mammogram. Accuracy of the score was also largely the same across the different racial and ethnic groups – and was better than that of many other breast cancer risk calculators using standard risk factor questionnaires or AI assessment of single mammograms.
Accurate estimates of breast cancer risk can help guide screening and prevention efforts that are personalized for individual women.
“Because this model is calibrated to match national guidelines, women with a 3% or higher risk of developing breast cancer over the next five years are recommended for more frequent breast imaging and should consider endocrine therapy, such as Tamoxifen, to reduce breast cancer risk,” says the study’s senior author, Graham Colditz, MD, DrPH, associate director of prevention and control at Siteman Cancer Center and chief of the Division of Public Health Sciences.
These findings build on previous research demonstrating the accuracy of the same AI model in Black and White women.
“While this is not yet FDA approved, it is the first model that looks at change in mammograms and is rigorously externally validated across North America and in a diverse population of women,” adds Colditz.
“That the model assesses mammograms done every two years based on Canadian recommendations—as opposed to annual screening in the United States—adds to its global applicability as well.”
The study is published June 6, 2025 in JAMA Network Open.