Breast cancer is the second leading cause of cancer death in women in the United States, but only about 15% to 20% of women have known risk factors, according to Vignesh A. Arasu, MD, PhD, a radiologist who specializes in breast cancer. “These traditional risk factors include a woman’s age, family history, prior benign biopsies, estrogen exposure, and breast density,” he says. “We have known for decades that a single imaging feature assessed on a mammogram—breast density—increases long-term risk. Today, artificial intelligence (AI) has allowed us to identify hundreds of new features on a mammogram beyond breast density.”

Mammography AI, Dr. Arasu notes, is mainly used to identify visible breast cancer. “But as a clinical radiologist and researcher, I knew that alongside identifying cancers, AI could also likely identify features of future risk, beyond the single factor of breast density,” Dr. Arasu says. “My colleagues and I thought it would be interesting to see how well these AI algorithms assess a woman’s future breast cancer risk compared with a standard risk model that incorporates these traditional risk factors.”

AI Algorithms Can “See” Imaging Biomarkers on Mammograms

For a study published in Radiology, Dr. Arasu and colleagues conducted a retrospective case-cohort study of women with a negative screening mammographic examination in 2016 and were followed until 2021. “We identified 324,000 women who had a mammogram in 2016 at Kaiser Permanente Northern California with no sign of breast cancer,” Dr. Arasu notes. “We then determined which women developed breast cancer between 2016 and 2021. Our team found 4,584 women with a breast cancer diagnosis. These women were then compared with a subgroup that included 13,435 of the 324,000 women who did not develop breast cancer.”

The study team evaluated five AI algorithms and generated a score for these negative mammograms from 2016. “These scores are intended for breast cancer detection, but we were studying whether they could predict future cancer risk out to 5 years,” he says. “We also used the Breast Cancer Surveillance Consortium (BCSC) clinical risk model to assess the women’s breast cancer risk based on their traditional risk factors from 2016. Finally, we evaluated whether AI or the BCSC had done a better job at predicting which women would have a breast cancer diagnosis.”

Dr. Arasu and colleagues observed that AI algorithms can “see” imaging biomarkers on mammograms, in addition to breast density. “This tells us there are changes happening in the breast tissue that can be used to predict risk,” Dr. Arasu notes. “This suggests that AI used alone or combined with current risk prediction models provides a new avenue for future risk prediction (Table).”

Going Beyond Early Detection of Breast Cancer to Risk Prediction

Most AI algorithms trained to read mammograms are used to flag a visible cancer to assist radiologists, Dr. Arasu explains. “This study suggests we can expand how we are using AI and mammograms to go beyond early detection to risk prediction,” he says. “Only 20% of women have a known breast cancer risk factor that puts them at higher risk for developing breast cancer. If we use mammography AI scores along with these traditional risk factors, we can potentially identify up to 60% of women who are at higher risk for developing breast cancer.”

This information, he continues, could potentially be used to develop personalized screening recommendations. For some women with high risk, this might mean more frequent screening, with the goal of identifying a cancer as early as possible, when treatment is more likely to be effective.

In the future, Dr. Arasu would like to see whether the accuracy of AI algorithms can be improved. “We also need future research to determine the best way to use the information clinically,” he adds.