NCI-funded researchers found that several newly developed artificial intelligence (AI) algorithms can predict a person’s five-year breast cancer risk better than a standard clinical risk model.

“We need to have more accurate and efficient ways to assess a woman’s future breast cancer risk,” said study lead author Vignesh Arasu, M.D., Ph.D., a research scientist at the Kaiser Permanente Division of Research and radiologist specializing in breast imaging with The Permanente Medical Group (TPMG). “Currently, only 20% of women who go on to develop breast cancer have identifiable risk factors.”

The retrospective study included 324,000 women who had a mammogram that found no sign of breast cancer in 2016 at Kaiser Permanente Northern California. Over the next five years, 4,584 of the women were diagnosed with breast cancer. The study compared these women with 13,435 of the 324,000 women in the original group who did not develop breast cancer.

The researchers had five AI algorithms report a five-year breast cancer risk prediction score from the mammography images made for these women in 2016. These scores were compared with Breast Cancer Surveillance Consortium (BCSC) risk scores. Risk estimates were calculated for incidences of breast cancer in the 0-5 years after the initial mammography examination.

When using the negative screening examination, AI algorithms performed better than the BCSC risk model for predicting breast cancer risk at 0 to 5 years. It also predicted missed or aggressive interval cancers.

The results of the study show that mammography AI algorithms may improve breast cancer risk prediction compared to current clinical risk models. “When combining AI with known traditional risk factors, we can now identify risk for approximately 60% of women who go on to develop breast cancer compared with 20% by using traditional risk factors alone, substantially improving our ability to better identify those at risk.”

This project is funded through an NIH Research Project Grant (R01 CA264987), NCI’s Quantitative Imaging Tools and Methods for Cancer Response Assessment grant (U01 CA225427), and the TPMG Delivery Science and Applied Research Physician Researcher Program.