Artificial intelligence improves mammography-based risk prediction

“We discuss recent advances in artificial intelligence-assisted breast cancer risk prediction, what this means for the future of breast cancer screening and prevention, and the critical research needed to translate mammographic signatures from research into clinical practice,” senior study author Erik Thompson said.

Breast tissue that appears white on a mammogram is radiologically dense, while breast tissue that appears black is considered non-dense. It is generally believed that women with denser mammograms have a higher risk of breast cancer compared with age and body mass index. Additionally, higher density makes it harder for mammograms to detect breast cancer, which is known as the “masking effect.”

Following policy changes in the United States, Canada and Australia, advocacy campaigns around the world are asking women to know their mammogram density. Mammography density screening is guiding the use of complementary imaging technologies in some places, with ultrasound and magnetic resonance imaging (MRI) improving cancer detection rates in clinical studies of women with very dense breasts. However, scientists and clinicians are still grappling with the complexities posed by masking effects, the risk of breast cancer associated with mammographic density, and how to best implement changes in clinical practice.

To predict future breast cancer diagnoses, advanced computational methods such as deep learning are now used to analyze mammograms. In particular, artificial intelligence methods are revealing mammographic signatures that may predict breast cancer risk better than any other known risk factor. These characteristics likely explain much of the association between mammographic density and breast cancer risk. The discovery of risk-predictive mammographic signatures generated by artificial intelligence provides new opportunities to identify women most likely to develop breast cancer in the future and distinguish them from those most likely to miss breast cancer risk due to masking effects .

“Women with mammogram characteristics associated with a higher risk of breast cancer detection may benefit from more frequent screening or risk-reducing medications,” Thompson said. “On the other hand, women who are diagnosed with breast cancer within the next five years could benefit from more frequent screening tests or risk-reducing medications.” Women with lower odds are offered longer screening intervals. In addition, women with high mammographic density but without high-risk mammographic characteristics may benefit from supplemental imaging such as MRI or ultrasound.

Research shows that some AI-generated mammogram features indicate early-stage malignancies that cannot be detected on mammograms read by radiologists, while other features may be benign conditions associated with increased breast cancer risk. The identity of AI-generated mammographic features that are not identified as cancer or benign disease remains unclear.

“It is critical that we identify the pathologies associated with mammographic features and the underlying mechanisms linking them to breast cancer tumorigenesis,” Thompson said. “This will be critical to determining their association with short- and long-term breast cancer risk and future efforts to reduce that risk.”