AI Tops Density in Predicting Breast Cancer Risk

Senior author Constance D. Lehman, M.D., Ph.D., professor of radiology at Harvard Medical School in Boston, Massachusetts, said traditional methods of assessing a woman’s risk for breast cancer, including age, family history, genetics and breast density, are inadequate. 

“Over two million women are diagnosed with breast cancer annually, and for most, it comes as a complete shock,” she said. “Only 5 to 10% of breast cancer cases are considered hereditary, and breast density alone is a very weak predictor of risk.” 

Clairity Breast, the first FDA-authorized image-only AI breast cancer risk model, was trained on 421,499 mammograms from 27 facilities in Europe, South America and the U.S. Using mammograms both from women who developed cancer and women who did not develop cancer over the subsequent five years helped the AI model to learn the patterns and differences in breast tissue that predict cancer risk. The model was calibrated on an independent test set using a deep convolutional neural network to generate five-year risk probabilities. 

“The model is able to detect changes in the breast tissue that the human eye can’t see,” Dr. Lehman said. “This is a job that radiologists just can’t perform. It’s a separate task from detection and diagnosis, and it will open a whole new field of medicine, leveraging the power of AI and untapped information in the image.

The model was applied to a study group of 236,422 bilateral 2D screening mammograms 
from five U.S. sites and 8,810 from one European site. The mammograms were acquired between 2011 and 2017. Radiologist-reported breast density (dense versus not dense) and five-year cancer outcomes were extracted from medical records and tumor registries, respectively. AI-predicted risks were categorized using National Comprehensive Cancer Network thresholds: average (less than 1.7%), intermediate (1.7-3.0%) and high (greater than 3.0%). 

The researchers compared the risk categories using statistical models that account for follow-up time and censoring. 

Accounting for breast density, women in the high-risk AI group had more than a fourfold higher cancer incidence than women in the average-risk group (5.9% vs. 1.3%). By contrast, breast density alone showed only modest separation (3.2% for dense vs. 2.7% for non-dense). 

“The results of this large-scale analysis demonstrate that AI risk models provide far stronger and more precise risk stratification for five-year cancer prediction than breast density alone,” said first author and presenter Christiane Kuhl, M.D., Ph.D., director, Department of Diagnostic and Interventional Radiology at University Hospital RWTH Aachen, in Germany. “Our findings support the use of image-only AI as a complement to traditional markers supporting a more personalized approach to screening.”

The American Cancer Society currently recommends that women at average risk have the option to start annual breast cancer screening with mammography at age 40. However, women under 40 are the fastest-growing group being diagnosed with breast cancer and advanced disease. 

“An AI image-based risk score can help us identify high-risk women more accurately than traditional methods and determine who may need screening at an earlier age,” Dr. Lehman said. “We already screen some women in their 30s when they are clearly at high risk based on family history or genetics. In the future, a baseline mammogram at 30 could allow women with a high image-based risk score to join that earlier, more effective screening pathway.”

Breast density legislation enacted in 32 states requires healthcare providers to inform women undergoing a screening mammogram of their breast density. 

“We’d like to see women given information on their breast density and their AI image-based risk score,” Dr. Lehman said. “We can do better than just looking at a mammogram and saying, ‘It is dense or not dense’ to inform women of their risk.”