New Study Explores the Role of AI in Tracking Breast Density

In breast cancer screening, consistency is key to tracking changes over time, especially when monitoring breast density. Dense breast tissue makes it harder to detect cancer on a mammogram and is itself a risk factor for breast cancer. Adding to the challenge, different radiologists may assess the same woman’s breast density differently.

WashU Medicine Mallinckrodt Institute of Radiology researchers and collaborators used an artificial intelligence model in one of the largest studies to date on AI and breast density. Aimilia Gastounioti, PhD, an assistant professor of radiology and principal investigator in MIR’s Computational Imaging Research Center, is the study’s senior author. “Our research team had representation from real-world practice, industry and academia, which allowed us to pursue this exciting and clinically relevant question,” Gastounioti said.

The team analyzed over 214,000 mammograms from more than 61,000 women across 50 Onsite Women’s Health outpatient radiology sites in the U.S. using a commercially available breast density AI model made in collaboration with Whiterabbit.ai. They compared the performance of the AI model to the interpretation of 39 radiologists, assessing how consistently each could classify breast density over time. They found that AI provided more longitudinally consistent breast density assessments compared to the interpreting radiologists, with the AI model classifying breast density consistently in 81% of cases, compared to 57% for radiologists. The model was also far less likely to flip-flop between different density categories, a trend that could cause confusion or unnecessary follow-ups for patients.

The results demonstrate the potential advantages of AI in breast density evaluation and light a path towards improved downstream care for patients.

The study’s results are published in BJR Artificial Intelligence.