AI Innovations in Mammogram Risk Scores Aid Future Breast Cancer Prediction
Recent advancements in artificial intelligence (AI) have enabled researchers to generate dynamic breast cancer risk scores from screening mammograms. This breakthrough, detailed in a study published in Radiology, suggests that changes in these AI-derived scores over time can effectively predict future breast cancer, presenting new opportunities for personalized risk assessment and prevention.
Using deep learning models, the study derived risk scores by analyzing entire mammogram images, rather than focusing solely on predetermined features like breast density. These models surpassed traditional methods in estimating a woman’s five-year risk of developing breast cancer. The study encompassed data from over 239,700 mammograms and 54,014 women, spanning 2009 to 2019, across various imaging sites.
The researchers, led by Dr. Constance D. Lehman from Harvard Medical School, evaluated longitudinal changes in risk scores using serial mammograms. The team discovered that women who eventually developed breast cancer exhibited progressively increasing risk scores over the years preceding a diagnosis. In contrast, scores for women who remained cancer-free showed little change, indicating stable risk levels.
The study revealed that risk scores for cancer patients increased from a median of 2.1 to 6.6 in the years ahead of diagnosis, highlighting a distinct upward trajectory compared to stable scores in those without cancer. This trend was observed regardless of age or breast density, underscoring the models' robust applicability across diverse patient subgroups.
Dr. Lehman noted the significance of these findings, emphasizing how AI-derived risk scores detected signals invisible to the human eye. Most notably, these scores alerted to potential risks as early as six years before a cancer diagnosis. This capability is crucial, as the majority of breast cancer cases are sporadic and not due to familial or genetic factors.
Beyond individual patient implications, the study suggests that dynamic risk assessment through imaging data could help mitigate disparities in screening performance across different populations. By offering a more personalized risk-based approach, AI in mammography may soon guide additional preventive strategies, such as managing conditions like high cholesterol or hypertension.
The incorporation of AI-based risk assessments into clinical practice is already underway, with models included in the 2026 National Comprehensive Cancer Network guidelines. These models propose that women aged 35 and older, identified as having an elevated five-year risk, consider supplemental breast MRI screenings alongside regular mammography.