Lunit – Published on November 3, 2025
For decades, progress in breast cancer prevention has been built on careful observation, modeling, and population-level programs. Yet despite these advances, our ability to predict an individual woman’s risk remains limited – constrained by self-reported data, incomplete family histories, and research that doesn’t always represent the full diversity of women we serve. Today, artificial intelligence offers a way to change that.
By Graham A. Colditz, MD, DrPH
Co-developer of Prognosia Breast
Image-based risk (IBR) modeling uses the mammogram itself and a woman’s age to estimate her likelihood of developing breast cancer within five years. It’s a breakthrough that could make precision prevention accessible to every woman who undergoes screening – no extra tests, no added effort.
When my colleague, Professor Shu (Joy) Jiang, and I began developing the Prognosia Breast model – recently recognized by the U.S. FDA with Breakthrough Device Designation – our goal was simple: to uncover subtle image-based patterns invisible to the human eye that predict future cancer development.
By deriving risk directly from the mammogram, we reduce the need for questionnaires or genetic testing. This not only simplifies preventive care but also extends its reach, particularly to women historically underserved by traditional risk tools.
In extensive validation studies123 across multi-ethnic cohorts, image-based risk outperformed established clinical models such as Tyrer-Cuzick and Gail. It demonstrated stronger accuracy and calibration for Black, Asian, and Indigenous women – a crucial finding that underscores a broader public-health truth: when we can measure risk equitably, we can intervene earlier and save more lives.
Of course, translating research innovation into clinical reality takes more than strong algorithms. It requires a partner capable of scaling responsibly. Lunit was the natural choice to bring Prognosia Breast to the world – combining deep expertise in medical AI regulation with a proven global footprint and complementary technologies spanning cancer detection, image quality analytics, and risk management. Together, we can deliver image-based risk through the same infrastructure radiologists already trust – embedding predictive power seamlessly into existing workflows.
Over the next one to two years, I expect to see image-based risk transition from validation to real-world use. Prospective trials are already underway to assess how the model performs in clinical settings and how it impacts workflow efficiency, patient engagement, and screening outcomes. Once regulatory clearance is achieved, radiologists will be able to stratify risk at the point of screening – identifying women who may benefit from supplemental imaging, closer monitoring, or preventive therapy long before disease develops.
For this innovation to succeed, three principles must guide adoption:
The convergence of imaging, epidemiology, and AI represents the most promising step forward since the arrival of digital mammography itself. With image-based risk, we have the opportunity to fulfill a long-held vision of personalized screening – turning every mammogram into a predictive tool for prevention, equity, and early intervention.
*Disclaimer: This product is currently under development and has received Breakthrough Device Designation by the US FDA. It has not been cleared or approved for commercial use.
References:
1 Jiang S et al. Validation of a Dynamic Risk Prediction Model Incorporating Prior Mammograms in a Diverse Population. JAMA Netw Open. 2025;8(6):e2512681
2 Jiang S et al. Development and Validation of Dynamic 5-Year Breast Cancer Risk Model Using Repeated Mammograms. JCO Clin Cancer Inform. 2024;8:e2400200
3 Jiang S et al. Deriving a Mammogram-Based Risk Score from Screening Digital Breast Tomosynthesis for 5-Year Breast Cancer Risk Prediction. Cancer Prev Res (Phila). 2025;18(6):347–354
