AI Reads Years of Mammograms to Predict Breast Cancer Risk Half a Decade Early

The Clues Were Already There

Every year, millions of women go in for their routine mammogram, get the all-clear, and move on. But what if those scans — stacked year after year — were quietly telling a more complex story? That's the premise behind a breakthrough AI tool developed at NYU Langone Health, and the results suggest that the future of breast cancer detection may already be sitting in radiology archives.

Researchers at NYU Langone Health and its Perlmutter Cancer Center built a deep-learning tool called NYU-DRP that analyzes women's past and recent annual 3D mammograms — a technique known as longitudinal digital breast tomosynthesis — to predict the five-year risk of developing breast cancer. The key word is "longitudinal." Rather than looking at a single snapshot, the AI reads how breast tissue changes over time, detecting subtle shifts that no single scan could reveal on its own.

How It Stacks Up Against Existing Tools

NYU-DRP was developed using 313,531 3D mammograms from 161,165 women without breast cancer, collected at NYU Langone hospitals between 2016 and 2020. That's a massive dataset — and the model learned from all of it. Published in the *American Journal of Roentgenology* in August, the study showed that NYU-DRP correctly ranked women at higher risk 72 percent of the time. Single 3D mammogram evaluations and AI-assisted 2D testing correctly predicted higher-risk cases 70 percent and 68 percent of the time, respectively.

The gap widens even further when compared to traditional, non-imaging methods. When researchers compared NYU-DRP with the Tyrer-Cuzick risk assessment — a widely used tool that estimates breast cancer risk based on personal and family medical history — NYU-DRP correctly predicted who was at higher risk after five years 67 percent of the time, while Tyrer-Cuzick accurately predicted five-year risk just 56 percent of the time. That's a meaningful difference when the stakes are a cancer diagnosis.

What the AI Is Actually Seeing

"Our findings demonstrate that repeated 3D mammograms contain information about a woman's future breast cancer risk that is not fully captured by either breast density or a single mammogram on its own," said study senior investigator Yiqiu Shen, PhD, an assistant professor in the Department of Radiology at NYU Grossman School of Medicine. That distinction matters enormously, because breast density has long been used as a rough proxy for cancer risk — and it turns out that proxy has real limits.

The AI model discerned that a significant percentage of women with dense breasts did not correspond to high actual risk, while a notable portion of women with less dense breast tissue was identified as higher risk. In other words, the tool is picking up on something deeper and more individualized than density alone. As Shen explained, "everybody's breast tissue is somewhat unique," and AI can "utilize this unique signal to come up with a personalized risk score."

What Could Change for Patients

Study co-investigator Laura Heacock, MD, said that if future experiments prove successful, NYU-DRP "could help physicians better tailor screening to a woman's actual risk, by identifying those women who may benefit from additional screening while avoiding unnecessary supplemental tests for those at lower risk." That kind of precision matters both medically and practically — fewer unnecessary procedures, less anxiety, and more targeted care for those who truly need it.

The tool isn't ready for clinical use just yet. The study used mammograms from NYU Langone hospitals and imaging equipment from one manufacturer, and the team plans to test the model using data from other medical centers and different 3D mammography systems. Researchers also plan to use the longitudinal DBT program to track women's breast health proactively, observing how the tool impacts their health and who develops breast cancer. If those broader tests hold up, the routine annual mammogram — long seen as a simple screening checkbox — could become one of the most powerful personalized risk tools in medicine.