AI Uses Past Mammograms To Predict Breast Cancer Risk

Researchers at NYU Langone Health have developed an artificial intelligence tool that predicts a woman’s five-year risk of developing breast cancer by analyzing a series of annual 3D mammograms. The study, reported by GN technics/ai (en-US), indicates that this approach outperforms methods relying on a single recent scan or traditional risk assessment tools that use genetic and family history data.

The model, known as NYU-DRP, correctly identified women at higher risk 72 percent of the time, compared to 70 percent for single-scan AI and 68 percent for 2D mammogram analysis. It also surpassed the widely used Tyrer-Cuzick risk assessment, which achieved a 56 percent accuracy rate. This improvement suggests that tracking changes in breast tissue over time provides a clearer picture of future health risks than static data points.

Longitudinal Data Improves Prediction Accuracy

The AI was trained on over 313,000 mammograms from more than 161,000 women who did not have breast cancer at the time of testing. By examining how breast tissue changes across multiple years, the system captures dynamic patterns that single images miss. This longitudinal perspective allows the tool to distinguish between normal variations and potential early signs of disease, offering a more nuanced view of individual risk.

The findings challenge the traditional assumption that breast density is the primary indicator of cancer risk. The study found that breast density alone did not always correspond to predicted risk levels. For instance, many women with extremely dense breasts were classified as having average risk, while some with less dense, fatty breasts were identified as high risk. This suggests that tissue evolution over time is a more reliable metric than density at a single point in time.

Tailoring Screening To Individual Needs

The practical implication of this technology is a more personalized screening strategy. Physicians could use the AI’s risk predictions to determine who needs additional testing, such as MRI or ultrasound, and who can safely continue with standard mammograms. This approach aims to reduce unnecessary procedures for low-risk women while ensuring high-risk individuals receive closer monitoring.

However, the tool is not yet a standalone diagnostic. It is designed to assist physicians in making informed decisions about screening frequency and intensity. The researchers emphasize that this is a predictive tool, not a definitive cancer detector. Future studies will track how these predictions hold up over time and whether they successfully guide patients toward appropriate care without causing undue anxiety or over-testing.

Limitations And Future Validation

The study had some constraints, including the fact that fewer than 3 percent of the participants developed breast cancer during the observation period. This low incidence rate makes it difficult to assess the tool’s performance in high-risk populations. Additionally, the model was developed using data from a specific hospital system, so its effectiveness in diverse populations with different healthcare access and demographics remains to be fully tested.

Before this technology becomes widely available, it must undergo rigorous validation in independent studies. Researchers plan to use the tool to proactively track breast health and observe its impact on patient outcomes. Until then, standard screening guidelines remain the primary recommendation for most women. The promise of AI-assisted screening is significant, but it requires careful integration into existing medical practices to ensure safety and efficacy.