An advanced artificial intelligence model developed by researchers at NYU Langone Health and its Perlmutter Cancer Center has demonstrated superior capability in predicting a woman's five-year risk of developing breast cancer. By evaluating past and recent annual three-dimensional mammograms taken over multiple years, known as longitudinal digital breast tomosynthesis, the deep-learning tool successfully surpassed traditional risk assessment frameworks and single-image artificial intelligence models.
The new study, published in the American Journal of Roentgenology, highlights how analyzing historical imaging data reveals structural tissue shifts that occur across successive screenings over time. This longitudinal approach marks a distinct departure from conventional evaluation methods that rely solely on single scans or patient medical histories, offering clinicians a more granular predictive mechanism for individual patient screening strategies.
Outperforming Traditional and Single-Scan Models
Researchers created the deep-learning tool, designated NYU-DRP, using a vast repository of 313,531 yearly three-dimensional mammograms collected from 161,165 women without breast cancer who underwent imaging at NYU Langone hospitals between 2016 and 2020. When evaluated against comparative testing models, NYU-DRP correctly ranked women who subsequently developed high-risk conditions 72 percent of the time. In contrast, single digital breast tomosynthesis and artificial intelligence-assisted two-dimensional testing achieved correct predictions in 70 percent and 68 percent of cases, respectively.
To establish a baseline against non-imaging methodologies, the research team also compared the tool's performance directly with the widely utilized Tyrer-Cuzick lifetime risk assessment. While the Tyrer-Cuzick method relies entirely on personal and family medical history, genetic mutations, and breast density without examining mammogram scans, NYU-DRP proved significantly more accurate. The artificial intelligence model correctly identified higher-risk cases after five years in 67 percent of matched comparisons, whereas the Tyrer-Cuzick framework accurately predicted the five-year outcome in 56 percent of cases.
Moving Beyond Breast Density Metrics
Among the study's key analytical insights was the finding that breast density alone fails to correspond reliably with a patient's predicted cancer risk. Although dense breast tissue is clinically recognized as a factor that heightens malignancy risk, the artificial intelligence model revealed substantial variance in actual outcomes across different tissue categories. Among patients classified with extremely dense breasts, the NYU-DRP model categorized 37.6 percent as average risk, whereas actual diagnosed cases after five years stood at a mere 0.7 percent.
Conversely, the algorithm identified 15.5 percent of patients with less-dense, fatty breast tissue as falling into high-risk categories, with actual five-year cases recorded at 2.5 percent within that cohort. "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," stated study senior investigator Yiqiu "Artie" Shen, Ph.D., an assistant professor in the Department of Radiology at NYU Grossman School of Medicine and Perlmutter Cancer Center.
Clinical Implications for Tailored Screening
The development of longitudinal artificial intelligence tools addresses a persistent limitation in diagnostic radiology, where uniform screening guidelines often fail to account for individual physiological trajectories. By evaluating sequential imaging rather than isolated snapshots, clinicians gain a refined mechanism to distinguish between patients who require supplemental diagnostic evaluations and those who can safely avoid unnecessary secondary procedures.
"Our study shows how AI models like NYU-DRP can be used to reliably determine a woman's future risk of breast cancer based on existing 3D mammograms, which hold information on how the breast tissue has changed across multiple screenings over time," said study lead investigator Yanqi Xu, Ph.D., a postdoctoral research fellow in the Department of Radiology at NYU Grossman School of Medicine and Perlmutter Cancer Center. The ability to parse subtle temporal changes in tissue composition provides a vital pathway toward personalized preventative care protocols.
Future Validation and Multi-Center Expansion
Following the retrospective conclusion of the study in 2025, during which less than 3 percent of the analyzed cohort developed breast cancer, the research team is preparing for broader validation phases. Future experimental initiatives will test the algorithm across diverse patient populations outside the NYU Langone network to evaluate its generalizability across different clinical environments and equipment manufacturers.
All imaging data evaluated in the initial study were acquired using breast imaging systems manufactured by Hologic Inc., located in Marlborough, Massachusetts. Co-investigator Laura Heacock, M.D., an associate professor of radiology, emphasized that successful external validation could allow physicians to "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."
Next steps for the research group involve deploying the longitudinal program prospectively to track patient breast health in real time, observe longitudinal health outcomes, and cross-reference data streams with external academic health centers utilizing varied imaging hardware.
