NYU-DRP AI Model Predicts Five-Year Breast Cancer Risk From 3D Mammograms

NYU Langone Health announced on September 10, 2026 that researchers at NYU Langone Health and its Perlmutter Cancer Center have developed a deep-learning tool, called NYU-DRP, that estimates a woman’s five-year risk of developing breast cancer by analyzing her 3D mammograms taken over multiple years.

A Deep Learning Model Built on Longitudinal 3D Mammograms

The tool analyzes longitudinal digital breast tomosynthesis, or longitudinal DBT, the 3D mammogram record a woman accumulates across repeated annual screenings. The validation study, titled “Predicting 5-Year Breast Cancer Risk From Longitudinal Digital Breast Tomosynthesis: A Single-Center Retrospective Study,” was published in the American Journal of Roentgenology online on August 12, 2026, after acceptance on August 3, 2026, and carries the DOI 10.2214/AJR.26.34951.

The paper states that its objective was to develop and evaluate a deep learning model that uses longitudinal DBT examinations to predict long-term breast cancer risk. Its authors note that imaging-based breast cancer risk prediction models have primarily used full-field digital mammography, or FFDM, and that although DBT has become a predominant screening modality in the United States, its potential for long-term risk prediction remains underexplored.

“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, PhD, a postdoctoral research fellow in the Department of Radiology at NYU Grossman School of Medicine and Perlmutter Cancer Center.

Retrospective Cohort and Reported Results

According to the published study, the retrospective analysis covered 313,335 DBT examinations from 161,077 women with a mean age of 58.5 years, imaged between January 2016 and August 2020 at a single health institution. The NYU Langone announcement describes NYU-DRP as created from 313,531 yearly 3D mammograms from 161,165 women without breast cancer who had tests performed at NYU Langone hospitals between 2016 and 2020.

The DRP model estimates two- to five-year breast cancer risk from longitudinal DBT examinations, patient age, and breast density. The researchers compared it with a single-timepoint DBT model, with the Mirai model using same-day FFDM, and with the Tyrer-Cuzick clinical risk model, measuring performance with the AUC, time-dependent concordance index, and integrated Brier score metrics.

In an independent test set of 34,570 examinations, the paper reports that the longitudinal DRP model achieved a five-year AUC of 0.721 (95% CI, 0.698–0.744), improving on the single-timepoint model at 0.707 (95% CI, 0.683–0.730) and the Mirai model at 0.687 (95% CI, 0.663–0.710), both at p <.001. The announcement characterizes those results as correctly ranking women at higher risk 72 percent, 70 percent, and 68 percent of the time, respectively.

In a matched case-control cohort of 432 women, the paper reports a five-year AUC of 0.676 (95% CI, 0.626–0.726) for the DRP model against 0.563 (95% CI, 0.509–0.619; p <.001) for the Tyrer-Cuzick model, a comparison the announcement reports as 67 percent versus 56 percent. According to the release, Tyrer-Cuzick involves no AI or mammogram scans and instead relies on personal and family medical information, such as age, genetic mutations, and breast density backed by biopsy results; the comparison covered 432 women, with half matched against women of similar age and background whose five-year outcomes differed.

The study further reports that breast density alone did not correspond to predicted risk. Among examinations of women with extremely dense breasts, the model classified 39.7 percent (746 of 1,877) as average risk, with an observed five-year cancer incidence of 0.8 percent (6 of 746); among examinations of women with fatty breasts, it classified 14.8 percent (386 of 2,605) as high risk, with an observed incidence of 2.6 percent (10 of 386). The announcement, which notes that dense breast tissue is known to heighten cancer risk, reports these shares as 37.6 percent and 0.7 percent, and as 15.5 percent and 2.5 percent, respectively.

Study senior investigator Yiqiu “Artie” Shen, PhD, an assistant professor in the Department of Radiology at NYU Grossman School of Medicine and Perlmutter Cancer Center, said the 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. The researchers noted that less than 3 percent of women tested during the study, which ended in 2025, developed breast cancer, and that all testing used breast imaging equipment manufactured by Hologic Inc. of Marlborough, Massachusetts; the paper identifies the design as a single-center retrospective study.

The authors conclude that a deep learning model using longitudinal DBT examinations improved long-term breast cancer risk prediction compared with FFDM-based and clinical risk models, and they state the clinical impact as the potential to inform dynamic risk assessment using screening images and to support future personalized screening strategies.

Planned Validation, Funding, and Screening Context

“If future experiments in other women with breast cancer prove successful, then AI-assisted 3D mammograms like 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,” said study co-investigator Laura Heacock, MD, an associate professor in the Department of Radiology at NYU Grossman School of Medicine and Perlmutter Cancer Center.

Shen said the team next plans 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. The researchers also plan to share and cross-check the tool with data from other academic health centers and with data from different 3D mammogram manufacturers.

Funding support for the study came from National Science Foundation grant 1922658, National Institutes of Health grant R01EB036530, the Milstein Pilot Project Fund, the Shifrin-Myers Breast Cancer Discovery Fund, and the Manhasset Women’s Coalition Against Breast Cancer. Besides Xu, Shen, and Heacock, the study’s co-investigators were Jungkyu Park, Felicia Pasadyn, Qi Lei, Alana Lewin, Krzysztof Geras, Linda Moy, and Freya Schnabel, and the paper lists Shen as corresponding author.

The announcement places the study against US screening statistics: an estimated 1 in 8 women in the United States will be diagnosed with breast cancer in their lifetime, with approximately 382,640 women expected to be diagnosed in 2026. The five-year relative survival rate exceeds 99 percent when breast cancer is caught in its earliest stages, more than 4 million breast cancer survivors currently live in the country, more than 43 million mammograms were performed in 2025, and experts now recommend annual mammography screening beginning at age 40.