NYU Langone AI Uses Multiple 3D Mammograms to Predict Breast Cancer Risk

NYU Langone Health researchers developed NYU-DRP, an AI tool that estimates five year breast cancer risk from multiple years of 3D mammograms, the health system said in a press release. The study was published online in the American Journal of Roentgenology on Aug. 12.

Researchers developed the model using 313,531 annual 3D mammograms from 161,165 women without breast cancer who received screening from 2016 to 2020. NYU-DRP correctly ranked women by five year risk 72% of the time, compared with 70% for a model using one 3D mammogram and 68% for a model using 2D mammograms. Against the Tyrer-Cuzick risk assessment, run on 432 women, it scored 67% to 56%. The margin over a single 3D mammogram is two percentage points, so most of the separation is from the older statistical model rather than from reading images repeatedly.

The more striking result concerns breast density, which currently drives supplemental screening decisions in many places. Density alone did not correspond with predicted risk. NYU-DRP classified 37.6% of women with extremely dense breasts as average risk, and 0.7% of that group developed cancer within five years. It classified 15.5% of women with less dense, fatty breasts as high risk, and 2.5% of that group did.

The researchers said repeated 3D mammograms carry information about future risk that density or a single scan does not capture, and that the tool would need validation in future studies, using data from other health centers and other mammogram equipment manufacturers, before it could be used to tailor screening.