NYU Langone Health's AI Tool Enhances Personalized Breast Cancer Screening

A recent study from NYU Langone Health highlights the potential of an artificial intelligence (AI) tool, NYU-DRP, to improve breast cancer risk prediction by utilizing longitudinal digital breast tomosynthesis (DBT). The research, published in the American Journal of Roentgenology, demonstrates that this deep-learning model is more effective at forecasting breast cancer risk over five years than existing methods.

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. Leveraging multiple years of mammogram data, the AI model more accurately ranked women's cancer risk, correctly identifying those at higher risk 72% of the time. This surpasses the accuracy of single 3D mammogram evaluations and AI models using 2D mammograms, which scored 70% and 68%, respectively.

The study also compared NYU-DRP with the Tyrer-Cuzick risk assessment, a widely used tool not based on AI. The AI model outperformed the Tyrer-Cuzick assessment, predicting higher risk with 67% accuracy after five years, compared to 56% for the traditional method. This underscores the AI tool's ability to incorporate data from 3D mammograms, revealing changes in breast tissue over time that are not considered by Tyrer-Cuzick.

Key insights from the study indicate that breast density alone is not a definitive predictor of breast cancer risk, although it is a recognized risk factor. 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.

The team at NYU Langone is continuing to refine this tool with plans to validate its effectiveness with data from other health centers and additional mammogram manufacturers. The goal is to offer more tailored breast cancer screening approaches, potentially reducing unnecessary procedures for low-risk individuals and enhancing surveillance for those at higher risk.