1. Longitudinal changes in deep learning-estimated breast density and their impact on the risk of screen-detected breast cancer: The DeepJoint Algorithm
- Author
-
Rakez, Manel, Guillaumin, Julien, Chick, Aurelien, Coureau, Gaelle, Chamming's, Foucauld, Fillard, Pierre, Amadeo, Brice, and Rondeau, Virginie
- Subjects
Statistics - Applications - Abstract
Mammography-based screening programs play a crucial role in reducing breast cancer mortality through early detection. Their efficacy is influenced by breast density, a dynamic factor that evolves over time, modifying breast cancer risk. Women with high breast density face increased breast cancer risk, coupled with reduced mammographic sensitivity. In this work, we present the DeepJoint algorithm, a pipeline for quantitative breast density assessment, investigating its association with breast cancer risk. First, we develop a lightweight deep-learning segmentation model that uses processed mammography images acquired from multiple manufacturers to assess quantitative breast density metrics such as dense area and percent density. Then, we fit a joint model to evaluate the association of these metrics with the time-to-breast cancer occurrence using an extensive database of 77,298 women participating in breast cancer screening in the United States. We demonstrate the impact of the current value and slope of the biomarker on breast cancer risk. We also derive individual and dynamic breast cancer risk predictions to describe how the individual longitudinal evolution of dense area and percent density impacts the risk of breast cancer. This innovative approach aligns with the growing interest in personalized risk monitoring during screening, offering valuable insights for improving breast cancer prevention strategies.
- Published
- 2024