Exploring synthesizing 2D mammograms from 3D digital breast tomosynthesis images

In this study, we propose a novel deep learning approach to synthesizing 2D mammograms from 3D digital breast tomosynthesis (DBT) images. The objective of our work is to eliminate the need for obtaining two separate mammography scans, 3D and 2D, by creating a method of projecting 3D DBT images to 2D. Our method is compared to the state-of-the-art proprietary Hologic C-VIEW software and two simple baselines. We identify several potential directions for improvement. We make our code and model weights available as open-source, thereby providing the first publicly accessible deep learning model for converting 3D DBT images to synthesized 2D mammograms.