Style Randomization Improves the Robustness of Breast Density Estimation in MR Images

Abstract

Breast density, a crucial risk factor for future breast cancer development, is defined bythe ratio of fat to fibro-glandular tissue (FGT) in the breast. Accurate breast and FGTsegmentation is essential for robust density estimation. Previous research on FGT segmen-tation in MRI has highlighted the significance of training on both images with and withoutfat suppression to enhance network’s robustness. In this study, we propose a novel dataaugmentation technique to further exploit the multi-modal training setup motivated by theresearch in style randomization. We demonstrate that the network trained with the pro-posed augmentation is resilient to variations in fat content, showcasing improved robustnesscompared to solely training with multi-modal data. Our method effectively improves FGTsegmentation, thereby enhancing the overall reliability of breast density estimation