From Unilateral to Bilateral Learning: Detecting Mammogram Masses with Contrasted Bilateral Network

Yuhang Liu, Zhen Zhou, Shu Zhang, Ling Luo, Qianyi Zhang, Fandong Zhang, Xiuli Li, Yizhou Wang, Yizhou Yu

Abstract

The comparison of bilateral mammogram images is important for finding masses especially in dense breasts. However, most existing mammogram mass detection algorithms only considered unilateral image. In this paper, we propose a deep model called contrasted bilateral network (CBN) to take bilateral information into consideration. In CBN, Mask R-CNN is used as a basic framework, upon which two major modules are developed to exploit the bilateral information: distortion insensitive comparison module and logic guided bilateral module. The former one is designed to be robust to nonrigid distortion of bilateral registration, while the latter one integrates the bilateral domain knowledge of radiologist. Experimental results on DDSM dataset demonstrate that the proposed algorithm achieves the state-of-the-art performance.
BibTeX
@inproceedings{liu2019from,
  author = {Yuhang Liu and Zhen Zhou and Shu Zhang and Ling Luo and Qianyi Zhang and Fandong Zhang and Xiuli Li and Yizhou Wang and Yizhou Yu},
  title = {{From Unilateral to Bilateral Learning: Detecting Mammogram Masses with Contrasted Bilateral Network}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2019},
  doi = {10.1007/978-3-030-32226-7\_53},
}

Loading related content...

Figure