Multi-level nested pyramid network for mass segmentation in mammograms

Runze Wang, Yide Ma, Wenhao Sun, Yanan Guo, Wendao Wang, Yunliang Qi, Xiaonan Gong

Abstract: Mass segmentation in mammograms is an important and challenging topic in breast cancer computer-aided diagnosis. In this work, we propose a novel multi-level nested pyramid network (MNPNet) for dealing with the limitations of intra-class inconsistency and inter-class indistinction that commonly existed in mass segmentation in mammograms. The MNPNet includes an encoder and a decoder. The former encodes contextual information, low-level detail information, and high-level semantic information in a multi-level multi-scale manner by multi-level nesting atrous spatial pyramid pooling (ASPP) module on the feature pyramid generated by modified ResNet34. The latter consist of a series of simple yet effective bilinear upsampling and feature fusion operations to refine the segmentation results along mass boundaries. Our proposed MNPNet is greatly demonstrated on two public mammographic mass segmentation databases including INbreast and DDSM-BCRP, respectively achieving the Dice index of 91.10% and 91.69% without any pre/post-processing.