AUNet: Attention-guided dense-upsampling networks for breast mass segmentation in whole mammograms

• DOI: 10.48550/arxiv.1810.10151
• Publication Date: 2018-01-01
• Authors: Hui Sun; Cheng Li; Boqiang Liu; Hairong Zheng; Dagan Feng; Shanshan Wang
• Keywords: Upsampling; Fuse (electrical)

Mammography is one of the most commonly applied tools for early breast cancer screening. Automatic segmentation masses in mammograms essential but challenging due to low signal-to-noise ratio and wide variety mass shapes sizes. Existing methods deal with these challenges mainly by extracting mass-centered image patches manually or automatically. However, manual patch extraction time-consuming automatic brings errors that could not be compensated following step. In this study, we propose a novel attention-guided dense-upsampling network (AUNet) accurate whole directly. AUNet, employ an asymmetrical encoder-decoder structure effective upsampling block, block (AU block). Especially, AU designed have three merits. Firstly, it compensates information loss bilinear dense upsampling. Secondly, designs more method fuse high- low-level features. Thirdly, includes channel-attention function highlight rich-information channels. We evaluated proposed on two publicly available datasets, CBIS-DDSM INbreast. Compared state-of-the-art fully convolutional networks, AUNet achieved best performances average Dice similarity coefficient 81.8% 79.1%