Automatic Polyp Segmentation via Multi-scale Subtraction Network

More than 90% of colorectal cancer is gradually transformed from polyps. In clinical practice, precise polyp segmentation provides important information in the early detection cancer. Therefore, automatic techniques are great importance for both patients and doctors. Most existing methods based on U-shape structure use element-wise addition or concatenation to fuse different level features progressively decoder. However, two operations easily generate plenty redundant information, which will weaken complementarity between features, resulting inaccurate localization blurred edges To address this challenge, we propose a multi-scale subtraction network (MSNet) segment colonoscopy image. Specifically, first design unit (SU) produce difference adjacent levels encoder. Then, pyramidally equip SUs at with varying receptive fields, thereby obtaining rich information. addition, build training-free “LossNet” comprehensively supervise polyp-aware bottom layer top layer, drives MSNet capture detailed structural cues simultaneously. Extensive experiments five benchmark datasets demonstrate that our performs favorably against most state-of-the-art under evaluation metrics. Furthermore, runs real-time speed \(\sim \)70fps when processing \(352 \times 352\) The source code be publicly available https://github.com/Xiaoqi-Zhao-DLUT/MSNet.