Bronchial Light Microscopy Image Segmentation Based on Boundary Attention

∗The identification of bronchus is of great significance in assisting the diagnosis of lung diseases. However, identifying the bronchus from tissue light microscopy images is a heavily repetitive task that requires a lot of time and effort. Most of the mainstream segmentation methods pay attention to the overall accuracy of the region, without special consideration for the boundaries. However, bronchi often have flexible shapes, which poses a challenge for accurate segmentation, especially for details at the edges. Therefore, this paper proposes a boundary-attention based bronchi segmentation network. This network is a “predict and refine” architecture. Specifically, a coarse segmentation result is first generated by the prediction network, and then the edge segmentation quality is improved by the refinement network. In addition, by specially designed hybrid loss, our network can focus on patch-level contextual information as well as pixel-level accuracy. At the same time, the global attention module and the local attention module enable our network to extract both multiscale features and focus on error-prone regions. Through our network, not only fine segmentation results can be achieved, but also superior performance at the bronchial boundary. Experiments on the BronSeg dataset show that our method outperforms mainstream methods in all metrics, especially in mIOU, which reaches 88.41%. ∗Corresponding author. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. CCRIS’21, August 20–22, 2021, Qingdao, China © 2021 Association for Computing Machinery. ACM ISBN 978-1-4503-9045-3/21/08. . . $15.00 https://doi.org/10.1145/3483845.3483890 CCS CONCEPTS • Artificial intelligence; • Computer vision; • Image segmentation;