基于空洞空间金字塔池化的U-Net网络在肺部图像分割上的应用-《中国医学物理学杂志》

文章信息/Info

Title:: Lung field segmentation using U-Net based on atrous spatial pyramid pooling

作者:: 夏文静1; 周腊珍1; 陈红池1; 李坊佐1; 2; 吴頲3; 张翔1; 2; 1.赣南医学院医学信息工程学院，江西赣州 341000； 2.赣南医学院组织工程材料与生物制造江西省重点实验室，江西赣州 341000； 3.赣南医学院第一附属医院，江西赣州 341000

Author(s):: XIA Wenjing1; ZHOU Lazhen1; CHEN Hongchi1; LI Fangzuo1; 2; WU Ting3; ZHANG Xiang1; 2; 1. School of Medical Information Engineering, Gannan Medical University, Ganzhou 341000, China 2. Key Laboratory of Biomaterials and Biofabrication in Tissue Engineering, Gannan Medical University, Ganzhou 341000, China 3. The First Affiliated Hospital of Gannan Medical University, Ganzhou 341000, China

Keywords:: Keywords: chest X-ray image lung field segmentation U-Net atrous spatial pyramid pooling

摘要:: 目的：胸部X线图像中肺野的自动分割是相关疾病筛查和诊断的关键步骤，为了适应计算机辅助诊断系统的要求，提出一种基于空洞空间金字塔池化的U-Net网络对胸部X线图像中肺野进行自动分割。方法：在编码和解码之间引入带有空洞卷积的空间金字塔池化用于扩大接受域；同时，在多个尺度上获取图像上下文信息，用于从胸片中分割肺野，使用Montgomery数据集及深圳数据集进行验证。根据医学图像分割常用指标准确性、Dice相似系数及交并比评价基于空洞空间金字塔池化的U-Net网络分割肺野的性能。结果：验证准确性为98.29%，Dice相似系数为96.61%，交并比为93.47%。结论：本文提出一种基于空洞空间金字塔池化的U-Net网络用于分割肺野，相较于其他方法学习到更多边缘分割特征，取得更好的分割结果。

Abstract:: Abstract: Objective The auto-segmentation of lung fields in chest X-ray images is a critical step in the screening and diagnosis of related diseases. In order to meet the requirements of computer-aided diagnosis system, a U-Net network based on atrous spatial pyramid pooling is proposed to automatically segment lung fields in chest X-ray images. Methods Atrous spatial pyramid pooling was introduced between encoding and decoding to enlarge the receptive field, and the image context was obtained at multiple scales to segment lung fields in chest radiographs. Montgomery chest X-ray set and the Shenzhen chest X-ray set were used for validation. The performance of atrous spatial pyramid pooling based U-Net in lung field segmentation was evaluated by the commonly used evaluation criteria for medical image segmentation (accuracy, Dice similarity coefficient and intersection over union). Results The validation showed that the accuracy, Dice similarity coefficient, and intersection over union were 98.29%, 96.61%, and 93.47%, respectively. Conclusion Compared with other methods, U-Net based on atrous spatial pyramid pooling for lung field segmentation learns more edge segmentation features, and achieves better segmentation results.