Edge-enhancement cascaded network for lung lobe segmentation based on CT images

Frontiers in Physics

Physics

Ye Yuan¹  Qingyao Luo²  Nan Bao²  Li-Bo Zhang³  Qiankun Li⁴ 

[1] College of Computer Science and Engineering, Northeastern University, Shenyang, China;Key Laboratory of Intelligent Computing in Medical Image, Ministry of Education, Northeastern University, Shenyang, China;College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China;Department of Radiology, General Hospital of the Northern Theater of the Chinese People’s Liberation Army, Shenyang, China;Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei, China;Department of Automation, University of Science and Technology of China, Hefei, China;

关键词: lung lobe segmentation;    CT images;    multi-stage cascaded network;    edge enhancement;    boundary response map;   

DOI  :  10.3389/fphy.2023.1098756

 received in 2022-11-15, accepted in 2023-02-17,  发布年份 2023

来源: Frontiers

【 摘 要 】

In order to reduce postoperative complications, it is required that the puncture needle should not pass through the lung lobe without tumor as far as possible in lung biopsy surgery. Therefore, it is necessary to accurately segment the lung lobe on the lung CT images. This paper proposed an automatic lung lobe segmentation method on lung CT images. Considering the boundary of the lung lobe is difficult to be identified, our lung lobe segmentation network is designed to be a multi-stage cascade network based on edge enhancement. In the first stage, the anatomical features of the lung lobe are extracted based on the generative adversarial network (GAN), and the lung lobe boundary is Gaussian smoothed to generate the boundary response map. In the second stage, the CT images and the boundary response map are used as input, and the dense connection blocks are used to realize deep feature extraction, and finally five lung lobes are segmented. The experiments indicated that the average value of Dice coefficient is 0.9741, which meets the clinical needs.

【 授权许可】

Unknown   
Copyright © 2023 Bao, Yuan, Luo, Li and Zhang.