SCOAT-Net: A novel network for segmenting COVID-19 lung opacification from CT images

doi:10.1016/j.patcog.2021.108109

Pattern Recognition

Volume 119, November 2021, 108109

https://doi.org/10.1016/j.patcog.2021.108109 Get rights and content

Highlights

•
We propose a novel spatial- and channel-wise coarse-to-fine attention network (SCOAT-Net) for segmentation of COVID-19 lung opacification from CT images and achieves state-of-the-art performance.
•
We use the attention mechanism so that the neural network can generate attention maps without external region of interest (ROI) supervision, increasing the interpretability of the neural network.
•
The generalization ability and compatibility of the proposed SCOAT-Net are validated on two external datasets, showing that the proposed model has specific data migration capability.

Abstract

Automatic segmentation of lung opacification from computed tomography (CT) images shows excellent potential for quickly and accurately quantifying the infection of Coronavirus disease 2019 (COVID-19) and judging the disease development and treatment response. However, some challenges still exist, including the complexity and variability features of the opacity regions, the small difference between the infected and healthy tissues, and the noise of CT images. Due to limited medical resources, it is impractical to obtain a large amount of data in a short time, which further hinders the training of deep learning models. To answer these challenges, we proposed a novel spatial- and channel-wise coarse-to-fine attention network (SCOAT-Net), inspired by the biological vision mechanism, for the segmentation of COVID-19 lung opacification from CT images. With the UNet++ as basic structure, our SCOAT-Net introduces the specially designed spatial-wise and channel-wise attention modules, which serve to collaboratively boost the attention learning of the network and extract the efficient features of the infected opacification regions at the pixel and channel levels. Experiments show that our proposed SCOAT-Net achieves better results compared to several state-of-the-art image segmentation networks and has acceptable generalization ability.

Keywords

COVID-19

Convolutional neural network

Segmentation

Lung opacification

Attention mechanism

Cited by (0)

Shixuan Zhao received the B.S degree from the University of Electronic Science and Technology of China (UESTC). He is now a Ph.D. student with the MOE Key Laboratory for Neuroinformation, School of Life Science and Technology, UESTC, China. His research interests are medical image analysis and computer vision.

Zhidan Li received the B.S degree from China Medical University. He is now a master student with the MOE Key Laboratory for Neuroinformation, School of Life Science and Technology, University of Electronic Science and Technology of China, China. His research interests are medical image classification and segmentation.

Yang Chen received the B.S. degree from Harbin Medical University, the M.S. degree from Sichuan University, and the Ph.D. degree from the University of Electronic Science and Technology of China. She has worked in the Imaging department of West China Hospital of Sichuan University for more than ten years and is now a postdoctoral fellow at the West China Biomedical Big Data Center. Her research interests artificial intelligence analysis of medical images.

Wei Zhao received the Ph.D. degree in imaging and nuclear medicine from Fudan University, China. He is a radiologist of The Second Xiangya Hospital. His research interests include chest CT imaging, radiomics and deep learning.

Xingzhi Xie received the B.S. degree in clinical medicine from Central South University, China. She is a graduate student in imaging and nuclear medicine at The Second Xiangya Hospital. Her research interests include CT imaging, radiomics and deep learning.

Jun Liu is the director of the radiology department of The Second Xiangya Hospital. He is also the leader of 225 subjects in Hunan Province, a National member of the Neurology Group of the Chinese Society of Radiology, National Committee of the Neurology Group of the Radiological Branch of the Chinese Medical Association. His research interests include brain functional imaging, radiomics and deep learning.

Zhao Di received his Ph.D. degree in computational science from Louisiana Tech University. Zhao Di has been engaged in post-doctoral research at Columbia University and Ohio State University. He is undertaking a number of national, provincial and ministerial research projects. He has good research experience in ”deep learning for medical image analysis”, and has published 25 academic journal papers and academic conference papers. He published one book and one translation. He holds a number of academic positions.

Yongjie Li received his Ph.D. degree in biomedical engineering from the University of Electronic Science and Technology of China (UESTC) in 2004. He is currently a Professor with the MOE Key Laboratory for Neuroinformation, School of Life Science and Technology, UESTC, China. He has published more than 90 peer-reviewed international journals and conference papers including Neuroimage, IEEE TPAMI, IEEE TIP, IEEE TBME, ICCV, CVPR, etc. He is also an active reviewer for more than ten leading journals and conferences. His research interests include visual mechanism modeling, and the applications in image processing for computer vision and medical diagnosis.

View Abstract