DCA-Net: A W-shaped network for segmenting COVID-19 infected area from medical images

Coronavirus disease (COVID-19) poses a significant threat to humans in 2019. Automated and accurate segmentation of the infected region of COVID-19 computed tomography (CT) images can help doctors diagnose and treat the disease. However, the variable shape of COVID-19 infected areas, which can be easily confused with other lung tissues, poses a challenge for CT image segmentation. To address this problem, a deep learning-based convolutional neural network is proposed for the automatic segmentation of COVID-19 lung infection regions. Our proposed segmentation method uses a U-Net network as the backbone, constructed as a coarse to fine segmentation network. Firstly, we introduce our designed contour-enhanced module (CA) in the coarse segmentation network to effectively extract the lung region; secondly, we introduce our designed multi-scale feature attention module (MFA) in the fine segmentation network to enable the network to extract spatial efficiently and channel information, better focus on quantifying the effective region, and enhance the network segmentation effect. Using the COVID-19 public dataset, the proposed network achieves the best segmentation results. The Dice, IOU, F1-Score, and Sensitivity metrics reach 88.74%, 78.73%, 86.58%, and 88.16%, respectively. DCA-Net can efficiently segment the COVID-19 infected region, which can be of great clinical benefit.