COVID Edge-Net: Automated COVID-19 Lung Lesion Edge Detection in Chest CT Images

Wang, K.; Zhao, Y.; Dou, Y.; Wen, D.; Gao, Z.

Coronavirus Disease 2019 (COVID-19) has been spreading rapidly, threatening global health. Computer-aided screening on chest computed tomography (CT) images using deep learning, especially, lesion segmentation, is an effective complement for COVID-19 diagnosis. Although edge detection highly benefits lesion segmentation, an independent COVID-19 edge detection task in CT scans has been unprecedented and faces several difficulties, e.g., ambiguous boundaries, noises and diverse edge shapes. To this end, we propose the first COVID-19 lesion edge detection model: COVID Edge-Net, containing one edge detection backbone and two new modules: the multi-scale residual dual attention (MSRDA) module and the Canny operator module. MSRDA module helps capture richer contextual relationships for obtaining better deep learning features, which are fused with Canny features from Canny operator module to extract more accurate, refined, clearer and sharper edges. Our approach achieves the state-of-the-art performance and can be a benchmark for COVID-19 edge detection. Code related to this paper is available at: https://github.com/Elephant-123/COVID-Edge-Net. © 2021, Springer Nature Switzerland AG.