Deep Learning-based Detection for COVID-19 from Chest CT using Weak Label

Chuansheng Zheng; Xianbo Deng; Qing Fu; Qiang Zhou; Jiapei Feng; Hui Ma; Wenyu Liu; Xinggang Wang

Our study provided a typical and successful solution for developing medical AI for emerging diseases, such as COVID-19. While we were developing this AI, doctors in Wuhan were still extremely busy with treating a huge number of COVID-19 patients and it may be impossible for them to annotate the lesions in CT volumes in the current austere fight against this epidemic. Thanks to the weaklysupervised algorithm in this study, locations of pulmonary lesions in CT volumes are not necessary to be annotated, and radiologists' annotating efforts can be minimized, i.e., only providing patient-level labels. Therefore, developing a helpful AI tool swiftly has become possible and available in the clinical application. In the future, the burden of AI experts could be lifted significantly by automatic machine learning (AutoML) [25] .