Development and Evaluation of an AI System for COVID-19

Cheng Jin; Weixiang Chen; Yukun Cao; Zhanwei Xu; Xin Zhang; Lei Deng; Chuansheng Zheng; Jie Zhou; Heshui Shi; Jianjiang Feng

In this study, we used clinically representative large-scale datasets from three centers in Wuhan and two publicly available chest CT datasets to develop and evaluate an AI system for the diagnosis of COVID-19. We compared the diagnostic performance of this system with that of five radiologists in a reader study of 200 cases, and the results showed the performance of this system is comparable to that of experienced radiologists in designated hospitals for COVID-19 in major epidemic areas. In addition, based on prediction score on every slice, we located the lesion areas in COVID-19 patients and statically analyzed lesion position in different subsets. We traced the specific phenotypic basis of the diagnosis output from the system on the original CT images through an interpretation network and apply radiomics analysis of the attentional region to understand the imaging characteristics of COVID-19. Figure 1 a shows the overall flow of the study.