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

There are still some drawbacks and future works of this research. First, collecting more data on other types of viral pneumonias or lung lesions can help improve its specificity further. Second, based on many chest CTs with detailed labelled lesions, a semantic segmentation algorithm can be trained to locate the outline of the lesion more accurately than Guided Grad-GAM, and distinguish the detailed category of the lesion. Overall, the proposed AI system has been comprehensively validated on large dataset with diagnosis performance comparable to human experts in diagnosing COVID-19. Unlike classical blackbox deep learning approaches, by visualizing AI system and applying radiomics analysis, it can decode effective representation of COVID-19 on CT imaging, and potentially lead to the discovery of new biomarkers. Radiologists could perform an individualized diagnosis of COVID-19 with the AI system, adding new driving force for fighting the global spread of outbreak.