False-Negative Results of Real-Time Reverse-Transcriptase Polymerase Chain Reaction for Severe Acute Respiratory Syndrome Coronavirus 2: Role of Deep-Learning-Based CT Diagnosis and Insights from Two Cases

Li, Dasheng; Wang, Dawei; Dong, Jianping; Wang, Nana; Huang, He; Xu, Haiwang; Xia, Chen

Of note, approximately 96% of patients with COVID-19 presented with chest CT abnormalities, such as multiple bilateral and peripheral ground-glass opacities and consolidation (3, 4) , making chest CT features essential in recognizing COVID-19. The National Health Commission of China revised the diagnostic criteria in the Hubei province, where a severe epidemic occurred (5) . A new diagnostic type called "Clinical diagnosis" was set according to the presence of pneumonia on chest CT, regardless of rRT-PCR results. To some extent, CT features and rRT-PCR results were complimentary in the diagnosis of COVID-19. From a clinical perspective, CT features could be utilized as the first and immediate reference for doctors to screen the highly suspected cases and to take necessary actions while rRT-PCR serves as a confirmation tool, the results of which could be utilized later to decide the subsequent action of continuing isolated treatment or discharge. Notably, our hospital was facilitated with a DL-based computer-aided diagnostic system (InferRead CT Pneumonia, Infervision, Beijing, China) for pneumonia, which greatly improved the detection efficiency for patients highly suspected with COVID-19 by alarming the technician within 2 minutes when any suspected cases was found after CT examination. The automatic lesion segmentation on CT was also helpful to evaluate the progression of COVID-19 quantitatively. With an integrated approach of DL, CT features, and rRT-PCR results, the screening and treatment of COVID-19 would be more effective.