An online AI diagnosis system for COVID-19, jointly developed by Beijing Institute of Genomics (China National Center for Bioinformation, CNCB), Guangzhou Regenerative Medicine and Health Guangdong Laboratory (Bioisland Laboratory), and the Computer Network Information Center of the Chinese Academy of Sciences, has been implemented recently in China Science & Technology Cloud, offering free online diagnosis service for COVID-19 to the world. Also, this AI diagnosis system could be accessed in the database 2019 Novel Coronavirus Resource (2019nCoVR), which is maintained by CNCB. According to the real test results in domestic hospitals, it takes 6~15 seconds to upload an adult’s lung CT image data (30-50 levels), and takes only 10~20 seconds for AI diagnosis. On April 25, Prof. ZHANG Kang from the Bioisland Laboratory published an article entitled “Clinically Applicable AI System for Accurate Diagnosis, Quantitative Measurements and Prognosis of COVID-19 Pneumonia Using Computed Tomography” in Cell. For this work, CNCB supported online AI diagnosis and provided international sharing service of chest CT images, clinical metadata and codes. Within one week of the data sharing, the downloads have reached over 1.5 million times, among which 36% are from abroad covering 97 countries/regions (USA (14%), India (2%), UK, France and Japan (1%, respectively)), with a total data volume exceeding 10 TB. Since the online of the AI diagnosis system for COVID-19 on February 25, its diagnosis accuracy has reached over 90% according to double-blind test. This system has already been put into use in hospitals including the Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University, Wuhan Jinyintan Hospital, etc., assisting for the rapid identification and diagnosis of COVID-19, lung-lesion segmentation and quantitative measurements, evaluation of drug treatment effects, and prognosis prediction of severe cases, contributing to the clinical treatment and decrease of mortality. COVID-19 pneumonia diagnosis results based on chest CT images (Image by CNCB)
