We introduce a comprehensive screening platform for the SARS-CoV-2(a.k.a. COVID-19). The proposed AI-based system works on chest x-ray (CXR) images to predict whether a patient is infected with the SARS-CoV-2. The design is to first filter out images of the normal category in the second stage and then, in the final stage, conduct fine-grained classification to divide the pneumonia candidates into two specific types, namely, SARS-CoV-2 and non-SARS-CoV-2. The proposed method has been integrated into the SARS-CoV-2 screening system maintained by Taiwan Centers for Disease Control and Taiwan National Health Insurance Administration.
The performance of the 3-stage model reaches a 89% sensitivity on clinical dataset and a 80% specificity. The study is also available for read online at https://arxiv.org/abs/2004.12786. For testing the platform, please access via HERE. (Username: guest / Password: guest)