A Novel Approach to Detect and Monitor COVID-19 Infection Using Transfer Learning Concept in AI

Abstract

COVID-19 is an ongoing pandemic that is continuing to spread after recording one hundred million cases, causing millions of casualties, overwhelming health care systems of many countries, and threatening the whole world. Monitoring and assessing the severity of COVID-19 through artificial intelligence would be a practical support for medical practitioners reviving patients and offloading the burden from medical system. Previous works exploited deep learning, for this purpose, which produces inexplainable diagnosis results and lacks medical evidence. Integrating clinical symptom into diagnosis with deep learning will support generating results more compelled and validated. In this study, we focus on verifying the effectiveness of applying the human lung lesion, specifically Ground Glass Opacity and Consolidation, caused by typical pneumonia for COVID-19 detection or severity assessment on chest X-ray image with deep learning technology. We have conducted multiple experiments with state-of-art machine learning architectures (MobileNetV2, ResNet, Faster R-CNN) on many datasets to establish the conclusion. The experiment result demonstrates that lung lesion is useful when incorporating with deep learning solutions for monitoring COVID-19 progression and will provide solid pathway to develop an improved model and support better research in the future.