Review of deep learning-based approaches for COVID-19 detection

Abstract

As the COVID-19 pandemic broke out worldwide, many deep learning-based methods are proposed to assist the doctors in COVID-19 diagnosis. This paper introduces open-source datasets of COVID-19 images and tests state-of-the-art COVID-19 diagnosis methods to provide a comprehensive review of these technologies. According to the experimental results, this paper introduces two interesting observations: 1) deep learning-based methods focus on big visual features rather than small detailed features; 2) the convolutional neural networks pay attention to the region of Lung Ultrasound images, which is also considered as crucial observation region from doctors' perspectives. These observations prove the efficiency of deep-learning solutions since they can learn essential doctors' COVID-19 diagnosis rules.