COVID-19 Classification From X-rays : A Comparative Study

Abstract

With the arrival of the most recent coronavirus pandemic, it was a must to find solutions to detect this dangerous virus. Analyzing X-ray images was among the exploited techniques to control this disease. However, the doctor's subjectivity in analyzing X-rays was the first obstacle in detecting this virus accurately. Applying new deep learning techniques to x-ray images can be a potential solution to reduce this subjectivity. This paper aims to conduct a comparative study between six different CNN architectures (VGG16, VGG19, Inception, Xception, DenseNet, and ChexNet) for COVID-19 detection from X-rays. The obtained results based on the transfer learning strategy confirm the efficiency of the VGG 16, where its achieved 98.69 % of accuracy on the COVID-19 Radiography Dataset.