Evaluation of Convolutional Neural Networks for COVID-19 Detection from Chest X-Ray Images

The paper presents a comparative analysis of five Convolutional Neural Networks (CNN's) for reliable detection of COVID-19 from chest X-ray (CXR) images. The transfer learning approach is applied to a dataset compiled from public sources, aiming at discriminating the presence of COVID-19 pathology from normal, lung opacity, and pneumonia situations. Experimental results indicate that best accuracy performances are obtained by the Resnet-50 network, while all CNN's compare favorably with existing solutions.