A Simplified Convolutional Neural Network Design for COVID-19 Classification on Chest X-ray Images

Abstract

COVID-19 is a respiratory virus that causes the spread of infection and has affected human around the world. The infection frequently results in pneumonia in human which can be detected using lung imaging, chest X-ray images. Deep learning models have been demonstrated to an effective COVID-19 interpretation on chest radiography. In this paper, we have proposed a simplified convolutional neural network model for COVID-19 screening that can classify the appearance of COVID-19 lesion into two classes. The proposed model; despite using fewer layers and the utilization of data augmentation approach in training process, can achieve the greater outcome. To evaluate the proposed model, we have used a partial of the public dataset, COVID-19 Radiography Database which is a collection of 13,808 chest X-ray images. At the final stage, the Grad-CAM visualization method has been used to enhance the important region of chest X-ray images in order to provide the explanations of COVID-19 predictions.