COVID-19 Diagnosis on Chest X-Ray Images using an Xception-based Deep Learning Classifier and Gradient-weighted Class Activation Mapping

Abstract

This paper proposes the development of a deep learning model for diagnosing COVID-19 through the analysis of chest X-ray images. First, data augmentation is implemented to avoid overfitting and improve model generalization. Then, instead of conventional image segmentation techniques, Gradient-weighted Class Activation Mapping (Grad-CAM) is used to highlight the important regions directly related to COVID-19. Subsequently, transfer learning is implemented to transform the data of the X-ray images to a reduced set of features using the Xception convolutional neural network. Finally, a classification neural network is designed, parameterized and trained, which is capable of recognizing healthy patients with 97% accuracy, while the detection rate for patients infected with COVID-19 was 92%.