Automatic Classification of COVID-19 from Chest X-Ray Image using Convolutional Neural Network

Abstract

COVID-19 has become one of the most virulent, acute, and life-threatening diseases in recent times. No clinically approved drug is available till now for its treatment. Therefore, early and swift detection is very essential for reducing overall mortality. The chest x-ray image is one of the possible alternative methods for detecting COVID-19. Researchers are exploring image processing techniques along with deep learning-based models like AlexNet, VGGNet, SqueezeNet, GoogleNet, etc. to detect COVID-19. This study aims to formulate, implement and investigate deep learning-based models and their probable hyperparameters tuning for obtaining the best results when identifying COVID-19 using chest x-ray images. To meet this objective, images from different publicly available databases were collected. In this paper, ResNet18, ResNet50V2, DenseNet121, DenseNet201, modified DenseNet201 and VGG16 were used to detect COVID-19. From the experimental results, modified DenseNet201 showed the best performance with 99.5% mean accuracy, 99.5% mean F1 score and 100% mean sensitivity in binary (COVID-19 and normal) classification and 98.33% mean accuracy, 98.34 mean F1 score, and 98.34% mean sensitivity (98% sensitivity for COVID-19) in 3-class (COVID-19, pneumonia, normal) classification. This may contribute to the process of designing and implementing a system that can detect COVID-19 automatically in the near future and enhance the quality of healthcare services.