COV-CTX: A Deep Learning Approach to Detect COVID-19 from Lung CT and X-Ray Images

Abstract

With the massive outbreak of coronavirus (COVID-19) disease, the demand for automatic and quick detection of COVID-19 has become a crucial challenge for scientists around the world. Many researchers are working on finding an automated and effective system for detecting COVID-19. They have found that computed tomography (CT-scan) and X-ray images of COVID-19 infected patients can provide more accurate and faster results. In this paper, an automated system is proposed named as COV-CTX which can detect COVID-19 from CT-scan and X-ray images. The system consists of three different CNN models: VGG16, VGG16-InceptionV3-ResNet50, and Francois CNN. The models are trained with CT-scan and X-ray images individually to classify COVID-19 and non-COVID patients. Finally, the results of the models are combined to develop a voting ensemble of classifiers to ensure more accurate and precise results. The three models are trained and validated with 9412 CT-scan images (4756 numbers of COVID positive and 4656 numbers of non-COVID images) and 3257 X-ray images (1647 numbers of COVID positive and 1610 numbers of non-COVID images). The proposed system, COV-CTX provides up to 96.37% accuracy, 96.71% precision, 96.02% F1-score, 97.24% sensitivity, 95.35% specificity, 92.68% Cohens Kappa score for CT-scan image based COVID-19 detection and 99.23% accuracy, 99.37% precision, 99.22% F1-score, 99.39% sensitivity, 99.07% specificity, 98.46% Cohens Kappa score for X-ray image based COVID-19 detection.