ABSTRACT
One of the essential tools in diagnosing and controlling COVID-19 is high-resolustion radiology images. However, radiology imaging systems are common in all hospitals; there is a shortage of image analysis techniques because of a lack of expertise in health centers. Computer-aided diagnosis (CAD) has become more important with the COVID-19 pandemic. Since the disease spreads rapidly and shows symptoms after a long incubation period, X-ray images and computed tomography images are used for early detection and treatment. The rapid interpretation of these images and the classification of them are of great importance in the detection and management of the disease. In this study, 14 deep learning models with four classifiers were compared to diagnose COVID-19 by using X-ray images. Seven main performance metrics are measured to compare the methods. This study presents that pre-trained models produce successful results in feature extraction, while traditional classifiers yield successful solutions in classification. The proposed approach increases test performance values and reveals the most effective methods for diagnosis of COVID-19. Seven of pre-trained models (AlexNet, DarkNet-53, DenseNet-201, GoogLeNet, Inception v3, MobileNetv2, ResNet-18) give the highest performance scores. The support vector machine is the most effective classifier together with 12 models.
