ABSTRACT
Coronavirus has been declared a pandemic by the WHO. It is an extremely transmissible disease with a long incubation period, which increases the problem of identifying patients infected with COVID-19. Therefore, experts all around the world are trying to find different solutions to this problem using machine learning and AI. In this study, we used the data containing 283 X-rays and 309 CT scans of COVID-19-positive and normal patients from Kaggle and GitHub. For X-ray classification, three models of CNN are used, namely VGG-16, VGG-19, and a custom CNN model. Out of these models, VGG-19 showed the best performance of classification. For the classification of CT scans, SVM and a Modified CNN model was used. For SVM classification, the data was first divided into four subsets of 16×16, 32×32, 48×48 and 64×64 patches. Then, using GLCM, GLSZM, and DWT features were extracted from these subsets. The most accurate results were obtained on 48×48 patches using DWT algorithm with 10-fold cross validation. The modified CNN model achieved a classification accuracy of 94.1%. To evaluate the classification, performance parameters like accuracy, sensitivity, specificity, and f-score were used.
