Comparative Study of Pretrained Network Feature Extraction and Classifiers for COVID-19 Detection

Abstract

Severe Acute Respiratory Syndrome (SARS-CoV-2) causes COVID-19, an infectious disease. It has since spread worldwide, leading to an ongoing pandemic. A coronavirus is a virus that infects the nose, sinuses, and upper throat. Fever, cough, trouble in breathing, loss of smell and taste are some of the symptoms. COVID-19 causes mild to moderate infection in most infected patients, who recover without the need for additional treatment. However, it is critical in the lives of older persons and persons with different diseases like diabetes, cancer, cardiovascular disease, and so on. In this study, we propose a method for detecting COVID-19 from CT images. Here the features are extracted using the pretrained network, ResNet-50, and categorized as COVID-19 infected or not using the KNN classifier. This study also focuses on the efficiency of pre-trained networks and other classification approaches for the automatic detection of COVID-19. The AlexNet, VGG-16, VGG-19, ResNet-50, ResNet-101, and DenseNet-201 pre-trained networks are used to extract features for analysis. We explored the Support Vector Machine(SVM), Ensemble based method, K Nearest Neighbour(KNN), Discriminant approach, Tree-based, and Naive Bayes classifiers to get the best classifier. The method was tested on the SARS-CoV-2 CT data set. The ResNet-50 with KNN classifier has a sensitivity, specificity accuracy, and F1-score of 95.99 %, 99.16%, 97.52%, and 97.56%, respectively, which is superior to the work reported.