A CNN-based hybrid model to detect Coronavirus disease

Abstract

In this paper, a hybrid classification technique for COVID-19 disease is proposed. The proposed model solves the two-class classification problem (covid, normal). In this study, we have presented hybrid models integrating superior deep learning and machine learning classifiers: Convolutional Neural Network (CNN) and Support Vector Machine (SVM), CNN and AdaBoost, CNN and K Nearest Neighborhood (kNN), CNN and Multilayer Perceptron (MLP), CNN and Naive Bayes (NB). In these models, CNN performs as a trainable deep feature extractor, and SVM, AdaBoost, kNN, MLP, NB behave as a recognizer. All experiments have been performed on COVID-CT and SARS-CoV-2 CT combined image datasets. As a result, proposed hybrid methods have been compared in terms of sensitivity, accuracy, precision, F1-score, AUC-score, specificity, FPR, FDR, and FNR. CNN+SVM, CNN+MLP, and CNN+kNN have achieved outperforming results according to the other models, respectively. Also, CNN+SVM performed the best (achieving 85.85% sensitivity, 85.86% precision, 85.86% accuracy, 85.85% F1-score, 85.85% AUC score, 86.47% specificity, 13.52% FPR, 13.86% FDR, and 14.76% FNR). When the results are examined, the proposed hybrid system is seen to be efficient to detect COVID-19. Also, the performance of the proposed hybrid system is better than the successful studies found on COVID-CT and SARS-CoV-2 CT combined image datasets in the literature.