Comparison of Convolutional Neural Network Models to Detect Covid-19 on CT-Scan Images

Coronavirus Disease-2019 (COVID-19) is a new type of coronavirus that attacks the respiratory tract and can be contagious. More severe cases of this infection can cause pneumonia, acute respiratory syndrome, kidney failure and even death. Detection of Covid-19 is a crucial step to identifying suspected Covid-19 patients early. CT Scan is a significant modality to detect Covid-19 disease. Automatic detection of Covid-19 on CT-Scan implemented various convolutional neural network models such as Mobile-Net, RestNet50, and DenseNet. These three architectures were implemented for Covid-19 detection but were not compared on the same database. In this study, the authors propose a performance comparison of these three models to detect Covid-19 using an open-source dataset of 2038 CT-Scan images. The methods involved data preparation, preprocessing, training, testing, and performance evaluation. The result showed that the best accuracy is obtained by the DenseNet121 model with 95.98% and MobileNet with 95.09%, while ResNet50 has 92.00% accuracy. Although the accuracy of ResNet50 is lower than the other two models, the travel time of the ResNet50 model is faster than the DenseNet121 model, with an average travel time of 16 seconds per epoch.