Analysing most efficient deep learning model to detect COVID-19 from computer tomography images

Abstract

COVID-19 illness has a detrimental impact on the respiratory system, and the severity of the infection may be determined utilizing a selected imaging technique. Chest computer tomography (CT) imaging is a reliable diagnostic technique for finding COVID-19 early and slowing its progression. Recent research shows that deep learning algorithms, particularly convolutional neural network (CNN), may accurately diagnose COVID-19 using lung CT scan images. But in an emergency, detection accuracy simply is not enough. Determinants of data loss and classification completion time play a critical element. This study addresses the issue by finding the most efficient CNN model with the least data loss and classification time. Eight deep learning models, including Max Pooling 2D, Average Pooling 2D, VGG19, VGG16, MobileNetV2, InceptionV3, AlexNet, NFNet using a dataset of 16000 CT scans image data of COVID-19 and non-COVID-19 are compared in the study. Using the confusion matrix, the performance of the models is compared and together with the data loss and completion time. It is observed from the research that MobileNetV2 provides the highest accurate result of 99.12% with the least data loss of 0.0504% in the lowest classification completion time of 16.5secs per epoch. Thus, employing MobileNetV2 gives the best and the quickest result in an emergency.