Deep convolutional neural network–based image classification for COVID-19 diagnosis

Abstract Initial cases of COVID-19 trace back to the end of 2019 which has laid foundations for the extensive spread of the disease risking lives worldwide. In response to the global coronavirus pandemic, early diagnosis of the disease is vital to prevent the virus from being spread to a larger population. Because of the unavailability of precise diagnostic toolkits, there arises a crying need to find efficient techniques that can be implemented for faster disease prediction while ensuring the accuracy of the prediction. Artificial intelligence (AI)–based solutions have the potential to help diagnose COVID-19 pandemic in an effective way. Automated image analysis with AI techniques can support clinical decision-making, improve workflow efficiency, and allow accurate and fast diagnosis of infection in a large number of patients. In the present study, existing convolutional neural network (CNN) models such as ResNeXt, Channel Boosted CNN, DenseNet, AlexNet, and VGG 16 were repurposed to assist in identifying the presence of COVID-19 before they reach mass scale. The dataset used in the study comprises of computed tomography (CT) images taken from 275 healthy individuals and 195 COVID-19 samples collected from 216 affected individuals. The proposed AI-based approach using deep learning models classifies COVID-19 affected cases by analyzing CT images and provides rapid detection of COVID-19 in a shorter time span. Ensemble classifier employed in this study proved to predict the presence of infection with a greater accuracy of 90.67% compared to the other similar works in the literature.