General Adversarial Networks: A Tool to Detect the Novel Coronavirus from CT Scans

Abstract

Detection of the novel Corona virus in the early stages is crucial, since no known vaccines exist. Artificial Intelligence- aided prognosis using CT scans can be used as an effective method to identify symptoms of the virus and can thus significantly reduce the workload on the radiologists, who have to perform this task using their eyes. Among the most widely used deep learning convolutional neural networks, research shows that the Xception, Inception and the ResNet50 provide the best accuracy in detecting Covid-19. This paper proposes that using General Adversarial Network (GAN) as a data augmentation technique, in combination with these models will significantly improve the accuracy and thereby increase the chances of detecting the same. The paper also compares and contrasts how each of the three GANs namely DCGAN, LSGAN, CoGAN, perform in association with the aforementioned models. The main aim of this paper is to determine the most credible GAN network to carry out the task of data augmentation as well to prove that involving GANs would improve the existing accuracy of our model, paving way for an effective approach to train the model.