CXR-based Diagnosis of COVID-19 using Deep Learning with CycleGAN for Data Augmentation

A generative adversarial network called CycleGAN is used for image dataset augmentation and successfully employed in transfer learning of several popular deep neural network architectures pretrained on ImageNet, to diagnose COVID-19 based on chest X-rays. The significant performance improvement of the classifiers indicates that CycleGAN is usable in the image-based diagnosis of COVID-19 when the available dataset is small or class imbalance affects the results. Therefore, it could accelerate the development of efficient tools for a fast response in rapidlychanging pandemic conditions.