Wasserstein GAN based Chest X-Ray Dataset Augmentation for Deep Learning Models: COVID-19 Detection Use-Case

• Main Authors: Hussain, B. Zahid, Andleeb, Ifrah, Ansari, Mohammad Samar, Joshi, Amit Mahesh, Kanwal, Nadia
• Format: Conference Proceeding
• Language: English
• Subjects: Biological system modeling; COVID-19; Data Augmentation; Data models; Deep learning; Generative Adversarial Network (GAN); Generative adversarial networks; Training; Training data; Wasserstein GAN
• ISSN: 2694-0604
• DOI: 10.1109/EMBC48229.2022.9871519

The novel coronavirus infection (COVID-19) is still continuing to be a concern for the entire globe. Since early detection of COVID-19 is of particular importance, there have been multiple research efforts to supplement the current standard RT-PCR tests. Several deep learning models, with varying effectiveness, using Chest X-Ray images for such diagnosis have also been proposed. While some of the models are quite promising, there still remains a dearth of training data for such deep learning models. The present paper attempts to provide a viable solution to the problem of data deficiency in COVID-19 CXR images. We show that the use of a Wasserstein Generative Adversarial Network (WGAN) could lead to an effective and lightweight solution. It is demonstrated that the WGAN generated images are at par with the original images using inference tests on an already proposed COVID-19 detection model.