Employing Generative Adversarial Network in COVID-19 Diagnosis

In recent years, many papers and models have been developed to study the classification of X-ray images of lung diseases. The use of transfer learning, which allows using already trained network models for new problems, could allow for better results in the COVID-19 disease classification problem. However, at the beginning of the pandemic, there were not very large databases of SARS-CoV-2 positive patient images on which a network could perform learning. A solution to this problem could be a Generative Adversarial Network (GAN) algorithm to create new synthetic data indistinguishable from the real data using the available data set. It would allow training a network capable of performing classification with greater accuracy on a larger and more diverse number of training data. Obtaining such a tool could allow for more efficient research on how to solve the global COVID-19 pandemic problem. The research presented in this paper aims to investigate the impact of using a Generative Adversarial Network for COVID-19-related imaging diagnostics in the classification problem using transfer learning.