Application of Generative Adversarial Networks on radiological images of COVID-19

Abstract

Image classification has been applied to several real problems such as remote data analysis, facial recognition, disease detection, among others. However, obtaining labeled data is a costly task, as it requires time, resources and specialists. In addition, some problems are very specific and present a minority class. These scenarios are challenging and degrade the performance of machine learning algorithms. In these cases, we can use data augmentation (DA) approaches to increase the number of examples labeled in a data set. The objective of this work is to analyze the use of Generative Adversarial Networks (GANs), which are networks capable of synthesizing artificial data from the original data, under an adversarial process of two neural networks. The GANs will be applied in the classification of radiological images of COVID-19 in order to improve the problem of unbalanced classes. (AU)