Algorithmic process that converts an image from a source domain to another different target domain (for example, convert the image of an orange to an image of an apple).
Published in Chapter:
Generation of Novel Synthetic Portable Chest X-Ray Images for Automatic COVID-19 Screening
Daniel Iglesias Morís (Grupo VARPA, CITIC, INIBIC, Universidade da Coruña, Spain), Joaquim de Moura (Grupo VARPA, CITIC, INIBIC, Universidade da Coruña, Spain), Jorge Novo (Grupo VARPA, CITIC, INIBIC, Universidade da Coruña, Spain), and Marcos Ortega (Grupo VARPA, CITIC, INIBIC, Universidade da Coruña, Spain)
Copyright: © 2022
|Pages: 34
DOI: 10.4018/978-1-6684-2304-2.ch008
Abstract
The diagnosis and the study of the evolution of COVID-19 is crucial to tackle the challenge that this disease represents for healthcare services. Chest x-ray imaging allows us to visualize the pulmonary regions, where COVID-19 causes its main affectation. In order to reduce the risk of cross-contamination, a crucial aspect in the pandemic, portable chest x-ray devices are advantageous being easier to decontaminate in comparison with the fixed machinery, despite offering a lower image quality. Furthermore, the recent emergence of COVID-19 implies a data scarcity that must be tackled. In this chapter, the authors present the analysis of a strategy that generates novel synthetic portable chest x-ray images using the CycleGAN, an architecture for image translation that is trained with unpaired data. The novel set of images is then added to the original dataset, improving the performance of the classification model.