Covid-19 detection and overcome the scarcity of chest X-Ray datasets based on transfer learning and GAN model

Diseases which affect the respiratory system are considered some of the most dangerous, since defects in the breathing process may lead to death. Currently, the coronavirus is one of the most complex strains of the corona family of diseases for which, as yet, there is no successful vaccine. Therefore, detecting the virus is of the utmost importance to global health. The use of deep learning techniques is considered to be a successful method by which to diagnose such diseases. However, there is a lack of data which models deep learning techniques required for training. In this paper, we suggest a method for data augmentation based on the CGAN model. To synthesize realistic chest X-ray images, we proposed to set edge information of the image to the generator, which is used as supporting information to increase the reality of the generated images. A MobileNet CNN model was used for diagnosing COVID-19. When this suggestion was applied to chest data, the diagnostic results were very satisfactory which the accuracy exceeds 99% in some cases.