Tympanic Membrane Generation with Generative Adversarial Networks

Abstract

Obtaining sufficient original data in most studies in the field of medical pattern recognition is a difficult and time consuming process. Different data augmentation methods are used to increase the amount of data to be used to train these systems. In this study, a generative adversarial networks based system that produces artificial images by using the tympanic membrane images taken from otoscope devices for data augmentation has been designed and trained. Artificial images that has been generated by different generative adversarial networks have been evaluated by specialist physicians with a Visual Turing test. Preliminary results show that artificial medical images can be perceived as real with high confidence scores.