Generative Adversarial Networks for Data Augmentation in X-Ray Medical Imaging
Authors: Iqball, T. and Wani, M.A.
Volume: 217
Pages: 341-355
DOI: 10.1007/978-3-030-91390-8_14
Abstract:This work focuses on the use of Generative Adversarial Networks for data augmentation in X-ray medical imaging. Data augmentation can be employed in situations where little data or imbalanced datasets are present. There are two main reasons why some medical datasets are limited or imbalanced: either there is little data available for some rare diseases, or the privacy policy of medical organizations does not allow it to share the data. But deep learning models often need a large and balanced dataset for efficient training of models, which can produce high accuracy results. A Progressively Growing Generative Adversarial Network (PGGAN) is proposed for data augmentation of X-ray images to mitigate the above problem. Extensive experimentation has shown that PGGAN generates good quality synthetic X-ray images for data augmentation to balance the dataset. The resulting balanced dataset used several classification models for testing. Various state-of-the-art classification models are adopted in transfer learning and fine-tuned to test the augmentation process.
Source: Scopus