Medical Image Compression Using Generative Adversarial Networks

Abstract

Medical images need to be efficiently compressed before transmission and storage due to the storage capacity and constrained bandwidth issues. An ideal image compression system must yield a high compression ratio with good quality compressed images. Machine learning models are proposed to perform tasks, whereas humans have difficulties in completing. In this paper, machine learning algorithms   such   as   Generative    Adversarial    Networks    (GANs), Conditional Generative Adversarial Networks (CGANs) and Deep Convolutionals Generative Adversarial Networks (DCGANs) are trained to relate the medical image contents to their compression ratio. The comparison of different methods is evaluated using various evaluation metrics such as PSNR, MSE, MAE, Compression Ratio, Compression Time and Decompression Time.