A Comparative Study of Generative Adversarial Networks (GANs) in Medical Image Processing: A Review

Abstract

Generative Adversarial Networks (GANs) have come to be as powerful in medical image processing with massive benefits in image quality, fusion, classification and segmentation tasks. This paper displays an in-depth analysis of GAN structures and their use cases in medical image analysis for data augment, anomaly detection, cross- modality synthesis, super-resolution, and image reconstruction. As the demand for automated diagnostic systems has increased, GANs provide an efficient way to synthesize realistic medical images, particularly in domains with limited data availability. In this review, we discuss new developments and issues on how to apply GANs to accurate and effective medical image analysis. Moreover, this work explores the strengths, limitations, and comparative performance of different GAN models across diverse datasets and clinical tasks. By identifying key differences among the GAN models, and analysing performance, this review will be a roadmap for future studies in developing GAN-based models for better diagnosis and health applications.