Generative adversarial networks (GANs) have emerged as a powerful technique in medical imaging for addressing challenges related to data scarcity and class imbalance. This paper reviews recent advancements in GAN-based data augmentation, specifically focusing on medical image generation and reconstruction. GANs have been employed to generate high-quality synthetic images that enhance the performance of deep learning models in tasks such as pneumonia detection and MRI reconstruction. Studies have shown that the application of GANs, including advanced architectures like HARA-GAN and ProGAN, significantly improves the sensitivity, specificity, and structural similarity of models trained on augmented datasets. For instance, the sensitivity of thyroid classification models improved from 76.8 to 84.2% when using GAN-augmented data (Islam in IEEE Access 8:200567–200580, 2020). Furthermore, GANs have been shown to generate high-fidelity images for physician training and anomaly detection (Zhang et al. in Medical Image Synthetic Data Augmentation Using GAN. CSAE, New York, NY, USA, 2020; Islam in IEEE Access 8:200567–200580, 2020). This paper provides a comprehensive analysis of 25 key research papers, discussing their methodologies and results, along with future directions for GAN applications in medical imaging.