This paper reviews the application of Generative Adversarial Networks (GANs) in medical imaging, highlighting their potential in various tasks such as image reconstruction, segmentation, classification, and cross-modality synthesis. GANs, which consist of a generator and a discriminator, have gained attention due to their ability to generate new samples without explicitly modeling the probability density function. The adversarial loss provided by the discriminator helps incorporate unlabeled samples and enforce higher-order consistency, making GANs useful in domain adaptation, data augmentation, and image-to-image translation. The review covers the background of GANs, including the vanilla GAN, its variants, and architectural improvements. It then delves into the applications of GANs in medical imaging, categorizing them into tasks such as reconstruction, image synthesis, segmentation, classification, detection, and registration. Each category is further detailed with examples and references to relevant studies. Key findings include the popularity of GANs in image synthesis, particularly cross-modality synthesis, and their effectiveness in improving segmentation performance through adversarial training. The paper also discusses challenges such as the lack of reliable evaluation metrics for medical images and the potential bias in unpaired image-to-image translation. The authors conclude by identifying future challenges, including the need for more robust evaluation metrics and addressing domain bias in unpaired translation. They emphasize the potential of GANs in synthesizing rare pathology cases and their role in data augmentation, especially for rare diseases.This paper reviews the application of Generative Adversarial Networks (GANs) in medical imaging, highlighting their potential in various tasks such as image reconstruction, segmentation, classification, and cross-modality synthesis. GANs, which consist of a generator and a discriminator, have gained attention due to their ability to generate new samples without explicitly modeling the probability density function. The adversarial loss provided by the discriminator helps incorporate unlabeled samples and enforce higher-order consistency, making GANs useful in domain adaptation, data augmentation, and image-to-image translation. The review covers the background of GANs, including the vanilla GAN, its variants, and architectural improvements. It then delves into the applications of GANs in medical imaging, categorizing them into tasks such as reconstruction, image synthesis, segmentation, classification, detection, and registration. Each category is further detailed with examples and references to relevant studies. Key findings include the popularity of GANs in image synthesis, particularly cross-modality synthesis, and their effectiveness in improving segmentation performance through adversarial training. The paper also discusses challenges such as the lack of reliable evaluation metrics for medical images and the potential bias in unpaired image-to-image translation. The authors conclude by identifying future challenges, including the need for more robust evaluation metrics and addressing domain bias in unpaired translation. They emphasize the potential of GANs in synthesizing rare pathology cases and their role in data augmentation, especially for rare diseases.