The classification of brain tumors using deep learning techniques has emerged as a pivotal area of medical research and diagnostics. In this study, we present an innovative approach to brain tumor classification through the application of Generative Adversarial Networks (GAN). Specifically, we have devised a GAN model with a modified ResNet architecture in the generator and a DenseNet architecture in the discriminator. By improving the precision and effectiveness of brain tumor classification, this innovative design offers substantial breakthroughs in the field of medical imaging by utilizing the capabilities of generative and discriminative networks. The generator, based on a modified ResNet, is designed to create realistic and high-resolution brain tumor images. It learns to generate synthetic brain scans that mimic the characteristics of actual tumor images, thus contributing to data augmentation and diversification. This augmentation process is crucial for training deep learning models effectively, especially when the availability of medical images is limited. The discriminator, on the other hand, employs a DenseNet architecture to distinguish between real brain tumor images and the synthetic ones generated by the ResNet-based generator. The DenseNet's ability to capture intricate details and features in medical images ensures that the discriminator can effectively discern between genuine and synthetic data, contributing to the GAN's overall learning process. Our proposed GAN model is trained on a diverse and well-curated dataset of brain tumor images, enabling it to identify and classify various tumor types and their characteristics with remarkable accuracy. The generated synthetic images aid in improving the model's ability to generalize and adapt to new, unseen data, thereby enhancing its performance in brain tumor classification tasks. © 2024 Elsevier B.V., All rights reserved.