Deep learning is rapidly advancing the medical field by improving the speed and accuracy of diagnoses, enabling personalized treatment plans, and automating traditionally manual tasks through innovative model architectures. However, training these models effectively often requires large, labeled datasets—something that remains challenging due to limited data availability and strict privacy regulations that hinder data sharing. Generative Adversarial Networks (GANs), originally designed for image-to-image translation, have shown promise in generating synthetic data that closely resembles real datasets. In this study, we propose an enhanced approach for generating synthetic brain MRI data using a modified SinGAN model, augmented with a dual-attention mechanism, to better handle the noise and graininess commonly found in such images.