Enhancing Medical Image Super-Resolution with GANs and Residual Attention Mechanisms

Medical image super-resolution is crucial for improving the quality of MRI scans, aiding better diagnosis and analysis. In this paper, we propose an enhanced super-resolution approach based on Generative Adversarial Networks (GANs) that incorporates residual attention mechanisms and perceptual loss functions to achieve high-quality image reconstruction. Specifically, we employ Squeeze-and-Excitation (SE) blocks within the generator network to improve feature learning and integrate perceptual and Structural Similarity Index Measure (SSIM) losses to enhance perceptual and structural quality. Experiments on the IXI MRI dataset demonstrate significant improvements over the baseline SRGAN, achieving a PSNR gain of 2.17 dB and an SSIM improvement of 0.062. Our approach effectively enhances visual fidelity and diagnostic relevance in reconstructed MRI images.