Enhancing Medical Image Clarity: Introducing L2R-SRGAN for Improved Super-Resolution

Artificial Intelligence (AI) has seen the emergence of Generative Adversarial Networks or abbreviated as GANs, a promising technology that mimics the actual world inputs to generate synthetic data. This capacity is quite promising, especially in the field of biomedical imaging, where increasing the resolution of X-rays and other medical picture data may significantly enhance clinical procedures and diagnosis. However, because medical technology has limits, obtaining high-resolution pictures of medical conditions can be difficult. There are now proven super-resolution (SR) methods available to address this issue. Rebuilding high-resolution images from their low-resolution counterparts is one of these methods, which provides more detailed information necessary for precise target identification. Super-resolution’s primary objective is to raise the image’s perceived quality. Employing generative adversarial network (GAN) based high-resolution representation learning, more precisely the SRGAN model, this study offers a unique approach for SR or super-resolution of medical images. The suggested technique, called L2RSRGAN, improves upon previous models by utilizing GANs’ strengths in maintaining high-frequency picture information and recreating visual quality. Notably, L2 regularization is integrated into the generator’s residual block, which smoothest out artifacts and reduces overfitting to improve the visual quality of generated high-resolution pictures. Our studies’ findings demonstrate that our model is effective in generating high-quality super-resolved medical pictures, with a SSIM or Structural Similarity Index Measure of 0.2098 and PSNR or peak signal-to-noise ratio of 9.18.