Biomedical Image Super-Resolution Using SRGAN: Enhancing Diagnostic Accuracy

Abstract

The super-resolution method, which has gained significant popularity today, aims to obtain high-resolution images from low-resolution ones, enhancing image quality and making details clearer. This technique allows for more detailed analysis of images, providing significant advantages in medical imaging, restoration of old photographs, and the analysis of security cameras. In medical imaging, super-resolution contributes to more accurate diagnosis of diseases by clarifying low-resolution MRI, CT, and ultrasound images. Similarly, in the restoration of old photographs, improving blurred visuals allows for the preservation and renewal of historically significant images. In the field of security, enhancing images obtained from low-resolution surveillance cameras makes it easier to identify suspects and allows for a more detailed analysis of events, playing a critical role in solving crimes. In recent years, deep learning-based approaches have made significant progress in the field of super-resolution. Notably, Convolutional Neural Networks (CNN) have achieved great success in solving these problems. However, one of the most remarkable developments in super-resolution is the SRGAN model, based on Generative Adversarial Networks (GAN). SRGAN has surpassed traditional methods by more effectively improving image quality. In this study, the SRGAN model was trained on three different biomedical datasets, achieving PSNR values of 31 and SSIM values of up to 94%. These results demonstrate the potential of super-resolution in enhancing biomedical imaging, offering clearer images for more accurate disease diagnosis, thereby improving the precision of medical analyses. Moreover, given that these developments can also be applied in fields such as security and restoration, the importance of super-resolution techniques across different disciplines is increasingly recognized.