High resolution image creation and evaluation with adversitive generative networks(GAN)

Accurate and reliable progression of disease diagnosis and treatment processes is crucial in biomedical imaging. However, biomedical images are sometimes of low resolution, which can negatively affect disease detection. For the accurate progression of treatment processes, it is necessary to work with high-resolution images or convert low-resolution images into high-resolution ones. In our study, the SRGAN model, one of the GAN architectures, was utilized to enhance and detail low-resolution color images into high-resolution versions. SRGAN provides an effective method for improving visual quality by achieving high performance in image enhancement and transformation processes. The datasets used in this thesis consist of skin cancer, blood cell cancer, and retinal fundus datasets, which are widely used in the field of medical imaging. In the skin cancer dataset, fine details on the skin surface were visualized more clearly, while in the blood cell cancer dataset, cell structures were highlighted to support diagnosis. In the retinal fundus dataset, vascular structures of the eye were detailed and resolution was enhanced to make them suitable for diagnostic analysis. To evaluate the performance of the model, Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) metrics were employed. The performance evaluation showed that the SRGAN model achieved promising results with a PSNR of 31.06 and an SSIM of 85.84% on the skin cancer dataset, a PSNR of 30.97 and an SSIM of 88.18% on the blood cell cancer dataset, and an SSIM of 94.30% and a PSNR of 30.71 on the retinal fundus dataset. The findings demonstrate that the model exhibits superior performance in terms of both structural and visual quality.