MRI Synthesis via CycleGAN-Based Style Transfer
Abha Gawande, Parshav Gandhi, Sharli Kude, Nikahat Mulla, Renuka Pawar · 2023
Medical imaging plays a crucial role in modern medicine by supporting illness diagnosis and therapy. However, there are difficulties in illness diagnosis using medical imaging, such as the costs and infrastructure to conduct an MRI scan, and the strenuous task of correctly interpreting an MRI scan. It is necessary to develop a method for producing MRI images with adjustable contrast levels. A cyclic Generative Adversarial Neural Network called CycleGAN, which performs well with unpaired data and is best suited for medical imagery, is used for the proposed healthcare platform to create artificial MRI scans. The results show that our method converts non-fat saturated MRI images into fat saturated images more effectively than a benchmark pix2pix approach, resulting in a PSNR (peak signal- to-noise ratio) of 12.89, structural similarity of 0.97, and KL divergence of 0.07, using the MRI dataset, a sizeable collection of freely accessible MR brain pictures.