Implementation of a Cyclegan Model for MRI Image Translation

Image-to-image translation has gained popularity within the medical field to transform images from one domain to another. Medical image synthesis via domain transformation is advantageous in its ability to generate images for instances where an abundance of images for a given class is limited, while protecting individuals' privacy. This study proposes the development of a CycleGAN model for translating neuroimages from a field strength of 3 Tesla to 0.5 Tesla. This model was compared to a DCGAN architecture. The CycleGAN model was able to generate the synthetic and reconstructed images with some success. The mapping function from domain A to domain B performed optimally with an average PSNR value of 30.02 dB and an MAE value of 589.73. The DCGAN model performed the best for the 3T DTI scans, however, the 1.5T model experience mode collapse. Future investigations could implement a model to classify the synthetic images.