MRI super-resolution using 3D cycle-consistent generative adversarial network

High-resolution magnetic resonance imaging (MRI) provides detailed anatomical information critical for clinical application diagnosis. However, current MRIs are acquired at clinical resolutions due to the limit of physical, technological, and economic considerations. On the other hand, existing approaches require paired MRI images as training data, which are difficult to obtain on existing datasets when the alignment between high and low-resolution images has to be implemented manually.Within the scope of project, we aim to provide an end-to-end system to solve the super-resolution method on 3D MRI. Our proposed method derives from recent neural network developments and does not require paired data for efficient training. By integrating different models with separated functions, our 3D super-resolution CycleGAN (SRCycleGAN) achieved compelling results on MRI volumes. The output is close with ground-truth, showing a low distortion on different scaling factors. Besides, we also compare our method against different GAN-based methods in this field to highlight the performance.