Anisotropic Super Resolution in Prostate MRI using Super Resolution Generative Adversarial Networks.

摘要: Acquiring High Resolution (HR) Magnetic Resonance (MR) images requires the patient to remain still for long periods of time, which causes discomfort and increases probability motion induced image artifacts. A possible solution is acquire low resolution (LR) process them with Super Generative Adversarial Network (SRGAN) create a super-resolved version. This work applies SRGAN MR prostate performs three experiments. The first experiment explores improving in-plane by factors 4 8, shows that, while PSNR SSIM (Structural SIMilarity) metrics are lower than isotropic bicubic interpolation baseline, able that have high edge fidelity. second anisotropic super-resolution via synthetic images, in input network anisotropically downsampled versions HR images. demonstrates ability modified perform super-resolution, quantitative comparable those baseline. Finally, third version super-resolve obtained from through-plane slices volumetric data. output contain significant amount frequency information make visually close their counterparts. Overall, promising results each show successful technique producing volumes an achievable goal.