We present a novel approach to synthesise high-resolution isotropic 3D abdominal MR images, from anisotropic 3D images in an unpaired fashion. Using a modified CycleGAN architecture with a gradient mapping loss, we leverage disjoint patches from the high-resolution (in-plane) data of an anisotropic volume to enforce the network generator to increase the resolution of the low-resolution (through-plane) slices. This will enable accelerated whole-abdomen scanning with high-resolution isotropic images within short breath-hold times.