A Comparative Study of CNN-Based Super-Resolution Methods in MRI Reconstruction.

Super-resolution (SR) and MRI reconstruction are two hot topics in the field of computer vision and medical imaging, respectively. Both of them have been attacked by the recent deep learning methodology. This work aims to investigate some typical CNN-based SR algorithms and their applications in MRI reconstruction. By dividing the whole network in MRI reconstruction to be data-consistency sub-network and image prior sub-network, we investigate the reconstruction performance by utilizing various CNN-based SR networks. Experimental results demonstrate that the ResNetlike and DenseNet-like hybrid SR networks can obtain very significantly superior performance than current state-of the-art approaches.