High-quality magnetic resonance (MR) image, i.e., with near isotropic voxel spacing, is desirable in various scenarios of medical image analysis. However, many MR images are acquired using good in-plane resolution but large spacing between slices in clinical practice. In this work, we propose a novel deep-learning based super-resolution algorithm to generate high-resolution (HR) MR images of small slice spacing from low-resolution (LR) inputs of large slice spacing. Notice that real HR images are needed in most existing deep-learning-based methods to supervise the training, but in clinical scenarios, usually they will not be acquired. Therefore, our unique goal herein is to design and train the super-resolution network without real HR ground-truth. Specifically, two-staged training is used in our method. In the first stage, HR images of reduced slice spacing are synthesized from real LR images using variational auto-encoder (VAE). Although these synthesized HR images of reduced slice spacing are as realistic as possible, they may still suffer from unexpected morphing induced by VAE, implying that the synthesized HR images cannot be paired with the real LR images in terms of anatomical structure details. In the second stage, we degrade the synthesized HR images to generate corresponding LR-HR image pairs and train a super-resolution network based on these synthesized pairs. The underlying mechanism is that such a super-resolution network is less vulnerable to anatomical variability. Experiments on knee MR images successfully demonstrate the effectiveness of our proposed solution to reduce the slice spacing for better rendering. (c) 2021 Elsevier Ltd. All rights reserved.