Unsupervised Representation Learning for 3D MRI Super Resolution with Degradation Adaptation

High-resolution (HR) MRI is critical in assisting the doctor's diagnosis and image-guided treatment but is highly time-consuming and costly to acquire. Therefore, deep learning-based super-resolution reconstruction (SRR) has been investigated to generate super-resolution images from low-resolution (LR) images. Training such neural networks requires authentic HR and LR image pairs, which are difficult to acquire due to patient movement during and between the acquisitions of LR and HR images. Rigid movements of hard tissues can be corrected with image registration. In contrast, the alignment of deformed soft tissues is challenging, making it impractical to train neural networks with authentic HR and LR image pairs. Existing studies in the literature focused on SRR using authentic HR images and down-sampled synthetic LR images. Yet, the difference in degradation representations between synthetic and authentic LR images suppresses the quality of SRR from authentic LR images. In this work, we propose a novel Unsupervised Degradation Adaptation Network (UDEAN) to mitigate this problem. Our network consists of the degradation learning network and the SRR network. The degradation learning network down-samples the HR images by addressing the degradation representation of the misaligned or unpaired LR images. The SRR network learns the mapping from the down-sampled HR images to the original ones. Experimental results show that our method outperforms state-of-the-art networks and can potentially be applied in real clinical settings.