Purpose: Deep learning (DL)-based super-resolution (SR) reconstruction for MRI has recently been receiving attention, however challenges hindering the implementation of these approaches remain. Low-resolution (LR) MRIs captured in the clinic exhibit complex tissue structures obfuscated by noise while training a robust network for a SR task requires abundant, perfectly matched pairs of LR and high-resolution (HR) images. We present the development of a novel MRI SR technique based on the concept of cascaded DL that allows for SR reconstruction in the presence of insufficient training data, an unknown translation model, and noise.
Methods: The proposed cascaded DL framework consists of three components: 1) a de-noising auto-encoder (DAE) trained using clinical LR noisy MRI scans processed with a non-local means filter that generates denoised LR data; 2) a down-sampling network (DSN) trained with a small amount of paired LR/HR data from volunteers that allows for the generation of perfectly paired LR/HR data; and 3) the proposed SR generative model (p-SRG) trained with data generated by the DSN that maps from LR inputs to HR outputs. After training, LR clinical images may be fed through the DAE and p-SRG to yield SR reconstructions of the LR input. The application of this framework was explored in two settings: 3D LR breath-hold MRI scans (< 3 sec/vol) and 4D-MRI LR acquisitions (0.5 sec/vol).
Results: The DSN produces LR scans from HR inputs with a higher fidelity to true LR clinical scans compared to conventional k-space down-sampling methods. Furthermore, HR outputs generated by the p-SRG exhibit considerably improved results compared to conventional approaches to MRI SR.
Conclusion: Four-fold enhancements in spatial resolution using the proposed method facilitate target delineation and motion management during radiation therapy, enabling precise MRI-guided radiation therapy with 3D breath-hold MRI within 3 seconds and 4D-MRI within 2 frames/sec.