Subdivision Features-Guided Brain MRI Super-Resolution via Forward and Backward Propagation

There have been many models designed for super-resolution (SR) tasks. However, the universal model rarely takes into account the potential characteristics of magnetic resonance imaging (MRI). Further, for real-time MRI, SR is a critical technology to improve spatial resolution while maintaining temporal resolution. Trying not to change the existing well-established models, we propose a subdivision feature-guided enhancement module which can be attached to the arbitrary generative networks for the brain MRI SR tasks. The experimental results proved the effectiveness of our designed module.