REMIX: Resolution Enhancement through Mixture of Experts

Sergio Morell-Ortega, Marina Ruiz Perez, Pierrick Coupe, Jose V Manjon

Keywords: Super-resolution, MRI, Deep Learning, Mixture Of Experts

Abstract: We propose a novel method to enhance the resolution of magnetic resonance images (MRI) using deep learning. Our approach is based on realistic MRI data degradation using real affine registrations. We demonstrate the efficacy of a Mixture of Experts approach in handling diverse input resolutions commonly present in clinical settings. This is done by training seven networks, each one for a specific native volume resolution, allowing more effective handling of low-resolution images when compared to a unique model.

Submission Number: 71