Learning to Sample MRI via Variational Information Maximization

Abstract

Accelerating MRI scans requires optimal sampling of k-space data. This is however a daunting task due to the discrete and non-convex nature of sampling optimization. To cope with this challenge, we put forth a novel deep learning framework that leverages uncertainty autoencoders to enable joint optimization of sampling pattern and reconstruction of MRI scans. We represent the encoder as a non-uniform Fast Fourier Transform that allows {\it continuous} optimization of k-space samples on a non-Cartesian plane, while the decoder is a deep reconstruction network. Our approach is universal in a sense that it can be used with any reconstruction network. Experiments with knee MRI shows improved reconstruction quality of our data-driven sampling over the prevailing variable-density sampling.