Experimental design for MRI by greedy policy search

Abstract

In today's clinical practice, magnetic resonance imaging (MRI) is routinely accelerated through subsampling of the associated Fourier domain. Currently, the construction of these subsampling strategies - known as experimental design - relies primarily on heuristics. We propose to learn experimental design strategies for accelerated MRI with policy gradient methods. Unexpectedly, our experiments show that a simple greedy approximation of the objective leads to solutions nearly on-par with the more general non-greedy approach. We offer a partial explanation for this phenomenon rooted in greater variance in the non-greedy objective's gradient estimates, and experimentally verify that this variance hampers non-greedy models in adapting their policies to individual MR images. We empirically show that this adaptivity is key to improving subsampling designs.

arXiv commentAccepted to NeurIPS 2020 (spotlight), 15-12-2020: Fixed typos, Figure 9, and pseudocode

References in corpus (12)

• What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
• Deep Convolutional Neural Network for Inverse Problems in Imaging
• Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
• fastMRI: An Open Dataset and Benchmarks for Accelerated MRI
• Advancing machine learning for MR image reconstruction with an open competition: Overview of the 2019 fastMRI challenge
• Self-Supervised Deep Active Accelerated MRI
• Invert to Learn to Invert
• Learning Sampling and Model-Based Signal Recovery for Compressed Sensing MRI
• Estimating Gradients for Discrete Random Variables by Sampling without Replacement
• Beyond Adaptive Submodularity: Approximation Guarantees of Greedy Policy with Adaptive Submodularity Ratio
• i-RIM applied to the fastMRI challenge
• Compressed Sensing: From Research to Clinical Practice with Data-Driven Learning