Experimental design for MRI by greedy policy search

TL;DR

This paper introduces a reinforcement learning approach to optimize MRI subsampling strategies, revealing that a greedy method performs nearly as well as complex non-greedy methods due to variance issues in gradient estimates.

Contribution

It demonstrates that a simple greedy policy search can effectively learn MRI subsampling strategies, challenging the assumption that more complex methods are necessary.

Findings

Greedy approximation performs nearly as well as non-greedy methods.

Variance in gradient estimates hampers non-greedy policy adaptation.

Adaptive policies improve MRI subsampling quality.

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.