Denoising Generalized Expectation-Consistent Approximation for MR Image Recovery

TL;DR

This paper introduces a novel plug-and-play algorithm for MR image recovery that uses a generalized expectation-consistent approximation to improve denoising performance by leveraging predictable error statistics, outperforming existing methods.

Contribution

It proposes a new PnP algorithm based on GEC approximation for Fourier-based operators, enhancing MR image recovery with better error modeling and a specialized DNN denoiser.

Findings

Outperforms existing PnP and AMP methods in MR image recovery

Provides predictable error statistics at each iteration

Leverages a new DNN denoiser tailored to error statistics

Abstract

To solve inverse problems, plug-and-play (PnP) methods replace the proximal step in a convex optimization algorithm with a call to an application-specific denoiser, often implemented using a deep neural network (DNN). Although such methods yield accurate solutions, they can be improved. For example, denoisers are usually designed/trained to remove white Gaussian noise, but the denoiser input error in PnP algorithms is usually far from white or Gaussian. Approximate message passing (AMP) methods provide white and Gaussian denoiser input error, but only when the forward operator is sufficiently random. In this work, for Fourier-based forward operators, we propose a PnP algorithm based on generalized expectation-consistent (GEC) approximation -- a close cousin of AMP -- that offers predictable error statistics at each iteration, as well as a new DNN denoiser that leverages those statistics. We apply our approach to magnetic resonance (MR) image recovery and demonstrate its advantages over existing PnP and AMP methods.