A Fast and Efficient Algorithm for Reconstructing MR images From Partial Fourier Samples

TL;DR

This paper introduces a fast, whole-image reconstruction algorithm for MR images from partial Fourier data, leveraging sparsity and ADMM, outperforming patch-based methods in efficiency and quality.

Contribution

It presents a novel whole-image reconstruction method using ADMM that avoids patch division, improving efficiency and performance over existing patch-based algorithms.

Findings

The proposed method achieves higher reconstruction quality.

It significantly reduces computational time.

Experimental results confirm its advantages over existing methods.

Abstract

In this paper, the problem of Magnetic Resonance (MR) image reconstruction from partial Fourier samples has been considered. To this aim, we leverage the evidence that MR images are sparser than their zero-filled reconstructed ones from incomplete Fourier samples. This information can be used to define an optimization problem which searches for the sparsest possible image conforming with the available Fourier samples. We solve the resulting problem using the well-known Alternating Direction Method of Multipliers (ADMM). Unlike most existing methods that work with small over-lapping image patches, the proposed algorithm considers the whole image without dividing it into small blocks. Experimental results prominently confirm its promising performance and advantages over the existing methods.