Robust Multi-echo GRE Phase processing using a unity rank enforced complex exponential model

TL;DR

This paper introduces a novel complex exponential model with a unity rank prior for multi-echo GRE phase processing, improving robustness against phase unwrapping errors, background effects, and noise, leading to enhanced susceptibility imaging.

Contribution

The paper proposes a rank-minimizing optimization framework using URA prior for GRE phase processing, enabling more accurate susceptibility feature recovery under challenging conditions.

Findings

Enhanced venous structure visualization in SWI

Reduced artifacts and errors in quantitative susceptibility maps

Detection of iron deposits not visible with standard methods

Abstract

Purpose: Develop a processing scheme for Gradient Echo (GRE) phase to enable restoration of susceptibility-related (SuR) features in regions affected by imperfect phase unwrapping, background suppression and low signal-to-noise ratio (SNR) due to phase dispersion. Theory and Methods: The predictable components sampled across the echo dimension in a multi-echo GRE sequence are recovered by rank minimizing a Hankel matrix formed using the complex exponential of the background suppressed phase. To estimate the single frequency component that relates to the susceptibility induced field, it is required to maintain consistency with the measured phase after background suppression, penalized by a unity rank approximation (URA) prior. This is formulated as an optimization problem, implemented using the alternating direction method of multiplier (ADMM). Results: With in vivo multi-echo GRE data, the magnitude susceptibility weighted image (SWI) reconstructed using URA prior shows additional venous structures that are obscured due to phase dispersion and noise in regions subject to remnant non-local field variations. The performance is compared with the susceptibility map weighted imaging (SMWI) and the standard SWI. It is also shown using numerical simulation that quantitative susceptibility map (QSM) computed from the reconstructed phase exhibits reduced artifacts and quantification error. In vivo experiments reveal iron depositions in insular, motor cortex and superior frontal gyrus that are not identified in standard QSM. Conclusion: URA processed GRE phase is less sensitive to imperfections in the phase pre-processing techniques, and thereby enable robust estimation of SWI and QSM.