A comparative study of gradient nonlinearity correction strategies for processing diffusion data obtained with ultra-strong gradient MRI scanners.

Umesh Rudrapatna,Derek K Jones,Jamie Roberts,Greg D Parker

doi:10.1002/mrm.28464

Abstract

PurposeThe analysis of diffusion data obtained under large gradient nonlinearities necessitates corrections during data reconstruction and analysis. While two such preprocessing pipelines have been proposed, no comparative studies assessing their performance exist. Furthermore, both pipelines neglect the impact of subject motion during acquisition, which, in the presence of gradient nonlinearities, induces spatio‐temporal B‐matrix variations. Here, spatio‐temporal B‐matrix tracking (STB) is proposed and its performance compared to established pipelines.MethodsDiffusion tensor MRI (DT‐MRI) was performed using a 300 mT/m gradient system. Data were acquired with volunteers positioned in regions with pronounced gradient nonlinearities, and used to compare the performance of six different processing pipelines, including STB.ResultsUp to 30% errors were observed in DT‐MRI parameter estimates when neglecting gradient nonlinearities. Moreover, the order in which B0 inhomogeneity, eddy current and gradient nonlinearity corrections were performed was found to impact the consistency of parameter estimates significantly. Although, no pipeline emerged as a clear winner, the STB approach seemed to yield the most consistent parameter estimates under large gradient nonlinearities.ConclusionsUnder large gradient nonlinearities, the choice of preprocessing pipeline significantly impacts the estimated diffusion parameters. Motion‐induced spatio‐temporal B‐matrix variations can lead to systematic bias in the parameter estimates, that can be ameliorated using the proposed STB framework.

Highlights

Recent developments in gradient technology have opened up immense opportunities to further our understanding of brain microstructure.[1,2,3] ultra-strong gradient systems often have compromised gradient linearity.[1,2] Figure 1 illustrates the typical gradient nonlinearities observed with the 300 mT/m gradient Connectom scanner used in this study.Even with careful head positioning, parts of the brain furthest from the isocenter can experience significant differences in gradient amplitude (>5%), leading to image distortion and deviations from the prescribed b-values
While this is benign for head scanning with most clinical scanners, where spatial uniformity of gradient amplitude is more achievable, the strong gradient nonlinearities in bespoke ultra-strong gradient systems has prompted the development of dedicated diffusion data analysis pipelines[10] that account for gradient-nonlinearity-induced spatial variations in B-matrices.[8,11,12,13,14,15]
Pipelines that ignore the impact of gradient nonlinearities on the B-matrix (MGH-B-matrix rotation (BR) and WUBR) fail to replicate MD estimates, while pipelines that account for spatial B-matrix changes are relatively immune to such discrepancies

Summary

Introduction

Recent developments in gradient technology have opened up immense opportunities to further our understanding of brain microstructure.[1,2,3] ultra-strong gradient systems often have compromised gradient linearity.[1,2] Figure 1 illustrates the typical gradient nonlinearities observed with the 300 mT/m gradient Connectom scanner used in this study.Even with careful head positioning, parts of the brain furthest from the isocenter can experience significant differences in gradient amplitude (>5%), leading to image distortion and deviations from the prescribed b-values (which scale with gradient amplitude squared). Even if an ideal preprocessing pipeline that addresses all the confounds listed above were to be developed, as noted above, gradient nonlinearities lead to spatial variations in diffusion weighting While this is benign for head scanning with most clinical scanners, where spatial uniformity of gradient amplitude is more achievable, the strong gradient nonlinearities in bespoke ultra-strong gradient systems has prompted the development of dedicated diffusion data analysis pipelines[10] that account for gradient-nonlinearity-induced spatial variations in B-matrices.[8,11,12,13,14,15] Here, we consider for the first time, the interaction of subject motion with gradient nonuniformity on diffusion measurements, as the effect of diffusion-weighting cannot be captured using spatially varying B-matrices alone. Spatio-temporal tracking of B-matrix at each voxel location is essential

Objectives

Methods

Results

Conclusion