Quantitative chemical shift-encoded (CSE) MRI techniques acquire complex-valued (magnitude and phase) images at multiple echo times (TE), enabling simultaneous mapping of fat-fraction, R2* (=1/T2*) and B₀ field. Applications of CSE-MRI include tissue fat quantification, iron quantification and quantitative susceptibility mapping (QSM). However, these techniques are vulnerable to unanticipated phase shifts in the acquired CSE images. Recently, phase shifts due to concomitant gradients (CG) [1] have been identified as a source of error for quantitative CSE techniques [2], however their effects on fat-fraction, R2* and B₀ maps are not well understood. Therefore, the purpose of this study was to characterize the effect of phase shifts due to CGs on CSE-based fat-fraction, R2* and B₀ measurements, at both 1.5T and 3T

In this study the phase shifts due to CGs were analyzed for different versions of a multi-echo gradient-echo pulse sequence: a monopolar flyback single echo train, monopolar flyback with two interleaved echo trains, and a bipolar single echo train, as shown in Fig. 1. The phase shifts due to CGs can be calculated from the known gradients [1]. Simulations were used to characterize the CG phase shifts and experimental measurements were used to demonstrate the correction of these shifts before parameter estimation.

Simulations

In our 3T simulations, complex data from a homogeneous water signal (fat-fraction=0%) were synthesized, and CG phase shifts were added. Representative CG phase shifts were calculated using gradients for a 3D six-echo CSE gradient echo sequence with an axial acquisition and a 48 cm z-axis field of view (FOV) for the three waveforms shown in Fig. 1. Subsequently, a complex fitting algorithm [3] was used to estimate the fat-fraction, R2*, and B₀ field map.

MRI Measurements

Measurements were made on a 1.5T Signa HDxt GE scanner and a 3T Discovery MR 750 GE scanner using an 8-channel and 32-channel torso coil, respectively. A NiCl₂ and NaCl doped-water phantom with 0% fat-fraction and homogenous R2* was scanned. At 3T, the scan was performed with FOV=210 x 210 x 480 cm (X x Y x Z), 6 echoes, TE₁=1.2ms, and ∆TE=0.97ms. At 1.5T two acquisitions with interleaved echo trains were made sequentially, one with FOV=210 x 210 x 480 cm, 6 echoes, TE₁=1.2ms, and ∆TE=0.98ms and a second with FOV=350 x 350 x 480 cm, 6 echoes, TE₁=0.9ms, and ∆TE=0.70ms. The strategy behind these two scans was to maintain the same B₀ field since the true field map inhomogeneity is unknown. This was accomplished by keeping the pre-scans and shims constant. Changing the FOV with the back to back scans results in different CG phase shifts and will produce different estimates of the field maps from the raw data.

Next, the CG phase shifts were calculated numerically using the known gradient waveforms. Then the fat-fraction and R2* maps were calculated from the raw data and data with the calculated CG phase shifts removed. At 1.5T, field maps were estimated from both sets of raw data and after the corresponding CG phase shifts were removed.

Fig. 2a demonstrates that two interleaved echo trains leads to a nonlinear accumulation of phase error, whereas the other sequences accumulate error linearly with respect to echo number. These simulations demonstrate that the fat-fraction in Fig. 2a and R2* in Fig. 2b are only affected with a two shot interleaved acquisition, but significant errors accumulate in the B₀ field map estimate in Fig. 1d for all acquisition types. The phantom experiments demonstrate that CG correction also results in improved estimates of the fat-fraction (Figs. 3a and 4), and R2* (Fig. 3b) for this phantom. Fig. 3c also demonstrates that CGs distort the field map estimate differently depending on the FOV, but removal of CG phase shifts in the sequential scans results in similar field maps with only a small constant offset.Nonlinear phase errors that result from interleaved echo train acquisitions will corrupt the simultaneous estimation of fat-fraction and R2*. These changes can be relatively large at off-isocenter locations (eg. >10% fat-fraction as shown in Fig. 4). CG correction of experimental data demonstrates that the detrimental effects of CG phase shifts can be removed before reconstruction to produce more accurate estimates of the fat-fraction, R2*, and field map measurements.