The echo planar imaging (EPI) Nyquist ghost often requires complex 2D phase error corrections in order to be robustly removed. Several methods exist but have not yet been systematically evaluated in high b-value diffusion-weighted (DW) EPI, where lower signal-to-noise ratios may affect the phase error estimation. We explore here the influence of different 2D phase-error corrected reconstruction methods on quantitative parameters derived from DW-EPI, and demonstrate that errors in parameter estimations relating to the Nyquist ghost can persist even after 2D phase-error correction.
Echo planar imaging (EPI) suffers from Nyquist ghost artefacts owing to alternating readout gradient polarities: gradient system imperfections cause a phase shift between opposing k-space trajectories which produces a ghost image offset by half the field-of-view (FOV)1. The 1D phase correction methods commonly implemented only correct shifts along the readout direction, and as such are unable to fully suppress the artefact when higher order phase errors are present2. Importantly, quantitative MRI methods based on EPI readout, such as diffusion weighted imaging (DWI), may be affected by incomplete corrections3. Several 2D phase correction methods have been proposed4,5,6; however the associated scan time increase, noise amplification or requirement for off-line reconstruction can be problematic. Crucially, 2D phase-corrected reconstruction methods have not been systematically studied in multi-shell high angular DWI (HARDI) with high b-values, where a low signal-to-noise ratio (SNR) may impact phase error estimations. Their widespread use in standard reconstruction pipelines has therefore been limited.
This work explores the extent to which widely used diffusion model parameters are influenced by different 2D phase-corrected reconstruction methods with varying requirements for reference scans.
Image Acquisition and Reconstruction. Multi-shell DW-EPI were acquired (3T Philips Ingenia CX) using the vendor’s 32-channel headcoil on 5 healthy volunteers (3 female; age=37±12years) with b=[1000,2000,3000]s/mm2 and 30 diffusion directions per shell; 9 volumes were acquired with b=0s/mm2. SENSE acceleration=2 was used, with resolution=2x2x2mm3, TR/TE=6848/85ms, bandwidth=3281Hz. A dual acquisition of each volume was acquired with opposing readout gradient polarities. Image reconstruction was performed off-line on the vendor’s 1D phase-corrected raw k-space data using Matlab; each dataset was reconstructed using five different methods (Figure 1). All datasets were inspected for artefacts: no corrections for motion or eddy currents were applied in order to avoid confounding the effects of post-processing with the reconstruction methods.
Image Analysis. SNR maps for each reconstruction were approximated using the non-DW volumes. Artefact power (AP) maps7 were generated for the same volumes, with S the voxel intensity and N the total voxel number:
$$AP=\Bigg|\left(\frac{\left( S-S_{dual}\right)^2}{S_{dual}^2}\right)^{\frac{1}{2}}-\frac{1}{N}\sum^N_{j=1}\left(\frac{\left(S\left( j\right)-S_{dual}\left( j\right)\right)^2} {S_{dual}\left( j\right)^2}\right)^{\frac{1}{2}}\Bigg|$$
Parameter maps, including fractional anisotropy (FA), mean diffusivity (MD) and mean kurtosis (MK) from DKI8, and orientation dispersion index (ODI) from the NODDI model9, were generated as test metrics for each reconstruction method and evaluated against the reference dual acquisition using the relative difference:
$$d = \frac{ S-S_{dual}}{S_{dual}}$$
Of the phase-corrected reconstructions implemented, REFB0 was most effective in reducing the influence of Nyquist ghost artefacts on parameter estimates. The additional scan time required for the reference image is minimal, and overall the correction is fast and easy to combine with simultaneous multi-slice acceleration without the need for complex reconstruction algorithms. The potential bias towards overestimation of MD requires further exploration, but could be related to the lack of correction applied to the DW volumes.
Optimisations to the PEC-SENSE method, such as phase maps estimated from low b-value data alone, could reduce the influence of SENSE noise amplification and improve the feasibility of this method. This could be relevant for different applications; intravoxel incoherent motion (IVIM) imaging, for example, utilises low b-value data that may require phase corrections that a non-DW reference scan cannot correct.
The influence of 2D phase-corrected reconstruction methods on diffusion parameters is therefore non-trivial: the tested parameters appear variably affected by the different reconstructions, and inconsistencies in expected estimates have the potential to mask pathologically relevant changes. This could impact multi-centre trials, where manifestations of the ghost may differ between centres.
1. Bernstein MA, King KF, Zhou X. Handbook of MRI pulse sequences. Elsevier. 2004.
2. Aldefeld B, Börnert P. Effects of gradient anisotropy in MRI. Magn Reson Med. 1998;39(4):606-614.
3. Porter DA, Calamante F, Gadian DG, Connelly A. The effect of residual Nyquist ghost in quantitative echo-planar diffusion imaging. Magn Reson Med. 1999;42(2):385-392.
4. Kellman P, McVeigh ER. Ghost artifact cancellation using phased array processing. Magn Reson Med. 2001;46(2):335-343.
5. Xie VB, Lyu M, Liu Y, Feng Y, Wu EX. Robust EPI Nyquist ghost removal by incorporating phase error correction with sensitivity encoding (PEC-SENSE). Magn Reson Med. 2018;79(2):943-951.
6. Chen NK, Wyrwicz AM. Removal of EPI Nyquist ghost artifacts with two-dimensional phase correction. Magn Reson Med. 2004;51(6):1247-1253.
7. Wargo CJ, Moore J, Gore JC. A comparison and evaluation of reduced-FOV methods for multi-slice 7T human imaging. Magn Reson Imaging. 2013;31(8):1349-1359.
8. Veraart J, Sijbers J, Sunaert S, Leemans A, Jeurissen B. Weighted linear least squares estimation of diffusion MRI parameters: Strengths, limitations, and pitfalls. Neuroimage. 2013;81:335-346.
9. Zhang H, Schneider T, Wheeler-Kingshott CA, Alexander DC. NODDI: Practical in vivo neurite orientation dispersion and density imaging of the human brain. Neuroimage. 2012;61(4)