1066

SENSE-based reconstruction for removal of spurious echo artifacts in MRS

Adam Berrington^1,2, Michal Povazan^1,2, and Peter B Barker^1,2

¹Russell H. Morgan Department of Radiology and Radiological Science, The Johns Hopkins University School of Medicine, Baltimore, MD, United States, ²F.M. Kirby Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States

Synopsis

Spurious echo artifacts appear in MRS spectra and originate from regions of unsuppressed and insufficiently crushed signal. We propose a reconstruction method to remove such artifacts using the sensitivity weighting of the receive channels at each spatial location. Data from phantom and simulations show that spurious echoes can be separated from underlying signal. The spatial distribution of unwanted signal was estimated in vivo using B₀ maps. Artifactual components of the signal were identified in sinus regions, however not entirely removed from the spectrum. Further work will aim to improve the conditioning of the SENSE reconstruction to remove spurious echo artifacts from MRS data.

Aim

To develop a reconstruction method based on sensitivity weighting of multiple receive channels to separate artifactual signal from localized signal in MRS.

Introduction

Ensuring high quality spectra is critical for MRS to be implemented routinely in the clinic.¹ However, quality is often degraded by the presence of spurious echo (or ghosting) artifacts, which overlap with metabolite peaks and affect quantification.² These artifacts arise from the excitation and refocusing of unsuppressed water signal from outside the MRS voxel, which is not properly crushed prior to acquisition.³ For instance, large susceptibility gradients around the sinuses have been shown to be a source for such artifacts.⁴ Methods such as alternating slice selection,⁴ optimizing crusher gradients,⁵ and deep learning,⁶ have been proposed to avoid or remove these artifacts. Given that such artifacts originate from spatially distinct regions, we propose a reconstruction method based on sensitivity encoding⁷ using multiple receiver coils to separate these signals from the MRS signal. A similar approach was recently demonstrated for dual-excited MRS voxels.⁸

Methods

An outline of the proposed method is shown in Fig.1 using a simulated phantom⁹ in combination with 16 receiver coils ($$$K$$$). Mean coil sensitivities, $$$S_{A_{j},k}$$$ at artifact regions, $$$A_{j}$$$ ($$$j=1\ldots N$$$) and voxel location, $$$V$$$, are calculated for each coil, $$$k$$$ ($$$k=1\ldots K$$$) generating a sensitivity matrix of size $$$[N+1, K]$$$. As in SENSE MRI, the unfolding matrix, $$$U$$$ is calculated from $$$S$$$ using the noise covariance of the channels, $$$\Psi$$$. Unfolding reveals spectra from the $$$N+1$$$ regions.

Scanning was performed on a 7 T Philips Achieva MR system with dual-transmit/32-channel receive array head coil. Sensitivity maps were median filtered and normalized to a sum-of-squares image. Noise covariance was calculated from background voxels in the sensitivity images. Phantom data were acquired from a GE ‘Braino’ Phantom (STEAM, TE/TR = 14ms/3s; NT = 32, 15x15x15mm³ voxel). Phantom data were reconstructed using a 3x1 grid of ROIs with no prior estimate of artifact regions. Spectra were also acquired from one volunteer (semi-LASER, TE/TR = 30ms/5s, NT = 64, 20x20x20mm³) in a frontal voxel shimmed to second order. In order to maximize the artifact, spoiler gradient amplitudes were reduced on all axes. To identify regions of spurious echo in vivo, a B₀ map was acquired with identical shim settings to the MRS scan. The B₀map was phase unwrapped¹⁰, smoothed and thresholded according to the bandwidth (150 Hz) of VAPOR water suppression pulses in MRS (Fig. 2). The subsequent image was segmented leaving regions of unsuppressed water in the brain. SENSE unfolded data from MRS voxels were compared to weighted and non-weighted channel-combined data.

Results

Fig. 1 shows a simulated spectrum containing 3 singlet resonances (V) and 1 spurious echo (A1) when using conventional channel combination. SENSE reconstruction was subsequently able to separate these two signals, matching the ground truth simulation.

In the phantom, SENSE reconstructed signals (A1 and A2) from outside the voxel contained pure spurious echo data occurring at two different time-points in the FID (Fig. 3B) and revealed a somewhat cleaner region around the water peak for the unfolded MRS voxel than normal channel combination (Fig. 3C), although there was not complete removal of the artifact.

Reconstructed spectra from one volunteer are shown in Fig. 4. Thresholding of the B0 image (Fig. 2) revealed 3 regions where residual water may be expected. One region, A2, was positioned close to the voxel (V) in the central sinus (g-factor = 1.4) and contained a large spurious echo component. Reconstruction did not completely remove the artifact. However, when compared to normal channel-combined data (Fig. 4C), the method was observed to produce cleaner spectra over the spectral range (3-4) ppm than unweighted coil-combination and similar to weighted combination.

Discussion

A straightforward reconstruction method is proposed to separate MRS signal from spurious echo artifacts. Simulation and phantom results show there is potential to identify artifact and MRS signal components based on sensitivity encoding. The degree of artifact separation depends on precise knowledge of the artifact’s location and, similar to parallel imaging, accurate sensitivity maps. Results in vivo support the previous finding that spurious echoes often originate from the frontal sinus (Fig. 4).⁴ The artifact was not entirely removed from the spectra. Gradient-echo sensitivity maps are vulnerable to signal dropout in such regions, particularly at 7 T, hence may result in poor estimation of sensitivities. Future work will endeavor to improve the quality of the sensitivity maps through interpolation and filtering⁷ and include regularization terms¹¹. In addition, the acquisition of residual water suppression images (rWSI) (Fig. 5) may aid truer identification of artifact regions.

Acknowledgements

We would like to acknowledge the following funding support: NIH P41EB015909

References

1. Öz, G. et al. Clinical Proton MR Spectroscopy in Central Nervous System Disorders. Radiology 270, 658–679 (2014).
2. Kreis, R. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR Biomed. 17, 361–81 (2004).
3. Carlsson, A., Ljungberg, M., Starck, G. & Forssell-Aronsson, E. Degraded water suppression in small volume 1H MRS due to localised shimming. MAGMA 24, 97–107 (2011).
4. Ernst, T. & Chang, L. Elimination of artifacts in short echo time1H MR spectroscopy of the frontal lobe. Magn. Reson. Med. 36, 462–468 (1996).
5. Landheer, K. & Juchem, C. Dephasing optimization through coherence order pathway selection (DOTCOPS) for improved crusher schemes in MR spectroscopy. Magn. Reson. Med. 1–14 (2018).
6. Kyathanahally, S. P., Döring, A. & Kreis, R. Deep learning approaches for detection and removal of ghosting artifacts in MR spectroscopy. Magn. Reson. Med. 80, 851–863 (2018).
7. Pruessmann, K. P., Weiger, M., Scheidegger, M. B. & Boesiger, P. SENSE: sensitivity encoding for fast MRI. Magn. Reson. Med. 42, 952–62 (1999).
8. Boer, V. O., Klomp, D. W. J., Laterra, J. & Barker, P. B. Parallel reconstruction in accelerated multivoxel MR spectroscopy. Magn. Reson. Med. 74, 599–606 (2015).
9. Guerquin-Kern, M., Lejeune, L., Pruessmann, K. P. & Unser, M. Realistic Analytical Phantoms for Parallel Magnetic Resonance Imaging. IEEE Trans. Med. Imaging 31, 626–636 (2012).
10. Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W. & Smith, S. M. FSL. Neuroimage 62, 782–790 (2012).
11. Lin, F. H., Kwong, K. K., Belliveau, J. W. & Wald, L. L. Parallel Imaging Reconstruction Using Automatic Regularization. Magn. Reson. Med. 51, 559–567 (2004).

Figures

Outline of SENSE artifact removal for spurious echo signals in MRS spectra. Simulations were performed with known spectra and spurious echo signal with 16 channel sensitivity maps generated. SENSE reconstruction is performed on the sensitivities, for simulations $$$\Psi$$$ = identity matrix. Step 5 shows the final reconstructed of the unfolded signal in panel 1 matching to the underlying simulated dataset.

A method for identify the location of spurious echo artifacts. A B₀ map (left) is acquired with same shim settings as MRS. This map is thresholded according to the water bandwidth of the MRS: $$$\left|\Delta f_{0}\right|>$$$ 150 Hz, revealing locations outside the water suppression bandwidth. This image is then labelled for SENSE reconstruction.

SENSE unfolding of signals from a 3x1 grid in a phantom. A) Reconstructed voxels overlaid onto a volume-coil image (A = artifact region, V = MRS voxel). B) Spectra contained large spurious echo artifacts. SENSE unfolding of two artifact regions (A₁ and A₂) revealed signals containing pure artifact data. A large spurious echo can be seen around halfway during the FID of A₁. In addition, signal from A₂ appeared to capture another echo which manifests downfield of the water peak. C) Comparing voxel spectra (V) to the channel-combined spectrum revealed a small improvement in the spectrum near the water peak.

In vivo demonstration of SENSE unfolding in one volunteer. A) The B₀ map, including voxel overlay in the frontal brain, and three labelled artifact regions (A₁, A₂ and A₃). B) The unfolded spectra: A₁ is a large region covering the inferior brain and resulted in small amplitude signal. The biggest artifactual signal component was from A₂ in the sinus region and contained a large spurious echo visible in the FID and to the left of the water peak. C) A close-up comparison of V to the weighted (and unweighted) coil combined spectra. The reconstruction performed similarly to weighted coil combination.

A potential method for determining artifact regions. Residual water suppression images (rWSI) were generated in one volunteer by including the VAPOR water suppression module before a T2-weighted turbo spin echo imaging sequence similar to previous literature³. When divided by an identical image acquired without water suppression, regions of unsuppressed water are visible. The rWSI is comparable to the B₀ map (right). An identical linear Z-shim is applied during both imaging sequences.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)

1066