Philipp Moser1,2, Bernhard Strasser3, Lukas Hingerl1, Michal Považan4,5, Gilbert Hangel1, Eva Heckova1, Stephan Gruber1, Siegfried Trattnig1,2, and Wolfgang Bogner1
1High Field MR Centre, Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria, 2Christian Doppler Laboratory for Clinical Molecular MR Imaging, Vienna, Austria, 3Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States, 4Russell H. Morgan Department of Radiology and Radiological Science, The Johns Hopkins University School of Medicine, Baltimore, MD, United States, 5F. M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States
Synopsis
The
optimal combination of signals from all receive elements is a
prerequisite in MRSI especially at high field (≥7T), not only for
SNR-efficient acquisition, but also for good parallel imaging
reconstruction [1,2]. Phantom and in vivo experiments showed superior
performance of MOSAIC including higher SNR, smaller FWHM and
anatomically detailed metabolic maps compared to Brown and WSDV coil
combination. MOSAIC is a flexible and robust approach for efficient
MRSI coil combination under challenging conditions (B0≥7T, many
coil elements, no reference coil, low SNR, possible spectral
artifacts, motion/instability related artifacts, 1st-order phase
error), especially with an outlook on parallel-imaging non-Cartesian
MRSI.
Introduction
The
optimal combination of signals from all receive coil elements is a
prerequisite in MRSI especially at high field (≥7T), not only for
SNR-efficient acquisition, but also for good parallel imaging
reconstruction [1,2]. Many coil combination methods have been
proposed: 1) Sensitivity maps (Sensmap) based approaches are the gold
standard, but require a reference coil that is not available at
modern pTx systems [3], 2) others need either a non-suppressed water
scan or weak-water suppression, which prolongs total scan times or
degrades
spectral quality
[4,5], 3) others perform poorly for low SNR or in the presence of
spectral artifacts. The recently published MUSCIAL coil
combination
[6] requires no reference coil, but
is designed for phase-encoded MRSI. With
the advent of spatial-spectral encoding methods for acceleration of
≥7T MRSI, the performance of MUSICAL has to be revisited. The
purpose of our study was to develop a new coil combination method
that capitalizes on all the benefits of MUSICAL, but is more general
with respect to arbitrary spatial-spectral encoded MRSI trajectories.
Our new approach termed Multi-channel
data Of
Spectroscopic
imaging Assembled
via Interleaved
Calibrations
(MOSAIC) acquires reference data in an interleaved fashion and is,
thus, insensitive to motion and scanner instabilities.Methods
MOSAIC
coil combination was integrated into a single-slice FID-MRSI sequence
with non-Cartesian spatial-spectral encoding using equidistant
concentric ring trajectories [7,8]. The calibration data was
directly measured before the WET water suppression module of
the MRSI sequence within
each TR. The calibration module is identical to the MRSI
excitation/acquisition module, but requires only a few trajectory
repetitions and negligibly low flip angles of 5° to avoid
possible saturation effects (Fig.1). The following FID-MRSI
parameters were used: TR: 600ms; acquisition delay: 1.3ms; FA: 45°;
FoV: 220×220mm²; matrix size: 64×64; spectral bandwidth: 2778Hz;
three temporal interleaves; TA: 6:18min.
The
performance of MOSIAC coil combination was evaluated on a 7T Siemens
Magnetom scanner with a 32-channel receive coil array. Three
different reference-coil independent approaches were compared:
MOSAIC, Brown (1st FID point) [5] and WSVD [9].
Phantom
measurements were performed with a Siemens phantom containing 8.2g/l
sodium acetat and 9.6g/l lithium
pyruvate. In vivo data were acquired from five
healthy volunteers.
Coil
combination efficiencies were evaluated qualitatively and
quantitatively based on SNR (adapted pseudo-replica method [10]) and line width (FWHM) maps of NAA
obtained via LCModel. CRLB values of tNAA, tCr, tCho and Glx were compared
among the three methods.
Mean
and standard deviation over all voxels within the brain were derived
per subject and compared using paired t-tests. The number of voxels
where spectral quality was too low to allow reliable spectral fitting
(i.e., CRLB NAA < 20) was assessed in all cases.Results
In
phantom
measurements all
three methods performed similarly well yielding an
average SNR of 18.9±3.7 for MOSAIC, 16.6±3.6 for Brown's method and
17.2±2.9 for WSVD. However,
in
vivo
MOSAIC performed better than Brown combination, followed by WSVD,
which performed worst. While
MOSAIC and Brown yielded comparable amounts of accepted voxels, the mean SNR over all volunteers dropped by 63% and
FWHM increased by 10% for
Brown compared to MOSAIC (p<0.001).
The mean CRLB values were 40% lower when using MOSAIC compared
to Brown combination (Tab.1).
Also, from a qualitative point of view, MOSAIC yields far more
homogeneous and anatomically
detailed
metabolic maps (Fig.2,3). Metabolic maps obtained with WSVD were very
inhomogeneous without any metabolic contrast. More than 15% less
voxels could be reliably fitted, with an overall SNR loss of 68% and
a 2.4fold increase in average metabolic CRLB values compared to
MOSAIC (p<0.001).Discussion/Outlook
In
vivo MOSAIC
performed significantly better than the other two reference-coil
independent methods (i.e.
Brown
and WSVD combination).
The better performance of MOSAIC can be explained by the ability to
obtain its reference data from an external water-unsuppressed source
(=extrinsic), while both other approaches rely on information from
the individual uncombined spectra itself (=intrinsic), which need
water suppression. Both, Brown and WSVD seem to have scaling problems
(see metabolic maps Fig.3), which is not so critical when looking at
ratio maps (Fig.4). Also,
Brown and WSVD combination seem to be more prone to lipid and angular
interleaving artifact, as
well as
1-st order errors than
MOSAIC.
In contrast to MUSICAL, MOSAIC can be directly applied with arbitrary
MRSI trajectories due to its interleaved acquisition. Overall, MOSAIC
is a flexible and robust approach for efficient coil combination of
MRSI data under challenging conditions (i.e., B0≥7T, many coil
elements, no reference coil, low SNR, possible spectral artifacts,
motion/instability related artifacts, 1st-order error),
especially with an outlook on parallel-imaging non-Cartesian MRSI.Acknowledgements
This study was supported by the FFG Bridge Early Stage Grant No. 846505, FWF Grant KLI 646 and FWF Grant P 30701.References
[1]
Kirchner T., et al., 2015, Reduction of voxel bleeding in highly
accelerated parallel 1H MRSI by direct control of the spatial
response function, Magn. Reson. Med. 73, 469–80
[2]
Strasser B., et al., 2017, (2 + 1)D-CAIPIRINHA accelerated MR
spectroscopic imaging of the brain at 7T, Magn. Reson. Med.
78(2):429-440
[3] Pruessmann KP., et al., 1999, SENSE:
sensitivity encoding for fast MRI, Magn. Reson. Med. 42: 952–962
[4]
Graham GD., et al., 2001, Spectroscopic Assessment of Alterations in
Macromolecule and Small-Molecule Metabolites in Human Brain After
Stroke, Stroke; 32:2797–2802
[5]
Brown MA., 2004, Time-domain combination of MR spectroscopy data
acquired using phased-array coils, Magn. Reson. Med. 52: 1207–1213
[6]
Strasser B., et al., 2013, Coil combination of multichannel MRSI data
at 7 T: MUSICAL, NMR Biomed. 26, 1796–805
[7]
Bogner W., et al., 2012, High-resolution mapping of human brain
metabolites by free induction decay 1H MRSI at 7T, NMR Biomed. 25,
873–82
[8]
Hingerl
L., et al., 2017, Density-Weighted
Concentric Circle Trajectories for High Resolution Brain Magnetic
Resonance Spectroscopic Imaging at 7T, Magn. Reson. Med.
[9]
Rodgers CT., et al., 2016, Coil combination for receive
array spectroscopy: Are data-driven methods superior to methods using
computed field maps?, Magn. Reson. Med. 75(2):473-87
[10]
Robson PM., et al., 2008, Comprehensive Quantification of
Signal-to-Noise Ratio and g-Factor for Image-Based and k-Space-Based
Parallel Imaging Reconstructions, Magn. Reson. Med. 60(4):895-907