Candace C Fleischer1, Xiaodong Zhong1,2, and Hui Mao1
1Department of Radiology and Imaging Sciences, Emory University, Atlanta, GA, United States, 2MR R&D Collaborations, Siemens Healthcare, Atlanta, GA, United States
Synopsis
Parallel imaging and spectroscopy are facilitated by
multi-channel phased array coils. An important step is the combination of individual
data from each channel, yet the effect of the non-uniformity of array coils on
signal-to-noise ratio (SNR) is poorly characterized. Here, we present a
systematic framework for identifying vulnerabilities in phased array coils for
MRS. We demonstrate the importance of voxel position and coil proximity on
overall SNR in a phantom and human subject, with significant SNR improvements after
selectively filtering individual spectra based on pre-determined SNR thresholds
which must be optimized for each phased array coil and volume of interest.
Introduction
Parallel imaging with phased array coils is widely implemented
in MRI to improve efficiency and accelerate acquisition.1-2 Phased
array coils require the combination of data collected from individual coil channels
to maximize the signal-to-noise ratio (SNR).3-4 However, SNR
variations in MRS due to the non-uniformity of each coil, particularly the
proximity of individual coil channels relative to the volume of interest are
not well understood. Here, we systematically investigate the effect of the
number and placement of individual coil channels and voxel position on spectral
SNR in a brain spectroscopy phantom and human subject.Methods
Single voxel MRS was acquired in a brain spectroscopy
phantom (GE #2152220) using a 32-channel head-array coil on a Siemens 3 T Tim Trio
and a 64-channel head and neck-array coil on a Siemens 3 T Prisma (Siemens
Healthcare, Erlangen, Germany) with the following parameters: TR/TE/averages:
2s/30ms/64; 1200 Hz bandwidth; 2048 complex data points; 2x2x2cm3 voxel
size. With IRB approved informed consent, brain spectra were acquired in a
healthy human subject (female, 20 years old) with the 64-channel coil (Prisma) using
the same parameters except with 128 averages. For phantom experiments, the
voxel position was systematically shifted by 30 mm in the left, right,
anterior, posterior, head, and feet directions and 50 mm in anterior and
posterior directions. For data acquisition in the human brain, voxel position
was shifted by 30 mm in the left and right directions (Figure 1).
SNR and peak full-width at half
maximum (FWHM) were used as spectral quality measures. Data was processed with
LCModel (version 6.3-1H) and Matlab (Mathworks, Natick, MA, R2015a). Comparison
of 32- and 64-channel array coils was performed using the vendor-supplied automated
combination algorithm3 and a paired two-tailed t-test. For the 64-channel
coil, data was acquired from either the 34 channels automatically selected by
the software based on the position of the phantom or from all 64 channels
manually selected. Spectra from each coil channel were processed individually and
selectively combined. Specifically, individual spectra above a SNR threshold
were combined with equal weighting and those below the threshold were excluded
from the final combined spectrum. The % change in SNR of the combined spectrum relative
to the maximum SNR for that position was calculated as: (SNR – SNRmax)/SNRmax
*100%. Comparisons of the combined spectral SNR as a function of position and
SNR threshold were performed using ANOVA and a post-hoc Tukey’s two-tailed
t-test.Results and Discussion
Table 1 summarizes the results of data collected from
multiple voxel positions using the 32- and 64-channel array coils along with data
acquired in the human subject with the 64-channel coil, with spectra from
individual channels combined using the vendor-supplied algorithm.3 No
significant difference was observed for peak FWHM (32-channel: 0.0127 ± 0.0043
ppm; 64-channel: 0.0130 ± 0.0036 ppm; p>.05). While both phased array coils produced
high spectral quality, the 32-channel coil produced significantly higher
overall mean SNR and lower standard deviation (47 ± 7) than the 64-channel coil
(42 ± 10; p<.05), demonstrating that simply increasing the number of phased
array channels does not necessarily lead to higher SNR. As the coil channels
are not uniformly arranged spatially (i.e. 32-channel coil contains 20
posterior and 12 anterior channels), and the location of the volume of interest
varies in practice, combining data from all coil channels may not be optimal particularly
when including channels with poor SNR (Figure 2A). We observed that overall SNR
was a function of voxel position (F=49.9, p=.002) at a given SNR threshold (≥5,
10, or 15). The change in overall SNR was dependent on the SNR threshold (F=6.0,
p=.03, Table 2). Tukey’s post-hoc t-tests revealed that relative to the
threshold of SNR≥5, both SNR≥10 (t=4.1, p<.05) and SNR≥15 (t=4.4, p=.04)
thresholds produced combined spectra with significantly higher SNR. The
difference between thresholds of SNR≥10 and SNR≥15 was not significant (t=0.2,
p>.05). Representative combined spectra acquired in a phantom and a human
subject at varying SNR thresholds and voxel positions are shown in Figures 2
and 3, respectively.Conclusions
Our results demonstrate the importance of considering coil
non-uniformity, position of individual coil channels, and data combination
strategies in order to improve spectral quality and increase SNR with array
coils. Overall SNR of spectra collected from phased array coils can be increased
by combining spectra with high SNR and discarding those with poor SNR at a
pre-determined threshold which must be optimized for each array coil and volume
of interest. Future studies will focus on developing improved automated algorithms
for phased array coil combination.Acknowledgements
This work was supported by NIH 1R01CA169937.References
1. Roemer,
PB et al. The NMR phased array. Magn Reson Med. 1990;16:192-225.
2. Deshmane,
A. et al. Parallel MR imaging. J Magn Reson Imaging. 2012;36:55-72.
3. Brown,
MA Time-domain combination of MR spectroscopy data acquired using phased-array
coils. Magn Reson Med. 2004;52:1207-1213.
4. Abdoli,
A and Maudsley, AA. Phased-array combination for MR spectroscopic imaging using
a water reference. Magn Reson Med. 2016;76:733-741.