M e d i c a l P hy s i c s a n d I n f o r m a t i c s O r i g i n a l R e s e a r c h

Filippi et al.
MRI of the Lumbar Spine

Medical Physics and Informatics

Original Research
Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved

Improvements in Lumbar
Spine MRI at 3 T Using
Parallel Transmission
Christopher G. Filippi1 OBJECTIVE. Parallel transmission MRI at 3 T improves image quality by reducing di-
Morgan Carlson2 electric effects with radiofrequency shimming. The purpose of this study was to determine
Jason M. Johnson 3 whether parallel transmission MRI improves signal-to-noise ratio (SNR) and contrast-to-
Heather N. Burbank1 noise ratio (CNR) in lumbar spine MRI at 3 T.
Gary F. Alsofrom1 MATERIALS AND METHODS. Ten healthy volunteers underwent T1-weighted MRI
and nine healthy volunteers underwent T2-weighted MRI of the lumbar spine. Sagittal and ax-
Trevor Andrews1,4
ial T1- and T2-weighted images were acquired using parallel transmission MRI and conven-
Filippi CG, Carlson M, Johnson JM, Burbank HN, tional MRI. The percentage improvements in SNR and CNR were calculated, and statistical
Alsofrom GF, Andrews T significance was determined using a two-tailed Student t test with p < 0.05 for significance.
RESULTS. The CNR and SNR showed statistically significant improvements at all levels
of the lumbar spine except SNR at T11 on axial T2-weighted imaging. For sagittal T1-weight-
ed imaging, the average improvement with parallel transmission MRI was 53% in CNR and
19% in SNR. For axial T1-weighted imaging, the average improvement was 48% in CNR and
23% in SNR. For sagittal T2-weighted imaging, the average CNR improvement was 38% and
the average SNR improvement, 20%. For axial T2-weighted scans, the average percentage
improvement in CNR was greater than 100% and the average SNR improvement was 18%
with parallel transmission MRI.
CONCLUSION. The parallel transmission sequence improves image quality of lumbar
spine MRI at 3 T, which is quantitatively supported by statistically significant improvements
Keywords: 3-T MRI, lumbar spine MRI, parallel in SNR and CNR.
transmission MRI

DOI:10.2214/AJR.11.8139
Received October 26, 2011; accepted after revision
January 27, 2012.
Some of the equipment used in this study is supported
by Department of Energy grant SC 0001753.
1
Department of Radiology, Fletcher Allen Health Care
University of Vermont School of Medicine,
111 Colchester Ave, Patrick One, Burlington, VT 05401.
Address correspondence to C. G. Filippi
(christopher.filippi@vtmednet.org).
2
Department of Physical Rehabilitation and Medicine,
University of Washington School of Medicine, Seattle, WA.
3
Department of Radiology, Massachusetts General
Hospital, Boston, MA.
4
Philips Healthcare, Cleveland, OH.
AJR 2012; 199:861867
0361803X/12/1994861
American Roentgen Ray Society
T
echnologic developments have
led to the widespread use of MRI
at 3 T. The benefits of imaging at
a higher magnetic field strength
include improvements in the signal-to-noise
ratio (SNR) and increased spatial resolution
[1, 2]. However, there are drawbacks includ-
ing dielectric shading and a signal intensity
inhomogeneity artifact (i.e., B1 inhomogene-
ity). Nonuniform radiofrequency distribu-
tion in the body is caused by changes of the
radiofrequency wave due to the electrical
properties of the tissue and is more pro-
nounced at 3 T because the radiofrequency
wavelength at 3 T ( 25 cm) approaches the
size of the body, resulting in standing waves
and B1 inhomogeneity [35]. The attenua-
tion of radiofrequency amplitude due to tis-
sue conductivity contributes to B1 inhomo-
geneity [4, 5]. Dielectric shading typically
produces regions of decreased brightness at
3 T and can even produce regions of absent
signal [6, 7].
Recent advances in the use of multiple ra-
diofrequency transmit sources, also known as
parallel transmission MRI or multitransmit
MRI, have made it possible to improve ra-
diofrequency uniformity. This correction can
be accomplished by performing a short addi-
tional calibration to determine the optimal out-
put for the independent radiofrequency trans-
mitters to produce radiofrequency uniformity
throughout the FOV. This calibration process
is called radiofrequency shimming or B1
shimming and is analogous to the B0 shim-
ming calibration that is routinely performed
on all scanners to make the static magnetic
field homogeneous. This improvement in ra-
diofrequency uniformity reduces dielectric
shading in images [8, 9]. With parallel trans-
mission MRI, radiofrequency shimming au-
tomatically adjusts the frequency, amplitude,
phase, and waveform of all radiofrequen-
cy pulses for optimal uniformity adapted to
each patients unique anatomy [8, 9]. Paral-
lel transmission imaging is not the same as

AJR:199, October 2012 861


parallel receive imaging, which uses multiple
receiver coil elements and was first introduced
in the 1980s [10, 11]. Parallel receive technol-
ogy can shorten scanning time [12] and in-
crease SNR [10].
Parallel transmission MRI has recently
Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved
been released for clinical use. The purpose
of this study was to evaluate the benefits of
this technology for improving image quality
in routine lumbar spine MRI at 3 T. In one
previous study describing the use of paral-
lel radiofrequency transmission on imaging
of the spine [9], investigators reported a 30%
reduction in total scanning time and image
quality comparable to that of standard sin-
gle-transmission sequences; however, the
image quality assessment for that study was
subjective and did not include quantitative
measurements of SNR or CNR [9]. In our
study, routine 3-T axial and sagittal T1- and
T2-weighted images were directly compared
quantitatively with 3-T axial and sagittal
T1- and T2-weighted images obtained with
parallel transmission MRI during the same
scanning session by calculating the SNR and
CNR. Image quality of single-transmission
and parallel transmission MR images was
compared qualitatively by two neuroradi-
ologists, both of whom have a certificate of
added qualification, in blinded fashion. We
hypothesized that parallel transmission MRI
improves SNR and CNR and that these im-
provements result in improved image quality.
Materials and Methods
This HIPAA-compliant and institutional re-
view boardapproved study was performed at the
University of Vermont MRI Center for Biomed-
ical Imaging. Ten healthy volunteers (nine men,
one woman; age range, 2547 years; average age,
30.7 years) were recruited for T1-weighted imag-
ing and nine healthy volunteers (six male, three
female; age range, 1646 years; average age, 30.2
years) were recruited for T2-weighted imaging.
T1- and T2-weighted scans of the same patient
were not obtained during the same session be-
cause of the scanning time needed for all of the
required scans.
All images were obtained on a 3-T Achieva MR
system (Philips Healthcare) after a commercial-
ly available dual-source parallel radiofrequency
transmission upgrade to an Achieva TX MR sys-
tem. The upgraded system uses two independent-
ly controlled transmitters to provide radiofrequen-
cy power to the separate ports of a standard body
transmit coil [13]. At the beginning of a session,
a radiofrequency shimming calibration test is per-
formed to estimate the radiofrequency transmis-
862 AJR:199, October 2012
Filippi et al.
sion distribution and calculate the output required
from each transmitter to minimize radiofrequen-
cy transmission inhomogeneity. The console soft-
ware allowed the technologist to acquire each
sequence thereafter either with or without radio-
frequency shimming during the same scanning
session. Each scan that was acquired with radio-
frequency shimming was followed by an identical
scan acquired without radiofrequency shimming.
All scans were obtained using a 15-channel sensi-
tivity-encoding spinal receive coil.
For all acquisitions, a 4-mm slice thickness
was used with a gap of 1 mm and without paral-
lel receive acceleration. The parameters for sag-
ittal T1-weighted imaging were as follows: TR/
TE, 425/8.8; number of signals acquired, 3; FOV,
320 mm; acquired matrix, 360 288; and band-
width, 257 Hz. For axial T1-weighted imaging, the
following parameters were used: TR/TE, 400/8;
number of signals acquired, 3; FOV, 150 mm; ac-
quired matrix, 192 144; and bandwidth, 291 Hz.
The scanning parameters for sagittal T2-weighted
imaging were TR/TE, 3500/110; number of sig-
nals acquired, 3; FOV, 320 mm; acquired matrix,
360 288; and bandwidth, 303 Hz. For axial T2-
weighted images, the scanning parameters were
TR/TE, 2800/120; number of signals acquired, 3;
FOV, 150 mm, acquired matrix, 192 144; and
bandwidth, 150 Hz.
The TR was kept the same for correspond-
ing shimmed and unshimmed scans so that im-
provements in SNR and CNR could be directly
attributable to the radiofrequency shimming. In
a multislice sequence when the number of slices
exceeds the maximum number of slices that can
be acquired in a single TR interval, the slices are
grouped in packages of slices that can fit in a effectively doubles the duration of the total scan-
single TR interval and the total scanning time
is proportional to the number of these packages.
On sagittal sequences, the number of packages
was kept identical. Thus, the scanning duration
for each sagittal T1-weighted scan was 4 minutes
8 seconds; for each sagittal T2-weighted scan, it
was 4 minutes 30 seconds. On axial sequences,
the number of packages was reduced from eight
to five on both the axial T1- and T2-weighted ra-
diofrequency-shimmed images because radiofre-
quency shimming reduced the specific absorption
rate (SAR). More specifically, because the SAR
safety limits of the system require a dead time
at the end of the TR interval (after the acquisi-
tion readouts) during which there are no radiofre-
quency transmissions, the reduced SAR also re-
duces the dead time and allows more slices to be
acquired in each TR interval. Thus, for axial T2-
weighted imaging with radiofrequency shimming,
the scanning duration was 3 minutes 58 seconds.
Without radiofrequency shimming, the scanning
time was 7 minutes 16 seconds because the SAR
limits required more slice packages. For simi-
lar reasons, the scanning duration for axial T1-
weighted imaging was 2 minutes 59 seconds with
radiofrequency shimming and 5 minutes 58 sec-
onds without radiofrequency shimming.
It is necessary to first estimate noise to ob-
tain SNR and CNR estimates. Several meth-
ods have been used in the past [14], but general-
ly these methods have involved either using pixel
values in the tissue of interest for a set of repeat-
ed scans or using a region of interest (ROI) in air
for a single image. For noise estimates in scans
of patients, the repeated-scans approach is often
avoided because the scanning session can be in-
tolerably long; in addition, patient movement be-
tween scans during these long sessions can domi-
nate the noise estimates and lead to a significant
and erratic patient-dependent overestimation bias.
Unfortunately with spinal imaging, it is also often
impractical to use the spinal image to draw a sig-
nal-free ROI for noise estimation because there is
typically little, if any, air within the FOV. In cases
in which a region of air can be found, the region is
often contaminated by artifactual signal due to re-
spiratory motion (as we noted in our preliminary
attempts to obtain these estimates). Therefore, to
obtain an accurate estimate of noise for SNR and
CNR calculations, each acquisition in the current
study was run twice with identical settings (e.g.,
with identical bandwidth and receiver scaling) ex-
cept that the gradients and radiofrequency trans-
mitter were switched off by the MR console dur-
ing the second scan. This method has been used
previously to estimate noise in coronary MR stud-
ies [15]. Although acquiring these noise scans
ning session, this approach made it possible to ob-
tain noise ROIs free of motion artifacts for all cor-
responding spinal images acquired in this study.
SNR and CNR measurements were calculated
from ROI measurements. To compare the SNR of
parallel transmission MRI versus the SNR of non
parallel transmission MRI, T1-weighted scans in
the axial plane of vertebral bodies T11, T12, and L1
were chosen because these areas represent the re-
gion in which there is the greatest degree of dielec-
tric shading. Equivalent ROI measurements were
obtained with respect to size at all levels using a
minimum ROI size of 15 15 mm for both scans
with and scans without radiofrequency shimming.
CNR values were calculated in the axial plane
by placing an ROI on the spinal cord and vertebral
marrow and comparing this ROI with the para-
spinal musculature on equivalent slices at the T11,
T12, and L1 levels on parallel transmission and
nonparallel transmission images. The minimum
ROI size for the spinal cord was 5 5 mm, and the

minimum ROI size for the paraspinal musculature
was 10 10 mm. Average pixel measurements were
obtained along with SDs of the measurements.
To compare SNRs in the sagittal plane on T1-
weighted scans, vertebral bodies L1 through and
including L5 were identified, and ROI averag-
Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved
es were obtained from equivalent slices on both
scans with and those without radiofrequency
shimming. CNR values were calculated for MR
images with and MR images without radiofre-
quency shimming by measuring ROIs of the T11
12 disk through and including the L45 disk space
levels. These values were compared with the ROIs
of the vertebral body marrow (T12 through and
including L5). For example, an average ROI mea-
surement from the T1112 disk was compared
with an ROI measurement from the T12 vertebral
body. The minimum ROI measurement for disk
spaces was 3 2 mm, and the minimum ROI mea-
surement for vertebral bodies was 15 15 mm.
The SNR calculations on sagittal T2-weighted
images were performed the same way as on the
sagittal T1-weighted images. For the CNR cal-
culations on sagittal T2-weighted images, verte-
bral body ROI averages were compared with CSF
and ROI averages in the spinal cord. The mini-
mum ROI size for a vertebral body was 15 15
mm; for spinal cord, 3 3 mm; and for CSF, 5 2
mm. SNR and CNR calculations were performed
on axial T2-weighted images the same way as on
T1-weighted images (Fig. 1). Noise measurements
were calculated from the background noise scans
that were acquired for each and every imaging
scan. The following formula was used to calcu-
late SNR [16]:
Meantissue
(SDnoise) 0.655
where Meantissue is the mean value of the ROI
drawn in the corresponding tissue and SDnoise is the
SD of the ROI on the corresponding noise scan.
In addition, CNR was calculated as follows [16]:
Meantissue1 Meantissue2
(SDnoise) 0.655
where Meantissue1 is the mean value of the ROI
drawn in tissue 1 and Meantissue1 is the mean value
of the ROI drawn in tissue 2.
Percentage improvements in SNR and CNR
values were calculated at all levels measured for
all the volunteers. The statistical significance of
the average percentage improvements in SNR
and CNR comparing images with radiofrequency
shimming to those without radiofrequency shim-
ming was obtained with a two-tailed Student t test
with significance set for p values < 0.05.
Two board-certified neuroradiologists with
certificates of added qualification (4 and 25 years
AJR:199, October 2012 863
MRI of the Lumbar Spine
Fig. 1Sagittal T2-weighted image (TR/TE,
3500/100; 3 signals acquired; 4-mm thickness;
1-mm gap; 320-mm FOV) of healthy 27-year-old
male volunteer shows regions of interest ()
placed on vertebral body marrow and within CSF
for measurement of signal-to-noise ratio. Min =
minimum, max = maximum, US = unsigned short
(integer that cannot be less than zero or greater than
216 1), avg = average, StDv = SD.
experience, respectively) were asked to judge im-
age quality of axial T1-, axial T2-, sagittal T1-,
and sagittal T2-weighted images with and with-
out radiofrequency shimming on the PACS in a
blinded fashion; neither identifiable imaging pa-
rameters nor text was on the images. The grad-
ing system the neuroradiologists used was as
follows: The image quality of the parallel trans-
mission MR image was graded as equivalent to
(2 points), better than (3 points), or worse than
(1 point) the nonparallel transmission MR im-
age. These results were recorded for all 19 cases
for all four sequences. The Cohen kappa statistic
was used to measure the degree of agreement be-
tween the two readers concerning diagnostic im-
aging quality. Kappa values of 0.400.75 are good
agreement and greater than 0.75, excellent agree-
ment. The Friedman rank coefficient test was used
to compare image quality of the parallel transmis-
sion MR images and routine MR images, with sta-
tistical significance set to p < 0.05.
Results
On qualitative examination of the MR im-
ages, one neuroradiologist judged the radio-
frequency-shimmed images to be superior to
the images without radiofrequency shimming
in nine of 10 cases on axial T1-weighted, eight
of nine cases on axial T2-weighted, nine of
10 cases on sagittal T1-weighted, and eight of
nine cases on sagittal T2-weighted imaging.
The other neuroradiologist judged the images
with radiofrequency shimming to be superior
to the corresponding images without radio-
frequency shimming in eight of 10 cases on
sagittal and axial T1-weighted imaging, eight
of nine cases on axial T2-weighted imaging,
and six of nine cases on sagittal T2-weighted
imaging. In all cases, radiofrequency shim-
ming images were judged to be as good as or
better than the images without radiofrequen-
cy shimming.
For all sequences studied, improvements
in image quality met statistical significance
with the Friedman rank coefficient test (p <
0.05). The Cohen kappa statistic exceeded
0.75, which is indicative of excellent agree-
ment between the two neuroradiologists. The
reasons noted for improvements in image
quality included less dielectric shading at the
thoracolumbar junction with better visual-
ization of conus medullaris; improved con-
trast between the spinal cord, vertebral mar-
row, and CSF; and more uniform fat signal
across the entire FOV.
With multitransmit MRI, there were marked
improvements on axial T1-, axial T2-, sagittal
T1-, and sagittal T2-weighted imaging with re-
spect to SNR and CNR values (Tables 1 and
2). On sagittal T1-weighted imaging, the av-
erage percentage improvement in CNR was
53% (p < 0.001 at all levels) and the average
percentage improvement in SNR was 19%
(p < 0.006 at all levels). On axial T1-weight-
ed imaging, the average percentage improve-
ment in CNR was 48% (p < 0.02 at all levels)
and the average percentage improvement in
SNR was 23% (p < 0.008 at all levels). On
sagittal T2-weighted imaging, the average
percentage improvement in SNR was 20%
(p< 0.05 at all levels) and the average per-
centage improvement in CNR was 38% (p<
0.005). On axial T2-weighted imaging, the
average percentage improvement in SNR
was 18%, and p was < 0.005 at T12 and L1,
but statistical significance was not observed
at T11. The average percentage improvement
on axial T2-weighted imaging in CNR was
greater than 100%, with p < 0.05 at all lev-
els (Figs. 25).
Discussion
Nonuniform radiofrequency distribution in
the body often results in dielectric shading,
which creates artifacts that can impede image
interpretation [1, 2, 17]. We have often ob-
served these artifacts on routine lumbar spine
MRI at 3 T, particularly near the lumbosacral
and thoracolumbar junctions on both sagittal
and transaxial imaging. Dielectric shading has
been reported to occur predominantly in heavi-
er patients [18]. However, we have observed
significant dielectric effects across the spectrum
of body types including children (Johnson A, et
 Filippi et al.

TABLE 1: Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR) Values on Sagittal and Axial T1-Weighted
Imaging and Percentage Improvement for Parallel Transmission Sequence Compared With NonParallel
Transmission Sequence
Mean SNR Mean CNR
T1-Weighted
Imaging Plane Multitransmit Nonmultitransit % Improvement Multitransmit Nonmultitransit % Improvement
Sagittal
Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved

L1 50.08 41.76 19.9a 14.26 9.46 50.7a

L2 48.19 41.83 15.2a 15.60 11.02 41.6a
L3 49.00 42.54 15.2a 17.08 11.72 45.7a
L4 47.06 40.02 17.6a 15.89 10.49 51.5a
L5 48.97 39.24 24.8a 18.62 10.71 73.9a
Axial
T11 26.16 20.41 28.2a 26.12 18.97 37.7b
T12 26.55 21.93 21.1a 20.78 13.85 50.0a
L1 24.91 20.90 19.2a 22.76 14.47 57.3a
ap < 0.01.
bp < 0.05.

TABLE 2: Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR) Values on Sagittal and Axial T2-Weighted
Imaging and Percentage Improvement for Parallel Transmission Sequence Compared With NonParallel
Transmission Sequence
Mean SNR Mean CNR
T2-Weighted
Imaging Plane Multitransmit Nonmultitransit % Improvement Multitransmit Nonmultitransit % Improvement
Sagittal
L1 29.36 24.72 18.8a 108.08 79.55 35.9a
L2 29.50 20.69 42.6b 107.94 77.96 38.5a
L3 30.18 27.37 10.3b 107.27 76.91 39.5a
L4 29.63 26.15 13.3b 107.82 78.12 38.0a
L5 30.48 26.11 16.7a 106.97 78.16 36.9a
Axial
T11 15.96 14.03 13.8 11.67 1.38 > 100b
T12 15.17 12.89 17.7a 12.58 1.30 > 100b
L1 15.20 12.47 21.8a 40.94 3.53 > 100b
ap < 0.01.
bp < 0.05.

al., 2010 annual meeting of the International
Society for Magnetic Resonance in Medicine
[ISMRM]), athletic and healthy adults [15],
average-weight and overweight adults [18],
patients with ascites [13, 19], and pregnant
women (Filippi CG, et al., 2010 ISMRM
meeting). We have previously attributed di-
electric effects to problems with the position
of the receiver coil, but this loss of signal uni-
formity seems more likely to be related to B1
inhomogeneity or difficulties with radiofre-
quency transmission. This nonuniformity of
signal from dielectric shading degrades im-
age quality on routine lumbar spine imaging
at 3 T and could potentially obscure impor-
tant pathologic findings.
In our study, there were statistically sig-
nificant improvements in both CNR and SNR
values on axial and sagittal T1-weighted scans
and on axial and sagittal T2-weighted scans
compared with conventional MR images at 3 T.
In our study, we kept both the TR and TE iden-
tical on the nonmultitransmit and multitrans-
mit MR images to assess how radiofrequency
shimming improved both CNR and SNR. On
axial T1- and axial T2-weighted images, the
number of packages was reduced, but all other
imaging parameters were kept constant includ-
ing the TR, TE, number of signals acquired,
FOV, and slice thickness; this reduction in the
number of packages resulted in 50% reductions
in scanning time.
To isolate the direct effect of parallel trans-
mission on SNR, CNR, and subjective image
quality, almost identical protocols were per-
formed using both conventional and parallel
transmission MRI at 3 T. However, because
these protocols were optimized for conven-
tional MRI, they were not necessarily opti-
mal for parallel transmission. Specifically, the
parallel transmission MR scans did not ben-
efit from reduced SAR that would enable de-
creases in TR and scanning time. In a more
practical imaging scenario, parallel transmis-
sion would allow either a substantial decrease
in scanning time or, if the scanning time is
maintained, an additional increase in SNR and
CNR through more signal averaging.

864 AJR:199, October 2012


Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved
A B
A B
In our study, because of the time need-
ed for radiofrequency shimming, obtaining
noise scans, and directly comparing parallel
transmission with nonparallel transmission
MR sequences, which surpassed 2 hours, we
opted to scan volunteers just for the sagittal
and axial T1-weighted scans and not for the
T2-weighted component. If the T2-weight-
ed component had been included, each ex-
amination would have required more than 4
hours of continuous imaging and none of the
volunteers had agreed to undergo imaging
for that long. This characteristic of our study
AJR:199, October 2012 865
MRI of the Lumbar Spine
design could be seen as a potential limitation
to this study. Furthermore, we recruited only
healthy volunteers so we cannot address is-
sues of improved lesion conspicuity or diag-
nostic accuracy for lesion detection because
we were interested just in the determination
of improvements in SNR and CNR with the
addition of parallel transmission MRI. How-
ever, image quality was judged to be supe-
rior by both neuroradiologists. Further stud-
ies will be helpful in determining whether
observed gains in SNR and CNR on rou-
tine lumbar spine MRI with radiofrequen-
Fig. 2Healthy 36-year-old male volunteer.
A and B, Routine axial T1-weighted image (A)
(scanning time, 5 minutes 58 seconds; TR/TE, 400/8;
3 signals acquired; 4-mm thickness; 1-mm gap; 320-
mm FOV) and axial parallel transmission T1-weighted
image (B) (scanning time, 2 minutes 59 seconds;
same parameters as A; number of stacks reduced
from 8 to 5). There is dielectric shading on routine
image (A), particularly across spinous process region
posteriorly and anterior to aorta and inferior vena
cava. These areas show more uniformity of signal on
parallel transmission image (B).
Fig. 3Healthy 38-year-old female volunteer.
A and B, MR images of lumbar spine obtained
using conventional (A) and parallel transmission
(B) sequences with identical parameters (TR/
TE, 425/8.8; 3 signals acquired; 4-mm thickness;
1-mm gap; 150-mm FOV). Only difference between
images is application of B1 shimming for parallel
transmission sequence. Parallel transmission image
(B) better shows less dielectric effect or shading
along superior aspect of image acquisition than non
parallel transmission image (A). Scanning time for
both sequences was 4 minutes 30 seconds.
cy shimming translate into improvements in
thoracic and cervical spine MRI at 3 T.
One of the reasons for obtaining estimates of
SNR and CNR for this study was to provide a
basis for quantitative comparison of scans with
potentially different scaling factors for the pur-
pose of examining the effects of radiofrequency
shimming. The gradient coils were disabled
during the noise scan to avoid any potentially
erratic noise bias due to the gradients them-
selves (e.g., radiofrequency spikes) because
even small contributions can, in principle, affect
the noise estimate without affecting the scaling.

Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved
A B
A B
One could argue, however, that obtaining such
noise scans with the gradients turned on is pref-
erable because having the gradient coils turned
on is more similar to the imaging scan. Such an
approach can provide additional insight into the
noise and the general performance of the MR
scanner; because the contribution of gradients
to the noise can vary from one scan to another,
however, accurately characterizing such effects
might require repeated noise scanning with a
correspondingly longer scanning session. Our
weekly quality control phantom tests showed
no significant artifacts or change in SNR during
the course of this experiment even for more gra-
dient-intensive functional MRI and diffusion-
tensor imaging sequences.
866 AJR:199, October 2012
Filippi et al.
Fig. 4Healthy 46-year-old male volunteer.
A and B, Axial T2-weighted images of lumbar
spine obtained using conventional MRI (A) and
parallel transmission (B) sequences. Routine image
(A; scanning time, 7 minutes 16 seconds) shows
dielectric shading or B1 inhomogeneity at level of
spinal canal that is obscuring visualization of nerve
roots. Signal contrast between nerve roots and CSF
is markedly improved on multitransmit image (B;
scanning time, 3 minutes 58 seconds). Images were
obtained using identical scanning parameters (TR/
TE, 2800/120; 3 signals acquired; 4-mm thickness;
1-mm gap; 150-mm FOV) but there were fewer stacks
for B. Scan time reduction of 50% was observed
for parallel transmission sequence compared with
conventional sequence.
Fig. 5Healthy 33-year-old male volunteer.
A and B, Sagittal T2-weighted images of lumbar
spine obtained using conventional (A) and
parallel transmission (B) sequences with identical
parameters (TR/TE, 3500/110; 3 signals acquired;
4-mm slice thickness; 1-mm gap; 150-mm FOV).
With use of B1 shimming for parallel transmission
sequence, there is less dielectric shading both
anterior to lumbar spine vertebral bodies and at level
of conus medullaris and contrast-to-noise ratio is
markedly improved, resulting in better conspicuity
between CSF and conus medullaris as well as
vertebral body marrow. Hemangioma at T11 level is
clearly better visualized on multitransmit image (B).
In conclusion, parallel transmission MRI fect of B1 inhomogeneity on breast MR imaging
shows statistically significant improvements at 3.0 T. Radiology 2007; 244:929930
in both SNR and CNR on axial and sagit- 3. Zhang Z, Yip C, Grissom W, Noll DC, Boada FE,
tal T1-weighted and on axial and sagittal T2- Stenger VA. Reduction of B1 inhomogeneity with
weighted images of the lumbar spine when transmit SENSE slice-select pulses. Magn Reson
compared with routine MRI at 3 T. The im- Med 2007; 57:842847
age quality of parallel transmission MR im- 4. Yang QX, Wang J, Zhang X, et al. Analysis of
ages was judged to be superior to that of wave behavior in lossy dielectric samples at high
nonparallel transmission MR images be- field. Magn Reson Med 2002; 47:982989
cause there was less dielectric shading and 5. Collins CM, Liu W, Schreiber W, Yang QX, Smith
the signal was more uniform. MB. Central brightening due to constructive inter-
ference with, without, and despite dielectric reso-
References nance. J Magn Reson Imaging 2005; 21:192196
1. Merkle EM, Dale BM. Abdominal MRI at 3.0 T: 6. Jin J, Chen J. On the SAR and field inhomogeneity
the basics revisited. AJR 2006; 186:15241532 of birdcage coils loaded with the human head.
2. Kuhl CK, Kooijman H, Gieseke J, Schild HH. Ef- Magn Reson Med 1997; 38:953963

7. Ibrahim TS, Lee R, Abduljalil AM, Baertlein BA,
Robitaille PM. Dielectric resonances and B(1) field
inhomogeneity in UHFMRI: computational analy-
sis and experimental findings. Magn Reson Imag-
ing 2001; 19:219226
8. Katscher U, Brnert P. Parallel RF transmission
Downloaded from www.ajronline.org by 36.82.102.157 on 03/04/17 from IP address 36.82.102.157. Copyright ARRS. For personal use only; all rights reserved
in MRI. NMR Biomed 2006; 19:393400
9. Nelles M, Knig R, Gieseke J, et al. Dual-source
parallel RF transmission for clinical MR imaging
of the spine at 3.0 T: intraindividual comparison
with conventional single-source transmission. Ra-
diology 2010; 257:743753
10. Hyde JS, Jesmanowicz A, Froncisz W, Kneeland
JB, Grist TM. Parallel image acquisition from
noninteracting local coils. J Magn Reson 1986;
70:512517
11. Roemer PB, Edelstein WA, Hayes CE, Souza SP,
AJR:199, October 2012 867
MRI of the Lumbar Spine
Mueller OM. The NMR phased array. Magn Re-
son Med 1990; 16:192225
12. Hutchinson M, Raff U. Fast MRI data acquisition
using multiple detectors. Magn Reson Med 1988;
6:8791
13. Willinek WA, Gieseke J, Kukuk GM, et al. Dual-
source parallel radiofrequency excitation body
MR imaging compared with standard MR imag-
ing at 3.0 T: initial clinical experience. Radiology
2010; 256:966975
14. Dietrich O, Raya JG, Reeder SB, Reiser MF,
Schoenberg SO. Measurement of signal-to-noise
ratios in MR images: influence of multichannel
coils, parallel imaging, and reconstruction filters.
J Magn Reson Imaging 2007; 26:375385
15. Yu J, Agarwal H, Stuber M, Schar M. Practical
signal-to-noise ratio quantification for sensitivity
encoding: application to coronary MR angiogra-
phy. J Magn Reson Imaging 2011; 33:13301340
16. Henkelman RM. Measurement of signal intensi-
ties in the presence of noise in MR images. Med
Phys 1985; 12:232233
17. Erturk SM, Alberich-Bayarri A, Hermann KA,
Martin-Bonnati L, Ros PR. Use of 3.0-T MR im-
aging for evaluation of the abdomen. Radio-
Graphics 2009; 29:15471563
18. Vernickel P, Roschmann P, Findeklee C, et al.
Eight-channel transmit/receive body MRI coil at
3T. Magn Reson Med 2007; 58:381389
19. Kukuk GM, Gieseke J, Weber S, et al. Focal liver
lesions at 3.0 T: lesion detectability and image
quality with T2-weighted imaging by using con-
ventional and dual-source parallel radiofrequency
transmission. Radiology 2011; 259:421428