Multiple-frequency excitation wideband MRI (ME-WMRI)

Multiple-frequency excitation wideband MRI (ME-WMRI)
Edzer L. Wu, Tzi-Dar Chiueh, and Jyh-Horng Chen
Citation: Medical Physics 41, 092304 (2014); doi: 10.1118/1.4893502
View online: http://dx.doi.org/10.1118/1.4893502
View Table of Contents: http://scitation.aip.org/content/aapm/journal/medphys/41/9?ver=pdfcov
Published by the American Association of Physicists in Medicine
Articles you may be interested in
Measuring signal-to-noise ratio in partially parallel imaging MRI
Med. Phys. 38, 5049 (2011); 10.1118/1.3618730
Spin-history artifact during functional MRI: Potential for adaptive correction
Med. Phys. 38, 4634 (2011); 10.1118/1.3583814
An anthropomorphic phantom for quantitative evaluation of breast MRI
Med. Phys. 38, 743 (2011); 10.1118/1.3533899
Partial volume estimation and segmentation of brain tissue based on diffusion tensor MRI
Med. Phys. 37, 1482 (2010); 10.1118/1.3355886
MRI/TRUS data fusion for prostate brachytherapy. Preliminary results
Med. Phys. 31, 1568 (2004); 10.1118/1.1739003
Multiple-frequency excitation wideband MRI (ME-WMRI)
Edzer L. Wu
Institute of Biomedical Engineering, National Taiwan University, Taipei 106, Taiwan
and MRI/MRS Interdisciplinary Lab, National Taiwan University, Taipei 106, Taiwan
Tzi-Dar Chiueha)
Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan
Jyh-Horng Chena)
Institute of Biomedical Engineering, National Taiwan University, Taipei 106, Taiwan; Department of Electrical
Engineering, National Taiwan University, Taipei, Taiwan; and MRI/MRS Interdisciplinary Lab, National
Taiwan University, Taipei 106, Taiwan
(Received 7 March 2014; revised 13 July 2014; accepted for publication 6 August 2014;
published 28 August 2014)
Purpose: In this study multiple-frequency excitation wideband MRI (ME-WMRI), a technique designed to acquire simultaneously excited images without blurring was proposed.
Methods: ME-WMRI consists of (a) a coherent acquisition sequence that applies refocusing gradient
during readout to mitigate signal dephasing, and (b) a signal enhancement procedure that further
reduces image blurring. This two-step method was tested on phantom imaging and in vivo imaging
with up to 3.84 pixel blur.
Results: Imaging results in both phantom and in vivo show that the lost details were successfully restored in the blur-mitigated images, and the accelerated images were comparable to standard images.
Conclusions: The proposed ME-WMRI method can effectively undo the image blurring, and thus
removes the limitation on fast simultaneous multislice MRI systems with a large acceleration factor.
© 2014 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4893502]
Key words: ME-WMRI, dephasing, blur, coherent acquisition, signal enhancement
1. INTRODUCTION
Magnetic resonance imaging has become a powerful tool
and the fastest growing medical imaging technique since it
was first invented four decades ago.1 The rich information
and contrast it provides has a profound impact on physicians, biochemists, neurologists, oncologists, psychologists,
and many others. Over the past few decades, MR imaging
speed has been greatly improved with various acceleration
techniques such as parallel imaging and non-Cartesian sampling strategies.2–4 Nevertheless, with the increasing need for
large volume coverage scans, we believe the overall MRI
throughput could be further improved with simultaneous multislice methods comparable to multicut method in CT.5–7
The use of simultaneous multislice (SMS) excitation to accelerate MR scans was first brought up more than two decades
ago.8 In this work, multiple-frequency signals spread over
a bandwidth wider than conventional MRI (wideband MRI)
was used in the excitation to increase acquisition efficiency.
Such technique can excite different locations of the subject simultaneously, as shown in Fig. 1. Since then, SMS technique
has gathered attention and several new methods have been
proposed. These methods are categorized into two classes,
shown in Fig. 2. The first class uses additional hardware. Multicoil array technique was proposed by Larkman et al., where
spatial encoding information in the multicoil receiver system
provides information to separate the simultaneously excited
slices.9 Another method reported by Breuer et al. implements
phase-modulated RF pulses and controlled aliasing to perform
092304-1
Med. Phys. 41 (9), September 2014
multislice imaging with phased array coils.10 Applications
such as functional and diffusion imaging have greatly benefitted from these techniques, making the imaging of brain faster
than ever.11–13 Other than adding RF coils, Lee et al. presented a method that superimposes a micro-B0 array onto the
main field to assign unique resonance frequencies to different
slices (MAMBA), thus separating the simultaneously excited
images.14–16 Setsompop et al. combined CAIPIRINHA technique with multislice echo planar imaging (EPI), and skillfully utilized blipped separation/refocusing gradients to prevent signal dephasing.17, 18
On the other hand, there are approaches that achieve multislice imaging without extra hardware. In 1988 Weaver proposed a method that applies a gradient during spatial encoding to modulate the carrier frequency for each excited slice
such that they become separated in the reconstructed image.8
Paley et al. combined SENSE imaging technique to further
increase the scanning speed of Weaver’s method.19 However,
the separation gradient causes dephasing along the slice direction and leads to voxel tilting/in-plane image blurring. Wu
et al., the authors of this study, have extended this method to
3D imaging sequences and achieved good results in isotropic
brain imaging.20
In addition to the above two classes of SMS methods,
Souza et al. reported the use of Hadamard-encoded RF pulses
as a means to separate the simultaneously excited/acquired
slices during image reconstruction. Similarly, phase-offset
multiplanar (POMP) technique proposed by Glover also applies an encoded RF pulse for excitation.21, 22 However, these
0094-2405/2014/41(9)/092304/10/$30.00
© 2014 Am. Assoc. Phys. Med.
092304-1
092304-2
Wu, Chiueh, and Chen: ME-WMRI
092304-2
F IG . 1. Illustration of a ME-WMRI: whole body scan. The five images were excited and acquired simultaneously. Imaging efficiency is improved by the
increased excitation/acquisition bandwidth.
two methods do not shorten the scan time of MRI due to the
extra number of excitations or phase encoding steps required
in their implementation.
Although much improvement has been obtained in the previous works, the solution for multislice acquisition of highquality MR images in commonly used gradient echo or spin
echo imaging without the use of added hardware still remains
elusive. In this study we start from the work of Weaver and
investigate the signal processing aspect of SMS MRI. The underlying mechanism for image blurring is formulated and two
new techniques that alleviate the image blurring of both 2D
and 3D wideband MR imaging are proposed. Experimental
results show that the proposed solution can increase the imaging speed by several folds without compromising the acquired
image quality.
2. METHODS
2.A. The effects of separation gradient
According to MR physics, separation gradients applied
during readout cause signal dephasing thus image blurring.
Without loss of generality, we consider only the signal corresponding to a single excited slice in a SMS MRI system.
Although derived with a single-slice scenario, the results can
be extended to SMS excitation cases under the assumption
of ideal field homogeneity, gradient linearity, and RF coil
performances.
The 2D k-space (spatial frequency space kx and ky ) MRI
signal S(kx , ky ) derived from Eq. (1), where z1 and z2 represent the two ends of the excited slice, and γ is the gyromagnetic ratio.23, 24 We also assume ideal RF excitation and
uniform proton density distribution across the excited slice
ρ(x, y):
S(kx , ky )
⎡
⎤
z2 ∞ ∞
ρ(x, y) exp{i2π (kx x + ky y)}dxdy ⎦dz
= ⎣
z1
−∞ −∞
∞ ∞
= (z1 − z2 ) ×
ρ(x, y) exp{iγ (Gx xτ
−∞ −∞
+ Gy yTpe )}dxdy.
F IG . 2. Technical progress of multislice MR imaging. Technical progress
of multislice MR imaging, showing the main methods that use simultaneous
multislice to accelerate MR imaging. These methods are grouped into two
classes: the first class uses additional RF coils or gradients while the second
class can operate in any general system configuration.
Medical Physics, Vol. 41, No. 9, September 2014
(1)
The proton density distribution function is subjected to a 2Dspatial encoding process [spatial encoding gradients and duration (Gx, τ ) and (Gy , Tpe )] and then it is integrated along X and
Y directions. The resulting MR image can be obtained through
two-dimensional inverse Fourier transform of the sampled kspace signal S(kx , ky ).
In SMS methods after slice excitation, an additional separation gradient was applied during spatial encoding and
this separation gradient (Gz ) has the same duration τ as the
092304-3
Wu, Chiueh, and Chen: ME-WMRI
092304-3
F IG . 3. Pulse sequence of wideband MRI and the effects in signal magnitude and phase caused by separation gradient. (a) Pulse sequence of 2D wideband
MRI with a separation gradient Gz is added during frequency encoding. The duration of the separation gradient is the same as frequency encoding gradient
and the gradient strength is given that the excited images can be fully separated after reconstruction. (b) The effects on signal magnitude and phase with the
application of separation gradient. Due to the phase difference of the spins across the excited slice (green and red arrows at the bottom), the collective magnitude
will undergo an attenuation with a sinc-shaped profile along sampling time. There is also a linear phase change that will result in image shift. (c) The normalized
signal magnitude and phase in the presence of a separation gradient measured by experiment (1–1.8 pixel blur).
readout interval. The 2D k-space MRI signal is then given by
S (kx , ky )
z2 ∞ ∞
=
ρ(x, y) exp{i2π (kx x
z1
−∞ −∞
+ ky y + kz z)}dxdy dz
∞ ∞
=
ρ(x, y) exp{iγ (Gx xτ
−∞ −∞
⎛ z
⎞
2
+ Gy yTpe )}dxdy × ⎝ exp(iγ Gz zτ )dz⎠ .
(2)
z1
Equation (2) indicates that the extra separation gradient Gz
introduces a term that is an integration over the slice thickness
along the Z direction, namely, from z1 to z2 , and
⎛ z
⎞
2
⎝ exp(iγ Gz zτ )dz⎠ = d × sinc(γ Gz τ (d/2))
z1
× exp[iγ Gz τ Zcen ],
(3)
where d = z2 − z1 and Zcen = (z1 + z2 )/2. The MRI signal can
be reformulated as
S (kx , ky ) = S(kx , ky ) × sinc(αkx ) × exp(iβkx ),
(4)
where α = π Rd, β = 2π RZcen , and R = Gz /Gx .
The sinc term in Eq. (4) causes signal attenuation in higher
spatial frequencies (kx ), which blurs the image along the X dimension. The degree of attenuation is determined by gradient
ratio R, duration of the separation gradient τ , and the excited
Medical Physics, Vol. 41, No. 9, September 2014
slice thickness d. Degree of blur is defined in terms of pixels,
in correspondence to voxel tilting in prior studies.8 From the
attenuation profile, the number of pixel blur is also equivalent
to the ratio of acquisition length to the width of the first zero
crossing of the sinc function.
The exponential term in Eq. (4) represents a shift along the
X direction in the acquired image, and the amount of shift
xs turns out to be the same amount as RZcen , where R is the
slope of the tilted voxels defined in previous SMA studies8
and Zcen is the distance between the slice and the center of the
separation gradient. The sequences and the effects of the separation gradient during readout derived in Eq. (4) are shown in
Fig. 3. Shown in the far right of Fig. 3 is the acquired attenuation profiles from 1 to 1.8 pixel blur to verify the equations.
At 1 pixel blur the span of the first zero crossing is about the
total acquisition length. Note how the span decreases as the
blurring becomes more serious.
For the sake of convenience, we define the number of simultaneous excited slices in ME-WMRI as the “wideband
factor” and use W to represent this factor in the following discussion. During acquisition, the FOV along readout as to be
expanded W folds to contain the separated slices, number of
sampling along readout NFE was also increased by W times to
maintain the same spatial resolution.
2.B. Coherent acquisition method
After identifying the source of image blurring, we propose a method that diminishes signal dephasing in the acquired k-space data. This method breaks one readout duration into several segments separated by refocusing gradients.
As a result, the signal during each segment will experience
less signal attenuation and the final attenuation profile will
become cycloid shaped with a series of sinc lobes. We call
092304-4
Wu, Chiueh, and Chen: ME-WMRI
092304-4
F IG . 4. Blur mitigation methods: (left) S = 2 coherent acquisition sequence and (right) steps to derive Wiener filter map used in signal enhancement. (a) (A)
The normalized signal magnitude and phase of the proposed S = 2 coherent acquisition method. (B) The attenuation profile of an S = 2 sequence compared to S
= 1 sequence (dotted line). It is clear that by refocusing the signal the second segment in S = 2 sequence prevents further dephasing, preserving signal intensity
in the second segment. (b) Flow chart of deriving the signal enhancement function. (1) and (2) For every horizontal line of the joined k-space, SNR is measured
within each sampling window and (3) and (4) the measured SNR is incorporated to produce the final Wiener filter map.
this method coherent acquisition method, and the number
of segments per readout duration is defined as “S.” Therefore, the conventional SMS MRI corresponds to the S = 1
case.
In Fig. 4(a), we show an S = 2 sequence and the evolution
of the effect on the signal over time, magnitude, and phase
drawn separately. The sequence depicts the actual gradient
waveform with finite slew rate, buffer points were added at
the guard-time intervals (shaded areas) right before and after refocusing to ensure equally spaced sampling between the
segments, also no readout gradient was applied during the
refocusing period. The magnitude and duration of the refocusing gradient is designed to minimize signal dephasing in
each segment. In the S = 2 sequence since MR signal is sampled during the refocusing period, redundant information such
as k-data acquired during refocusing, unequally spaced samplings, and overlapping kx will be removed and the remaining
k-space data segments joined together before further signal
processing.
2.C. Signal enhancement method
Even after coherent acquisition sequence, there still remains some residual k-space signal attenuation located in the
valleys of the cycloid shaped profile. The degree of blurring
was decreased for certain structures, and the Fourier transform of this attenuation profile introduced smaller spikes that
would cause additional image artifacts. Toward this end, we
propose to further apply k-space signal enhancement to restore the original signal intensity.25
Medical Physics, Vol. 41, No. 9, September 2014
Considering the image property and attenuation level we
choose Wiener filtering, which is given by
1
|H (u, v)|2
K̂(u, v) =
K(u, v), (5)
H (u, v) |H (u, v)|2 + P (u, v)
where the acquired k-space data are K(u, v) and the attenuation curve H (u, v). The attenuation curve H (u, v) can either be measured or simulated. The advantage of Wiener filter
over a direct inverse filtering is that by adding a weighting
term P (u, v) (P = 1/SNR), the Wiener filtering is able to adjust the level of signal enhancement in images of low SNR,
avoiding unwanted noise amplification. A P-weighting map
is calculated over the acquired k-space to determine the level
of enhancement at different spatial frequencies. The process
is described in Fig. 4(b).
2.D. Experiment setup
The data obtained for this study are performed on a Bruker
3T Medspec system using a T8826 transmit/receive head coil
as well as a 7 T Biospec animal system using a volume coil.
Images of phantom made by Bruker for quality testing are
acquired to evaluate the effectiveness of the blur mitigation
method. The S = 2 images have a FOV of 19 × 19 cm,
flip angle (FA) = 30◦ , TR/TE = 100/8 ms, slice thickness
= 1.5 mm, resolution = 0.781 × 0.781 mm, echo position
was set at 30%, falling on the center of the first readout segment and away from the gradient switching. Separation gradient strength is set so that the image blur ranges from 0 to
2 pixels, in steps of 0.2 pixel. Gradient strength = 70% of
092304-5
Wu, Chiueh, and Chen: ME-WMRI
092304-5
F IG . 5. Effects of the two-stage blur mitigation methods and SNR analysis. Contrasts of low-resolution detail (L), high resolution detail (H), and signal-to-noise
ratio are plotted with respect to different degrees of blurring. After applying blur mitigation methods, the detail contrast of both frequencies remains on par with
standard images up to 1.8 pixel blur, but the low resolution detail contrasts start to degrade over 1.8 pixels. SNR of all degrees of blur remains the same due to
the adaptive Wiener filtering that prevents noise from being enhanced.
3.02 G/cm, ramp time = 450 μs, slew rate = 4.69 mT/m ms.
Attenuation profiles corresponding to various degrees of blur
are also taken to verify the equations and to perform signal enhancement. The profiles are normalized by the free induction
decay baseline to derive the actual attenuation. To quantify
the effectiveness of the blur mitigation methods, the image
profiles over vertical stripes of high- and low-spatial frequencies at the lower part of the phantom are taken. Detail contrast is defined as the average peak-to-valley over mean signal
strength, and will decrease as the image becomes blurrier. As
for the W = 2, S = 3 experiment a custom-made phantom
is built to show blur mitigation effects in fine structures. Two
slices 3 cm apart are excited by a 51 kHz modulated sinc pulse
and the power is adjusted to maintain the equivalent flip angle
(FA = 30◦ ). The W = 2 images have a FOV = 12 × 6 cm,
TR/TE = 100/9 (ms), a 1.28 ms increase in TE was caused by
the added guard points. Gradient strength = 15% of 3 G/cm,
ramp time = 450 μs/slew rate = 1 mT/m ms. Echo position
was set at 50%, to appear at the center of the second readout segment. The in-plane resolution = 0.3125 mm and slice
thickness = 0.6 mm, and the image is blurred 3.84 pixels. Signal and noise were measured, respectively, to derive the SNR
before and after applying blur mitigation methods in all phantom experiments. Noise was measure from the corners of the
image and signal as the average of various ROIs, as shown in
Fig. 5.
Other than phantom, a W = 2, S = 3 pig’s tail image is
taken to demonstrate the capability of blur mitigation in in
vivo scans. The in vivo image has a 3 × 3 cm FOV, 0.156
× 0.156 mm in-plane resolution, flip angle = 30, TR/TE
Medical Physics, Vol. 41, No. 9, September 2014
= 100/9 ms, also a 1.28 ms increase in TE. Gradient strength
= 20% of 3 G/cm, ramp time = 450 μs, slew rate = 2.66 mT/
m ms, and 0.6 mm slice thickness. The two excited slices were
45 mm apart, resulting in a 2.56 pixel blur.
As For 3D implementation of ME-WMRI, the human
brain imaging has a W factor of 3, 19.2 × 19.2 × 25.6 cm3
FOV, 0.75 mm isotropic resolution, flip angle = 25◦ , TR/TE
= 30/6 ms gradient strength = 70% of 3 G/cm, ramp time
= 450 μs, slew rate = 4.66 mT/m ms. The W = 6 mice
whole body image has a 3 × 2.5 × 9 cm3 FOV, 0.156 mm
isotropic resolution, flip angle = 30◦ , TR/TE = 30/7 ms gradient strength = 50% of 3 G/cm, ramp time = 450 μs, slew
rate = 2.66 mT/m ms.
3. RESULTS
The quantitative results of blur mitigation are illustrated
in Fig. 5. The two plots on the right display detail contrast
and SNR before and after applying the blur mitigation methods. The high- and low-frequency detail contrasts of the standard image are represented by the red and black horizontal
lines on the bottom-right plot. Successful blur mitigation is
achieved when the detail contrasts approach these two lines,
respectively.
3.A. Phantom imaging
3.A.1. S = 1 image with signal enhancement
In S = 1 imaging the difference between the standard image and blurred images becomes obvious when the blur exceeds 1 pixel, both the high and low frequency stripes in the
092304-6
Wu, Chiueh, and Chen: ME-WMRI
phantom are blurred by the separation gradient (S1 L and S1
H). Having known the attenuation profile for each degree of
blur we apply signal enhancement to the acquired S = 1 kspace. After signal enhancement, the high resolution details
increase (S1 enhanced H) while the low resolution details remain the same (S1 enhanced L).
3.A.2. S = 2 blur mitigated images
It is clear from the start that the S = 2 images have better
detail contrast in high spatial frequency (S2 joined H) than the
S = 1 blurred images (S1 H) up to 2 pixel blur. However, the
low frequency contrast did not improve for any degree of blur
(S2 joined L). Such result indicates that in S = 2 coherent acquisition, high spatial frequency signals are recovered while
low frequency signals still suffer attenuation. The detail con-
092304-6
trast of high spatial frequency drops to almost zero for images
with a blurring greater than 1.8 pixels.
By enhancing the S = 2 acquired image signals, both lowand high-resolution contrasts are further improved (S2 enhanced L and S2 enhanced H). The detail contrasts of low
spatial frequency in S = 2 enhanced images (S2 enhanced L)
have the equivalent results as the S = 1 enhanced images (S1
enhanced L), while the high spatial frequency details were improved across all degrees of blur and become comparable to
the standard image.
3.A.3. W = 2 blur mitigated phantom images
Figure 6 shows the W = 2 accelerated ME-WMRI with S
= 3 sequence and signal enhancement applied. The comparison images are displayed to show the effectiveness of the blur
F IG . 6. Results of W = 2 (two-slice), S = 3 blur mitigated images. (a) A set of W = 2, 3.84 pixel blurred images with and without S = 3 blur mitigation methods
applied, compared to the standard image at the top. Image profiles of phantoms A and B are measured to quantize the difference. (b) and (c) The blur mitigated
images (red) were far superior to the blurred images (blue) in terms of sharpness and are almost comparable to the standard images (black). (c) The measured
signal, noise, and SNR of the two excited slices. The slight decrease in image signal using S = 3 sequence was recovered with signal enhancement.
Medical Physics, Vol. 41, No. 9, September 2014
092304-7
Wu, Chiueh, and Chen: ME-WMRI
092304-7
F IG . 7. S = 3 Blur mitigation applied to W = 2 in vivo imaging. (a) Three sets of pig’s tail images: standard, blurred, and blur mitigated. The image was blurred
2.56 pixels with W = 2 acceleration and S = 3 sequence, causing unclear boundaries between tissues. From the individual close-up comparisons in the (b) and
(c), the applied blur mitigation method restored the sharpness of the tissue structures.
mitigation methods. The measured image profiles of vertical
stripes show noticeable improvement after applying blur mitigation, recovering most of the peak-to-valley contrast. This
result indicates the use of a higher S can recover image detail
from a more severe degree of blur incurred by acceleration
with large W acceleration or demanding geometrical setting.
We have also measured the signal and noise of the image, respectively, and derived SNR for each slice during each blur
mitigation step.
3.B. In vivo imaging results
Figure 7 shows the potential of blur mitigation methods for
W = 2, S = 3 in vivo imaging. The axial images of a pig’s tail
were blurred 2.56 pixels, causing boundaries and structures
to be soft and indiscernible. With the blur mitigation methods applied, the contours of tissue and cartilage maintain their
sharpness, which undoubtedly helps to improve the diagnosis
accuracy.
3.C. 3D implementation
A straightforward extension of the proposed methods is
Wideband MRI in 3D, where the equivalent thickness can be
easily decreased to produce sharper images through adding
the number of phase encoding steps along slab thickness.
Shown in Fig. 8 are the sequence and results of 3D wideband
MRI. With a thinner image slice, 3D wideband MRI can be
acquired with image blur that can be neglected. One concern
is the modulated RF pulse for multiple slab excitation may
Medical Physics, Vol. 41, No. 9, September 2014
contain wider transitional bands between the excited slabs and
result in image dark bands.
4. DISCUSSION
4.A. Image quality: Detail contrast and SNR
In the S = 1 phantom image high-frequency detail contrast drops rapidly and reaches bottom at 1.8 pixel blur while
low-frequency detail contrast maintains approximately the
same value up to 1 pixel blur. After enhancing the S = 1
image, it is obvious that the low-frequency detail contrasts
have greatly improved but the high-frequency contrasts do
not change much. That is due to the attenuation at high frequency components cannot be recovered after being strongly
attenuated. On the other hand the S = 2 images show a much
better high-frequency contrast even before enhancement, but
the low-frequency contrasts are on par with the unenhanced
S = 1 images. Both low- and high-frequency contrasts of the
S = 2 images are improved after applying Wiener filter signal enhancement. The low-frequency contrasts slowly decay
and the high-frequency contrasts stay at a relatively high level
throughout all sets. In this case, the two blur mitigation methods can deal with structures of different spatial frequencies in
the phantom. It is conceivable that for images without very
sharp details, signal enhancement alone is sufficient to mitigate the blur. One can also design the attenuation profile of
the coherent acquisition scheme to retrieve desired frequency
components by adjusting the refocusing gradients and length
of each readout segments.
092304-8
Wu, Chiueh, and Chen: ME-WMRI
092304-8
F IG . 8. 3D ME-WMRI sequence and W = 3 isotropic brain and W = 6 mice whole body imaging. (a) In 3D ME-WMRI, the equivalent image thickness can
be easily made thinner with the cost of more z encoding steps. Accelerated isotropic resolution images can be acquired with minimal blurring. (b) A W = 3,
0.75 mm isotropic whole brain image was taken under 10 min of scan time whereas a standard scan will take about half an hour without other acceleration
techniques. (c) Another example is a W = 6 isotropic mice whole body imaging, the imaging time was reduced from 1 h to less than 10 min, demonstrating the
advantage of the proposed ME-WMRI imaging scheme for scanning long objects.
On the subject of image contrast, the addition of refocusing period leads to a slightly longer acquisition duration and
a longer TE if a greater receiving bandwidth was not used.
Segments of the readout will have different effective TE’s
but the final image contrast will be dominated by the effective TE where the image echo locates. In the W = 2, S = 3
experiment the TE of image echo is prolonged by 1.28 ms,
explaining the minor difference in image contrast. This difference in TE could be minimized by optimized sequence
design.
Other than longer TE, the additional separation/refocusing
gradients could cause differences in echo intensity thus affecting SNR. In the S = 2 situation where the image echo is
located in the first segment, the additional gradient switching generated a refocusing effect causing the echo to have
a slightly higher value (∼5% increase compared to standard
image). However, in the S = 3 experiments, image echo that
was acquired in the second segment shows some decrease (the
joined k-space has a ∼8% decrease compared to standard image, but was later restored by Wiener filtering). This effect
is mainly governed by the field homogeneity, gradient performance, and sequence design; fine shimming and careful
gradient adjustments could minimize the difference.
Since the number of frequency encoding increases along
with the receiving bandwidth (both W times increase), noise
per pixel in ME-WMRI remains unchanged from the conventional MRI. Therefore, the SNR per pixel in ME-WMRI
should remain approximately the same for all W factors given
equivalent excitation/reception.26
4.B. Field homogeneity and gradient performance
For an ideal gradient the switching process in coherent
acquisition sequence should be instantaneous and therefore
Medical Physics, Vol. 41, No. 9, September 2014
the resonance signal peaks caused by the refocusing effect
should appear precisely at the designated time instants. However, in real-world hardware the gradients have limited slew
rates and cannot be switched instantaneously, the ramped gradients and induced Eddy current could cause errors in readout sample values. We have inserted a few dummy sample points (a short guard-time interval) in our sequences to
serve as a buffer between each gradient switching and prevent k-data discontinuities. There will be an overlapping in
the k-space trajectory that can be either discarded or used
for gridding during image reconstruction. The guard-interval
duration depends on the capability of the gradient coils in
the MR scanners. For sequences that use greater gradient
strengths, shorter rise time, and less guard points, the accurate knowledge of k-space trajectory might be required for
reconstruction.
As for clinical considerations where the switching of gradients might cause stimulation of peripheral nervous systems,
the trapezoid gradient waveforms used in this study produced
a maximum slew rate of 6.5 mT/m ms, which is well under
the PNS stimulation threshold.27
As for transmission and reception, the use of multichanneled coils and algorithm designed pulses will increase the
excitation uniformity among slices,28–31 while traveling wave
technique could provide unvarying reception quality across a
larger coverage.32
Other than hardware limitation, the decay of FID is inevitable across the acquisition period. Major factors such as
field homogeneity, substance composition, etc., will determine the speed of decay. Such decay causes discontinuity in
the k-space data among the acquired segments and may very
well induce image artifacts. Precise shimming at the locations
of multiple-slices is suggested to avoid such issue.
092304-9
Wu, Chiueh, and Chen: ME-WMRI
092304-9
4.C. Advanced sequence design and implementation
ACKNOWLEDGMENTS
The overall effectiveness of the proposed method was contributed by k-space signal distribution and the multisegmented
sequence design. In general, the higher the number of segments, the less attenuation of the k-space signals and the
lower distortion in enhanced images. A higher S factor will
be needed to refocus the rapidly attenuating signals in scans
with higher W factor. Locations and lengths of the segments
could also be modified according to the signal distribution of
k-space to optimize blur mitigation. Considering signal enhancement, the sinc attenuation profiles of a magnetic field
of good linearity would be highly identical to the simulation
results, meaning that there is no need for attenuation profile
acquisition in each scan.
As for advanced implementations of coherent acquisition sequences, the readout segments need not to be equally
spaced along the acquisition. One can choose specific regions of the image k-space and adjust the gradients accordingly to preserve the most meaningful frequency components from being attenuated. Furthermore, a smaller S factor will be enough with schemes that employ shorter readout durations such as partial k-space methods. In combination with compressed sensing, the peaks of severely attenuated profiles can be placed at the sparsely sampled points,
obtaining fine k-space data with high W acceleration factors.
SMS imaging was put together with other methods to achieve
a higher acceleration rate in clinical studies,10–12 parallel
imaging and other acceleration techniques could be added
onto ME-WMRI to further increase the scanning throughput
as well.33
Pushing coherent acquisition sequence to the limits is a
set of fast switching separation and readout gradients with refocusing between each sampling point to prevent dephasing.
There are two advantages in this design: at first, the total acquisition time can remain unchanged since no additional time
is spent on refocusing; and second the required magnitude of
refocusing gradient would decrease.
The authors thank the interdisciplinary MRI/MRS laboratory and C.-H. Hsieh of Instrumentation Center, National
Taiwan University for MRI experiments. The work was supported by Grant No. NSC 101-2119-M-002-024-MY3. The
National Taiwan University Institutional Review Board approved the study protocol and all participants provided written
informed consent.
5. CONCLUSIONS
The proposed ME-WMRI method can effectively undo
the image blurring caused by the separation gradient during
acceleration. Coherent acquisition sequence alleviates signal
dephasing by breaking a continuous readout into multiple segments, while signal enhancement further restores the
attenuated frequency components. With the two proposed
methods, ME-WMRI can successfully mitigate the image
blur in both phantom and in vivo imaging, and thus removes the limitation on fast SMS MRI systems with a large
W factor.
With the image blurring solved by the proposed methods, the high temporal/spatial resolution capability of MEWMRI could provide a new MRI experience for practitioners and patients alike. We believe that ME-WMRI provides
new possibilities for future generations of MRI, in which
increased temporal/spatial resolution contributes greatly to
whole body screening, early cancer detection, and other new
MRI applications.34–37
Medical Physics, Vol. 41, No. 9, September 2014
a) Authors
to whom correspondence should be addressed. Electronic mail:
[email protected] and [email protected]
1 P. C. Lauterbur, “Image formation by induced local interactions: examples
employing nuclear magnetic resonance,” Nature (London) 242, 190–191
(1973).
2 K. P. Pruessmann et al., “SENSE: Sensitivity encoding for fast MRI,”
Magn. Reson. Med. 42, 952–962 (1999).
3 M. A. Griswold et al., “Generalized autocalibrating partially parallel acquisitions (GRAPPA),” Magn. Reson. Med. 47, 1202–1210 (2002).
4 M. Lustig et al., “Sparse MRI: The application of compressed sensing for
rapid MR imaging,” Magn. Reson. Med. 58, 1182–1195 (2007).
5 M. Endo et al., “Development and performance evaluation of the first
model of 4D CT-scanner,” IEEE Trans. Nucl. Sci. 50, 1667–1671 (2003).
6 S. Mori et al., “Physical performance evaluation of a 256-detector row CTscanner for 4-dimensional imaging,” Med. Phys. 31, 1348–1356 (2004).
7 T. Kido et al., “Cardiac imaging using 256-detector row four-dimensional
CT: Preliminary clinical report,” Radiat. Med. 25, 38–44 (2007).
8 J. B. Weaver, “Simultaneous multislice acquisition of MR images,” Magn.
Reson. Med. 8, 275–284 (1988).
9 D. J. Larkman, J. V. Hajnal, A. H. Herlihy, G. A. Coutts, I. R. Young,
and G. Ehnholm, “Use of multicoil arrays for separation of signal from
multiple slices simultaneously excited,” J. Magn. Reson. Imaging 13, 313–
317 (2001).
10 F. A. Breuer, M. Blaimer, R. M. Heidemann, M. F. Mueller, M. A. Griswold, and P. M. Jakob, “Controlled aliasing in parallel imaging results in
higher acceleration (CAIPIRINHA) for multi-slice imaging,” Magn. Reson. Med. 53, 684–691 (2005).
11 S. Moeller, E. Yacoub, C. A. Olman, E. Auerbach, J. Strupp, N. Harel,
and K. Ugurbil, “Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to high spatial
and temporal whole-brain fMRI,” Magn. Reson. Med. 63(5), 1144–1153
(2010).
12 D. A. Feinberg, S. Moeller, S. M. Smith, E. Auerbach, S. Ramanna,
M. F. Glasser, K. L. Miller, K. Ugurbil, and E. Yacoub, “Multiplexed echo
planar imaging for sub-second whole brain FMRI and fast diffusion imaging,” PLoS ONE 5(12), e15710 (2010).
13 B. Bilgic et al., “Wave-CAIPI for highly accelerated 3D imaging,” Magn.
Reson. Imaging (2014) (unpublished).
14 K. Lee, M. N. J. Paley, I. D. Wilkinson, and P. D. Griffiths, “Rapid 2 dimensional imaging using multiple acquisition micro B0 arrays (MAMBA),”
Magn. Reson. Imaging 20, 119–125 (2002).
15 M. N. J. Paley and K. J. Lee, “Methods and apparatus for magnetic resonance imaging,” US patent#6,980,001 B2 (Dec. 27, 2005).
16 M. N. J. Paley et al., “B1AC-MAMBA: B1 array combined with multipleacquisition micro B0 array parallel magnetic resonance imaging,” Magn.
Reson. Med. 49, 1196–1200 (2003).
17 K. Setsompop, B. A. Gagoski, J. R. Polimeni, T. Witzel, V. J. Wedeen, and
L. L. Wald, “Blipped-controlled aliasing in parallel imaging for simultaneous multislice echo planer imaging with reduced g-factor penalty,” Magn.
Reson. Med. 67(5), 1210–1224 (2011).
18 J. Xu, S. Moeller, J. Strupp, E. Auerbach, L. Chen, D. Feinberg, K. Ugurbil,
and E. Yacoub, “Highly accelerated whole brain imaging using alignedblipped-controlled aliasing multiband EPI,” Proc. Int. Soc. Mag. Reson.
Med. 20, 2306 (2012).
19 M. N. Paley, K. J. Lee, J. M. Wild, P. D. Griffiths, and E. H. Whitby,
“Simultaneous parallel inclined readout image technique,” Magn. Reson.
Imaging 24(5), 557–562 (2006).
20 E. L. Wu, T. D. Chiueh, and J. H. Chen, “Multislice wideband MRI: a new
dimension of MR image acceleration,” Proc. Intl. Soc. Mag. Reson. Med.
17, 342 (2009).
092304-10
Wu, Chiueh, and Chen: ME-WMRI
21 S.
P. Souza, J. Szumowski, C. L. Dumoulin, D. P. Plewes, and G. Glover,
“SIMA: Simultaneous multislice acquisition of MR images by Hadamardencoded excitation,” J. Comput. Assist. Tomogr. 12, 1026–1030 (1988).
22 G. Glover and A. Shimawaka, “POMP (phase offset multi-planar) imaging: A new high efficiency technique,” in Proceedings of the 7th Annual
Meeting of SMRM, San Francisco, CA, 1988.
23 Z. P. Liang and P. C. Lauterbur, Principles of Magnetic Resonance Imaging:
A Signal Processing Perspective (IEEE, New York, 2000), pp. 153–173.
24 M. Bernstein, K. F. King, and X. J. Zhou, Handbook of MRI Pulse Sequences (Elsevier, Amsterdam, 2004), pp. 43–57.
25 R. C. Gonzalez and R. E. Woods, Digital Image Processing 3rd ed. (Pearson Education, London, 2009).
26 E. L. Wu, T. D. Chiueh, and J. H. Chen, “A study of Wideband MR
imaging- SNR and CNR,” Proc. ISMRM 18 (2010).
27 C. L. Ham et al., “Peripheral nerve stimulation during MRI: effects of high
gradient amplitudes and switching rates,” J. Magn. Reson. Imaging 7(5),
933–937 (1997).
28 U. Katscher and P. Börnert, “Parallel RF transmission in MRI,” NMR
Biomed. 19(3), 393–400 (2006).
29 Z. Zhang et al., “Reduction of transmitter B1 inhomogeneity with transmit SENSE slice-select pulses,” Magn. Reson. Med. 57(5), 842–847
(2007).
Medical Physics, Vol. 41, No. 9, September 2014
092304-10
30 J.
Pauly, P. Leroux, D. Nishimura, and A. Macovski, “Parameter relations
for the Shinnar-Le Roux selective excitation pulse design algorithm,” IEEE
Trans. Med. Imaging 10(1), 53–65 (1991).
31 P. Balchandani, J. Pauly, and D. Spielman, “Designing adiabatic radio frequency pulses using the Shinnar-Le Roux algorithm,” Magn. Reson. Med.
64(3), 843–51 (2010).
32 D. O. Brunner et al., “Travelling-wave nuclear magnetic resonance,” Nature (London) 457, 994–998 (2009).
33 F. H. Wu, E. L. Wu, L. W. Kuo, T. D. Chiueh, and J. H. Chen, “Wideband
parallel imaging,” Proc. Intl. Soc. Mag. Reson. Med. 17, 2677 (2009).
34 E. L. Wu, T. D. Chiueh, and J. H. Chen, “3D Isotropic brain imaging using
multislice wideband MRI,” Proc. Intl. Soc. Mag. Reson. Med. 17, 2679
(2009).
35 I. T. Lin, E. L. Wu, H. C. Yang, and J. H. Chen, “HTS volume coil enhanced
SNR in Wideband mice whole body screening,” Proc. ISMRM 19, 1871
(2011).
36 E. L. Wu, K. H. Cho, T. D. Chiueh, and J. H. Chen, “Reduction of diffusion tensor imaging acquisition time with wideband MR imaging,” Proc.
ISMRM 17, 2680 (2009).
37 E. L. Wu, Y. A. Huang, T. D. Chiueh, and J. H. Chen, “2X accelerated
whole brain isotropic MRA using multislice wideband MRI,” Proc. Intl.
Soc. Mag. Reson. Med. 20, 2327 (2012).