Magnetic Resonance in Medicine 62:468 – 475 (2009)
High Resolution Diffusion-Weighted Imaging Using
Readout-Segmented Echo-Planar Imaging, Parallel
Imaging and a Two-Dimensional Navigator-Based
Reacquisition
David A. Porter,1* and Robin M. Heidemann2
Single-shot echo-planar imaging (EPI) is well established as the
method of choice for clinical, diffusion-weighted imaging with
MRI because of its low sensitivity to the motion-induced phase
errors that occur during diffusion sensitization of the MR signal.
However, the method is prone to artifacts due to susceptibility
changes at tissue interfaces and has a limited spatial resolution. The introduction of parallel imaging techniques, such as
GRAPPA (GeneRalized Autocalibrating Partially Parallel Acquisitions), has reduced these problems, but there are still significant limitations, particularly at higher field strengths, such as 3
Tesla (T), which are increasingly being used for routine clinical
imaging. This study describes how the combination of readoutsegmented EPI and parallel imaging can be used to address
these issues by generating high-resolution, diffusion-weighted
images at 1.5T and 3T with a significant reduction in susceptibility artifact compared with the single-shot case. The technique uses data from a 2D navigator acquisition to perform a
nonlinear phase correction and to control the real-time reacquisition of unusable data that cannot be corrected. Measurements on healthy volunteers demonstrate that this approach
provides a robust correction for motion-induced phase artifact
and allows scan times that are suitable for routine clinical
application. Magn Reson Med 62:468 – 475, 2009. © 2009
Wiley-Liss, Inc.
Key words: readout-segmented EPI; parallel imaging; GRAPPA;
2D navigator; diffusion weighted imaging
The application of diffusion-weighted imaging in clinical
studies is well established, particularly in the evaluation
of acute stroke (1–3). These studies typically rely on single-shot diffusion-weighted echo-planar imaging (EPI; 4)
which provides reliable clinical images free from motioninduced artifact. However, as is well documented, singleshot EPI is sensitive to susceptibility artifacts at tissue
interfaces and suffers from a limitation to the maximum
resolution that can be achieved. Although the introduction
of parallel imaging techniques to single-shot acquisition
methods like EPI (5– 8) has reduced these effects, there are
still significant residual problems. This is particularly true
at higher field strengths where there is an increased level
1Siemens
AG, Healthcare Sector, Erlangen, Germany.
Planck Institute for Human Cognitive and Brain Sciences, Department of
Neurophysics, Leipzig, Germany.
*Correspondence to: David A. Porter, Siemens AG, Healthcare Sector, MED
MR PLM AW Neurology, Allee am Roethelheimpark 2, 91052 Erlangen, Germany. E-mail: david.a.porter@siemens.com
Received 25 August 2008; revised 19 February 2009; accepted 25 February
2009.
DOI 10.1002/mrm.22024
Published online 15 May 2009 in Wiley InterScience (www.interscience.
wiley.com).
2Max
© 2009 Wiley-Liss, Inc.
of susceptibility artifacts and increased blurring due to a
shorter T*2 relaxation time. The effect of field strength is a
significant consideration because of the increasing amount
of routine clinical imaging that is being performed at field
strengths of 3 Tesla (T) and above.
Motion during the diffusion-sensitizing gradients leads
to a spatially dependent phase variation that is different
from one excitation to the next. This produces a very high
level of artifact if standard multishot imaging techniques
are used. In the case of rigid body motion, this phase
variation is a linear function of position (9,10). However,
the effect of cerebrospinal fluid (CSF) pulsation on the
brain is to induce deformations, which do not conform to
a rigid-body model. In this case, the resulting phase variation has a more general nonlinear behavior in two dimensions (11).
To compensate for this phase variation, navigator echo
acquisitions were introduced into multishot sequences to
monitor the shot-to-shot phase changes and apply a suitable correction to the data (9,10). Initial implementations,
based on spin-echo sequences, used an additional echo to
reacquire the central line of k-space at each excitation and
applied a linear phase correction along the readout direction. Subsequently, the technique was extended to use 2D
navigators to perform linear phase correction in both readout and phase-encoding directions in interleaved segmented EPI acquisitions (12,13).
More recently, alternative sequence types and modified
phase-correction techniques have made it possible to apply a nonlinear, 2D phase-correction using navigator data.
This is an essential requirement to achieve a robust correction for the phase errors that occur in diffusionweighted imaging of the brain. This type of correction
typically involves transforming to the image domain and
performing a complex multiplication of the imaging signals with data derived from the corresponding 2D navigator. It is also possible to apply an equivalent correction
using a direct deconvolution of the k-space data (11). As
only a subset of k-space points are sampled at each excitation, the image domain data used in this process do not
provide an accurate representation of the actual 2D spatial
variation of signal. In particular, k-space sampling strategies that do not fulfill the Nyquist condition for the required field of view (FOV) at each shot result in aliasing,
which complicates the image-domain 2D phase correction.
Promising results have been shown using iterative phase
correction algorithms to address the problem of aliasing in
several sequences, including true FISP (11), spiral imaging
(14), and interleaved EPI (15). However, these techniques
468
EPI With Parallel Imaging and 2D Reacquisition
require intensive computation and may not always provide a robust estimate for the required phase correction.
The image-domain 2D nonlinear phase correction is
much less challenging when acquisition schemes are used
that fulfill the Nyquist condition by sampling a contiguous
set of k-space points at each excitation. In this case, an
unaliased image can be reconstructed from the data from
each shot and the phase correction can be applied as a
noniterative complex multiplication. One sequence that
fulfils this condition is diffusion-weighted PROPELLER
(16), which uses a train of spin echoes to sample a band of
contiguous k-space sample points, centered at the k-space
origin. At each shot, the orientation of the band is varied to
provide complete k-space coverage. The spin echo train
makes the method insensitive to susceptibility-based distortion, but introduces the possibility of artifacts relating
to the failure of the CPMG condition (Carr-Purcell-Meiboom-Gill) due to motion-induced phase shifts during the
diffusion preparation (17). This problem is more severe at
higher fields due to a greater deviation in refocusing pulse
angle from 180 degrees due to the increased B1 field variation. An alternative variant of PROPELLER avoids the
CPMG issue by using an EPI echo-train to acquire the
blades (18), but is sensitive to blurring related to T*2 decay
and off resonance effects.
Another sequence, which meets the Nyquist sampling
condition at each shot, is diffusion-weighted, readout-segmented EPI (19), which can be modified to acquire a 2D
navigator echo for nonlinear phase correction (20). This
technique uses a much shorter echo-spacing than singleshot EPI, thereby reducing susceptibility and T*2 blurring
artifacts. The sequence uses a Cartesian k-space sampling
scheme, resulting in fewer spin excitations and a shorter
scan time than a corresponding PROPELLER sequence.
Although 2D navigator correction is usually effective at
removing the shot-to-shot nonlinear phase variation in
multishot diffusion-weighted imaging, the correction fails
for severely corrupted data sets, in which signal voids
occur in the navigator images. These signal voids correspond to a shift of low spatial-frequency components outside of the k-space acquisition window used for the navigator. In these cases, a complete correction of the data is
not possible. Although this is an infrequent problem, it is
essential that the issue is addressed to provide a robust
technique that can be used with confidence in clinical
studies. One way to deal with the problem is to use cardiac
gating to avoid acquiring data from the affected slices
during the systolic part of the cardiac cycle, which corresponds to the highest level of CSF pulsation and related
brain deformation (21,22). In addition to the loss of efficiency compared with ungated acquisitions, this approach
to the problem has additional practical disadvantages, including variability in repetition time (TR) and interval
between heartbeats. As an alternative, some multishot
techniques address the problem of uncorrectable data sets
by simply omitting these data from the reconstruction (16).
This is obviously only an option if there is sufficient redundancy in the sampling scheme for data from some
shots to be discarded without severely degrading the final
image.
A more elegant and efficient approach to the problem of
uncorrectable data is to reacquire the unusable data as part
469
FIG. 1. Sequence diagram for a single readout segment: Following
a diffusion preparation (GD), two spin echoes are used to sample
imaging and navigator echo. For the first spin echo a variable
amplitude prephasing gradient pulse (colored blue) is applied to
define an offset along kx, which varies from shot to shot to sample
a different segment in k-space. For the second echo a constant
amplitude prephasing gradient pulse is used to acquire the central
kx segment every time.
of the acquisition process. Previous work with 1D navigators demonstrated that an appropriate reacquisition
scheme can substantially reduce artifacts in diffusionweighted spin-echo images, while limiting the additional
scan time required for the reacquisition (23,24).
This study describes an extension of this reacquisition
technique to 2D navigator sequences, in which the navigator data are used during the scan to identify and resample data that cannot be corrected by 2D nonlinear
phase correction. It is demonstrated that this 2D-navigatorbased reacquisition scheme can be combined with readout-segmented EPI and parallel imaging to provide a robust method for high resolution diffusion-weighted imaging, which has a low level of image distortion and is
insensitive to motion-induced phase error. A preliminary
account of this work has been presented elsewhere (25,26).
Also note that subsequent clinical application of diffusionweighted (DW) readout-segmented EPI with GRAPPA
(GeneRalized Autocalibrating Partially Parallel Acquisitions) has also shown promising results in other studies
(27).
THEORY AND METHODS
Readout-Segmented EPI
The readout-segmented EPI sequence used in this study is
shown schematically in Figure 1. After a diffusion preparation, two spin echoes were used to acquire imaging and
navigator data, respectively, using a sinusoidal EPI readout, which sampled a subset of contiguous points in the
readout direction. For the first spin echo, a variable amplitude prephasing pulse (colored blue in Fig. 1) was applied to define an offset along kx, which was varied from
one shot to the next. In this manner, each shot was used to
sample a different segment in the kx direction (see Fig. 2).
The smaller readout gradient moment, corresponding to
the reduced kx coverage, allows a substantially shorter
470
FIG. 2. Depiction of the k-space coverage for readout segmented
EPI. In this example, a five-shot EPI acquisition is used, which
divides k-space into five segments along the readout direction.
Each shot acquires data from a set of contiguous k-space sample
points, which is beneficial for 2D navigator phase correction. In
practice, a small overlap is used at the interface between readout
segments (see main text). [Color figure can be viewed in the online
issue, which is available at www.interscience.wiley.com.]
echo-spacing than with single-shot EPI, significantly reducing the effect of susceptibility and T*2 decay. The second echo was used to generate 2D navigator data by sampling the central kx segment at each shot. A blipped phaseencoding gradient was used to traverse the full ky range.
A small overlap was used between the individual readout segments, which for clarity is not shown in Figure 2.
This overlap serves two purposes. The first purpose relates
to the regridding required for the sinusoidal readout gradient waveform. By sampling additional kx points at the
start and end of each readout, the regridded kx points that
correspond to the actual interface between readout segments can be more precisely determined by including the
contribution from these additional points in the convolution process used during regridding. It is critical that a
precise regridding process is applied to avoid a mismatch
at the interface between the readout segments. The second
purpose for the overlap concerns the 2D navigator correction and is addressed in detail below. For the images
acquired in this study a fixed overlap of eight kx sample
points was used, so that four additional data points were
sampled at the start and end of each readout.
With readout-segmented EPI it is essential that the kspace trajectory along kx is carefully controlled so that
there are no discontinuities at the interface between readout segments, which can cause ghosting in the final image.
This places some strict requirements on the data sampling
and gradient hardware, which may not always be fulfilled.
In particular, the readout gradient moment sampled at
each echo in the EPI echo-train must be consistent with the
increments used for the readout prephasing pulse (colored
Porter and Heidemann
blue in Fig. 1). Small discrepancies can arise between
these two gradients if there is a temporal displacement
between the data sampling window and the applied readout gradient or if the actual readout gradient waveform
deviates from the ideal case. During the current work, the
time-shift problem was overcome by using a predetermined gradient delay and applying a standard first-order
phase correction along the kx dimension of the raw data
using non–phase-encoding reference data. In addition, to
correct the readout gradient moment arising from deviations of the waveform from the ideal sinusoidal shape, the
gradient amplitude was manually adjusted to ensure optimum image quality. This was achieved by acquiring a set
of phantom images with gradient amplitude scaling factors
in the range 0.98 to 1.02. A visual inspection of the images
was then used to select the optimum scaling factor, corresponding to the image with minimum ringing artifact in
the readout direction. Typically, this optimum scaling factor represented a change in gradient amplitude of less than
1% compared with the nominal value. This optimum scaling factor was then used to scale the readout gradients
during the subsequent acquisition of images in vivo.
The readout-segmented EPI technique was combined
with parallel imaging by undersampling k-space in the ky
direction during both imaging and navigator echo-trains.
This provides a further reduction in susceptibility and T*2
decay artifacts by shortening the EPI echo-train length and
decreasing the effective echo-spacing. The GRAPPA
method (28) was used to reconstruct the undersampled
data, for which the autocalibration signals (ACS) were
acquired using a single-shot EPI readout that sampled data
from the central region of k-space in both kx and ky directions.
2D-Navigator-Based Reacquisition
During the measurement, a reacquisition process was used
to repeat scans that resulted in unusable data. These unusable data sets are characterized by navigator images with
a large nonlinear phase error, which corresponds to a
broadening of the signal distribution in k-space. This kspace broadening was used to identify the affected scans
directly from the k-space data by using the width of the
signal distribution in the kx direction as a measure for the
extent of the nonlinear phase error.
This distribution width, Wx, was characterized using the
following expression for the first moment of the signal
distribution along the kx direction:
Wx ⫽
冘冘 冘
i
ky
兩S i共k x,k y兲兩 䡠 兩k x ⫺ ⍀ x兩
kx
[1]
where the summation is performed over the set of sampled
k-space points in the 2D navigator for each receive coil i, Si
is the complex k-space signal for each receive coil and ⍀x
is the kx coordinate corresponding to the maximum signal
amplitude when summed across all receive channels (this
is taken to represent the centre of the signal distribution).
Note that this expression is particularly convenient for
parallel imaging applications, as it can be applied directly
to undersampled data sets without any further processing.
This makes it suitable for the real-time calculations that
are required during the reacquisition process.
EPI With Parallel Imaging and 2D Reacquisition
The reacquisition was performed at the end of the standard measurement when all readout segments had been
scanned for all b-values and diffusion-encoding directions. During reacquisition an independent decision about
which scans to reacquire was made for each slice position.
The first step in this decision-making process was to compare all shots for a given image (i.e., for a given b-value and
diffusion direction) and identify the segment with the
smallest distribution width Wx. This width was then used
as a reference point for all other readout segments for that
image and in each case a relative distribution width was
calculated. Segments with relative widths exceeding an
empirical threshold of 1.05 times the reference width were
marked for reacquisition. This process was repeated for all
images to identify the complete set of scans to be reacquired. The order of the reacquisitions was weighted toward data sets with large relative distribution widths but,
similar to the approach taken in previous work (23), an
increased weighting was also given to readout segments
closest to the centre of k-space, where corrupt data has a
greater influence on image quality. In addition, an overall
time limit was placed on the time spent reacquiring data,
which was set to 20% of the standard measurement time.
Image Reconstruction
After using a standard regridding procedure (29) along the
kx direction to compensate for the sinusoidal readout gradient waveform, Nyquist ghost phase correction was performed using reference data from a non–phase-encoded
EPI echo-train. An adapted GRAPPA reconstruction was
then applied to the regridded, Cartesian sampled data from
each readout segment to reconstruct the missing ky data
points. For this GRAPPA reconstruction, the coil weights
were derived from the ACS data using a Moore-Penrose
pseudo inverse and a 2D convolution kernel (30) was used
with four source points along the phase-encoding direction and three source points along the readout direction.
The same processing was applied to both imaging and
navigator data, yielding in each case a separate k-space
data set for each receive channel.
2D nonlinear navigator phase correction was then applied independently to the data from each readout segment and receive channel as follows. The imaging data
were zero-filled and Fourier transformed to provide a complex image domain data set. In a similar way, the corresponding 2D navigator data were also transformed to the
image domain, although in this case a k-space Hanning
filter was applied in the kx direction to minimize truncation artifacts. The navigator data were then normalized
and used to phase correct the imaging data by performing
a pixel by pixel complex multiplication. Although the
image domain data from an individual readout segment
only represent a subset of the spatial frequencies required
for the final image, this image domain phase correction
procedure is not corrupted by aliased signal contributions
because the sampled points are contiguous in k-space,
thereby fulfilling the Nyquist sampling condition for the
specified FOV.
One important point to consider in relation to this phase
correction procedure is the effect of the limited number of
kx points in the segment of data being corrected. The
471
nonlinear, image-domain phase correction corresponds to
a redistribution of the raw data signals in k-space, in
which signals from some regions of the object are shifted
by more than those from other regions, according to the
local motion-induced phase variations. Due to this k-space
redistribution, a potential problem arises at the edge of the
readout segment when the effect of the phase correction is
to try to recover data that have been shifted outside the
segment. These data are of course not available because
they fall outside the sampled region, thereby limiting the
effectiveness of the phase correction. This provides a further motivation for sampling additional kx points at the
start and end of each readout event. This has the effect of
making the sampled region wider in the kx direction than
required, so that shifted signals at the edge of the readout
segment can be recovered. Note also that the zero-filling in
the kx direction at the start of the phase-correction procedure is necessary to avoid aliasing in k-space when the
image-domain phase correction shifts k-space signals outside of the sampled region.
After navigator phase correction, the data were transformed back to k-space and the additional readout columns at the edge of each segment were discarded. The data
from all readout segments were then combined, before
applying a final transformation back into the image domain to produce a separate image for each receive channel.
Image data from all receive channels were then combined
in a standard way using the square root of the sum of
squares.
Imaging Experiments
The readout-segmented EPI sequence with 2D navigatorbased reacquisition was implemented on a 1.5T MAGNETOM Avanto system and a 3T MAGNETOM Trio system
(Siemens Healthcare, Erlangen, Germany). For protocols
without parallel imaging, image reconstruction was performed on the scanner using a modified reconstruction
program and for protocols with parallel imaging, the reconstruction was performed off-line using MATLAB (The
Mathworks Inc., Natick, MA). All studies were performed
on healthy volunteers without cardiac triggering using
standard 12-channel head coils. Informed consent was
obtained from all subjects before each study.
At 1.5T, readout-segmented EPI images were acquired
with the following parameters: no parallel imaging, FOV
230 mm, matrix 224 ⫻ 224, pixel size 1.0 mm ⫻ 1.0 mm,
slices 17, slice thickness 5 mm, number of readout segments per image 11, echo-spacing 300 s, TR 3860 ms,
echo time (TE) 82 ms, one scan at b ⫽ 0 s/mm2 and three
at b ⫽ 1000 s/mm2 in orthogonal directions, one average,
total measurement time (including reacquisitions, a
Nyquist ghost phase-correction scan and a dummy scan for
magnetization preparation) 3 min 25 s.
At 3T, readout-segmented EPI images were acquired
with the following parameters: parallel imaging using
GRAPPA with an acceleration factor of 2, FOV 210 mm,
matrix 224 ⫻ 224, pixel size 0.9 mm ⫻ 0.9 mm, slices 19,
slice thickness 5 mm, number of readout segments per
image 11, echo-spacing 320 s, TR 3380 ms, TE 68 ms, one
scan at b ⫽ 0 s/mm2 and three at b ⫽ 1000 s/mm2 in
orthogonal directions, one average, total measurement
472
Porter and Heidemann
FIG. 3. Example showing the effect of the 2D navigator based reacquisition. a: The navigator data from the central seven shots of a
diffusion-weighted data set before data reacquisition. Top row: low resolution navigator images; bottom row: corresponding k-space data.
Shot 6 has a signal void artifact in the navigator image, which is seen as a widened signal distribution in k-space. b,c: The corresponding
artifact in the reconstructed diffusion- and trace-weighted images, respectively. d: The navigator data after reacquisition. Data sets with
wide k-space signal distributions have been replaced by newly acquired data. e,f: The corresponding artifact-free images reconstructed
using the reacquired data.
time (including reacquisitions, a Nyquist ghost phase-correction scan, a GRAPPA auto-calibration scan and a
dummy scan for magnetization preparation) 3 min 6 s.
At 3T, standard DW single-shot EPI images were also
acquired for comparison using the following parameters:
parallel imaging using GRAPPA with an acceleration factor of 2, FOV 230 mm, matrix 192 ⫻ 192, phase partial
Fourier factor 6/8, pixel size 1.2 mm ⫻ 1.2 mm, slices 19,
slice thickness 5 mm, echo-spacing 900 s, TR 2800 ms,
TE 87 ms, one scan at b ⫽ 0 s/mm2 and three at b ⫽ 1000
s/mm2 in orthogonal directions, four averages, total measurement time (including a GRAPPA autocalibration scan
and a dummy scan for magnetization preparation) 50 s.
RESULTS
Figure 3 shows data from a 1.5T examination, which illustrates the effectiveness of the reacquisition procedure for
an imaging slice at the base of the brain. The top row of
Figure 3a shows the low-resolution navigator images corresponding to the central seven readout segments (or
shots) of the standard measurement before data reacquisition. The corresponding k-space data in modulus format
are shown in the second row of Figure 3a. With the exception of shot number six, the navigator images and corresponding k-space signal distributions are very similar. The
navigator image from shot six, however, has bilateral signal voids in the anterior part of the cerebellum with a
corresponding wide signal distribution in k-space. The
effect of these corrupt data is to produce a localized artifact
in the final reconstructed image as shown in Figure 3b.
The artifact is most noticeable on the right side of the brain
in the region indicated by the arrow in the figure. Similarly, Figure 3c shows the effect of the corrupt data when
the image from Figure 3b is used to produce a traceweighted image (by taking the geometric mean of the three
DW images with mutually orthogonal diffusion-encoding
directions).
Figure 3d shows the corresponding set of navigator images after the automatic data reacquisition procedure,
which consisted of 7 reacquisitions, requiring an additional 27 s of measurement time. The readout segments
with relatively large k-space signal distributions have been
reacquired, which is most noticeable with shot number
six. As shown in Figure 3d,e, the modified data produce
images without the artifact seen with the standard data set.
The trace-weighted image in Figure 3e demonstrates that
the readout-segmented EPI method can provide good quality diffusion-weighted images at 1.5T with low susceptibility artifact. Note that these images were acquired without parallel imaging, which would provide a further reduction in susceptibility and T*2 decay effects.
Figures 4 and 5 provide a comparison between readoutsegmented EPI and single-shot EPI at 3T. In both figures,
the top row shows the single-shot EPI images, acquired
using a GRAPPA acceleration factor of 2, and the bottom
rows show the corresponding high-resolution readout-segmented EPI images, also acquired using a GRAPPA acceleration factor of 2. In this case, the automatic reacquisition
procedure resulted in eight reacquisitions, corresponding
to an additional 27 s of measurement time. In both figures,
the left column shows images with a nominal b-value of
zero and the right column shows trace-weighted images
with a nominal b-value of 1000 s/mm2. Figure 4 shows
comparative images for a slice position at the base of the
brain, where susceptibility artifacts are generally at their
most severe. This region is also particularly challenging
for multishot DW imaging due to the high level of CSF
pulsation around this part of the brain, leading to large
nonlinear phase errors. The readout-segmented EPI data
show a substantial reduction in susceptibility artifact and
anatomical detail compared with the single-shot images
EPI With Parallel Imaging and 2D Reacquisition
473
this blurring effect is more significant at 3T than at 1.5T,
making it more difficult to achieve a high-resolution with
single-shot EPI.
The readout-segmented EPI method significantly reduces these limitations by allowing a shorter echo-spacing
in the EPI echo-train compared with the single-shot case
due to the smaller number of kx points that are sampled
during each readout gradient. The method is easily combined with GRAPPA, using the same methodology as that
developed for single-shot EPI, resulting in an effective
echo-spacing that is many times shorter than that of the
equivalent single-shot EPI scan with the same resolution.
Taking the GRAPPA acceleration factor of 2 into account,
the readout-segmented EPI images in Figures 4 and 5,
correspond to an effective echo-spacing of 160 s, whereas
an equivalent single-shot EPI protocol with GRAPPA
would have had an effective echo-spacing of around 700
s. In the case of the comparison shown in Figure 4, the
single-shot EPI scan still had a significantly longer effective echo-spacing (450 s) despite the lower resolution.
Furthermore, for the single-shot EPI measurement there
was a time period of 43.2 ms between the acquisition of the
central k-space echo and the end of the echo-train, where
the outer ky line was sampled. The equivalent time period
of 17.9 ms for the readout-segmented EPI sequence was
much shorter despite the larger k-space coverage. Consequently, the improvement in the anatomical detail seen in
the readout-segmented EPI images of Figures 4 and 5 is
FIG. 4. Direct comparison between single-shot EPI and readout
segmented EPI at the base of the brain. a,b: Single-shot EPI acquisition with b ⫽ 0 s/mm2 (a) and corresponding trace-weighted
image with b ⫽ 1000 s/mm2 (b). c,d: Readout segmented EPI with
b ⫽ 0 s/mm2 (c) and corresponding trace-weighted image with b ⫽
1000 s/mm2 (d)
and without any evidence of residual motion-induced
phase error. Similarly, Figure 5d shows typical artifactfree images for an axial slice at the level of the lateral
ventricles, a region that is also affected by CSF pulsation.
Although single-shot EPI images at this slice position do
not exhibit a marked susceptibility artifact, the readoutsegmented EPI images still demonstrate a higher image
quality due to the substantial improvement in anatomical
detail.
DISCUSSION
As seen in Figure 4, standard single-shot EPI protocols are
prone to significant susceptibility artifact at 3T, even when
parallel imaging techniques, such as GRAPPA, are used to
reduce the effective echo-spacing. In addition, although
parallel imaging has made it possible to increase the resolution that can be achieved with single-shot EPI, by reducing the length of the echo-train for a given k-space
coverage, the resolution is still limited compared with
other MR imaging methods. This resolution limitation is
imposed by the T*2 decay during the echo-train, which
results in a blurring of the signal in the phase-encoding
direction and sets a practical limit on the length of echotrain that can be used for data acquisition. Furthermore,
because T*2 values reduce with increasing field strength,
FIG. 5. Direct comparison between single-shot EPI and readout
segmented EPI in the brain at the level of the lateral ventricles. a,b:
Single-shot EPI acquisition with b ⫽ 0 s/mm2 (a) and corresponding
trace-weighted image with b ⫽ 1000 s/mm2 (b). c,d: Readout segmented EPI with b ⫽ 0 s/mm2 (c) and corresponding trace-weighted
image with b ⫽ 1000 s/mm2 (d).
474
attributable both to the higher nominal resolution and to a
reduction in T*2 blurring effects.
The implementation of readout-segmented EPI described in this study did not use partial Fourier in the
phase-encoding direction, as is typically the case with
single-shot EPI protocols, including those used in the current work. Although this would also be an option for the
readout-segmented EPI technique, the reduction in TE
would be less significant than with single-shot EPI due to
the shorter echo-train. Consequently, it might be preferable to use a full k-space acquisition along ky and avoid
some of the difficulties that are associated with the reconstruction of phase partial Fourier diffusion-weighted EPI
data, leading to artifact or loss of spatial resolution (31).
Another advantage to using a full ky acquisition, which
could be explored in future work, is that it may allow some
of the readout segments on one side of k-space to be
omitted from the acquisition and a partial Fourier reconstruction used for the kx direction. This would have the
benefit of decreasing the number of spin excitations required for a given image resolution, which would be useful
for scan time reduction, particularly if the technique were
used to scan multiple diffusion directions in a DTI study.
This study has demonstrated that the readout-segmented EPI acquisition scheme can be combined with a
simple image-based complex multiplication to provide a
robust correction for the nonlinear phase errors that arise
from subject motion in diffusion-weighted imaging. Compared with acquisition schemes that do not sample contiguous k-space sample points, there is a considerable
reduction in computational effort during image reconstruction, as there is no requirement for an iterative phase
correction procedure.
All imaging techniques, which rely on 2D navigator
phase correction will fail when the spatial frequency of the
motion-induced phase errors becomes too large. As illustrated by the data shown in Figure 3, the incidence of these
highly corrupt data sets is quite low for diffusion-weighted
imaging of the brain, affecting particular brain regions at
certain points in the cardiac cycle. This makes it possible
to avoid the corresponding image artifacts by reacquiring a
small subset of scans, which are identified as unusable in
real time during the acquisition. A measure of the relative
width of the k-space signal distribution has been shown by
this study to be an effective way of controlling this reacquisition process. The procedure adopted in this work
used a relative width threshold of 1.05 and restricted the
maximum time spent reacquiring data to 20% of the standard measurement time. In preliminary measurements,
this approach was found to be sufficient to avoid the
occasional artifacts that otherwise occur. To ensure a good
compromise between scan time and artifact suppression,
further work with a large number of subjects is required to
fully optimize these two criteria used to determine the
number of reacquisitions in a range of measurement protocols.
A recent clinical application of the DW segmented EPI
method (27) acquired images with an implementation of
the sequence that did not include the navigator-controlled
reacquisition. To compensate for this, these acquisitions
used a very large overlap between readout segments, corresponding to as much as 50% of the width of the segment,
Porter and Heidemann
making it possible to correct for large k-space signal displacements. A disadvantage to this approach is that a
substantial increase in the overlap region is required to
accommodate the kind of k-space distribution seen in the
navigator data of shot number 6 in Figure 3. The resulting
increase in the width of the readout segments leads, in
turn, to a significant increase in the echo-spacing and TE
compared with a reacquisition strategy, in which these
infrequent, highly corrupt data sets are discarded. Nevertheless, this approach may be a useful option in the absence of real-time feedback between data analysis and data
acquisition.
There are two further outstanding issues that affect the
routine application of DW readout-segmented EPI. The
first of these is the manual adjustment used in this work to
scale the readout gradient to ensure k-space continuity at
the interfaces between readout segments. As seen in the
images of Figures 3–5, this adjustment procedure is effective at avoiding ringing artifact in the subsequent images
acquired in vivo. However, this process will need to be
automated for routine use, so that adjustment parameters
can be easily established for a wide range of measurement
protocols. This could be done either by using an automated version of the adjustment process described above,
or by leaving the readout amplitude fixed and using information from a k-space trajectory measurement to re-grid
the data along kx (32). In this latter approach the additional
kx points, sampled at the edges of readout segments, would
allow the effective moment of the readout gradient to be
increased if required.
The second outstanding issue is that of rigid-body motion. The 2D image-domain phase correction used in the
current work addresses the problem of motion during the
diffusion preparation, but not the problem of rigid-body
motion between the acquisition of different readout segments, which leads to a loss of detail in the final image due
to misregistration of the data. In future work, this rigidbody problem could be addressed by performing an image
registration of the navigator images from different shots
and using the information to realign the data from the
individual readout segments. This could be achieved using techniques derived from motion correction procedures
described elsewhere, such as those used with the selfnavigated PROPELLER sequence (33).
CONCLUSION
This study has described a new multishot technique for
acquiring high-resolution DW images, which have low
susceptibility based image distortion and T*2 blurring and
a robust correction for motion-induced phase artifact. The
technique provides significant image quality improvement
compared with DW single-shot EPI at 3T with relatively
short scan times, making the technique of interest for DWI
and DTI in routine clinical studies.
REFERENCES
1. Warach S, Gaa J, Siewert B, Wielopolski P, Edelman RR. Acute human
stroke studied by whole brain echo planar diffusion-weighted magnetic
resonance imaging. Ann Neurol 1995;37:231–241.
EPI With Parallel Imaging and 2D Reacquisition
2. van Everdingen KJ, van der Grond J, Kappelle LJ, Ramos LM, Mali WP.
Diffusion-weighted magnetic resonance imaging in acute stroke. Stroke
1998;29:1783–1790.
3. Wu O, Koroshetz WJ, Ostergaard L, Buonanno FS, Copen WA, Gonzalez
RG, Rordorf G, Rosen BR, Schwamm LH, Weisskoff RM, Sorensen AG.
Predicting tissue outcome in acute human cerebral ischemia using
combined diffusion- and perfusion-weighted MR imaging. Stroke 2001;
32:933–942.
4. Turner R, Le Bihan D, Maier J, Vavrek R, Hedges LK, Pekar J. Echoplanar imaging of intravoxel incoherent motion. Radiology 1990;177:
407– 414.
5. Griswold MA, Jakob PM, Chen Q, Goldfarb J, Manning WJ, Edelman RR,
Sodickson DK. Resolution enhancement in single-shot imaging using
simultaneous acquisition of spatial harmonics (SMASH). Magn Reson
Med 1999;41:1236 –1245.
6. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Wang J, Kiefer B,
Haase. A Off-resonance artifacts in single shot EPI using partially
parallel imaging. In: Proceedings of the 9th Annual Meeting of ISMRM,
Glasgow, Scotland, 2001. (abstract 446).
7. Heidemann RM, Griswold MA, Porter D, Kiefer B, Nittka M, Wang J,
Haase A, Jakob PM. Minimizing distortions and blurring in diffusion
weighted single shot EPI using high performance gradients in combination with parallel imaging. In: Proceedings of the 9th Annual Meeting of ISMRM, Glasgow, Scotland, 2001. (abstract 169).
8. Bammer R, Keeling SL, Augustin M, Pruessmann KP, Wolf R, Stollberger R, Hartung HP, Fazekas F. Improved diffusion-weighted singleshot echo-planar imaging (EPI) in stroke using sensitivity encoding
(SENSE). Magn Reson Med 2001;46:548 –554.
9. Ordidge RJ, Helpern JA, Qing ZX, Knight RA, Nagesh V. Correction of
motional artefacts in diffusion-weighted MR images using navigator
echoes. Magn Reson Imaging 1994;12:455– 460.
10. Anderson AW, Gore JC. Analysis and correction of motion artifacts in
diffusion weighted imaging. Magn Reson Med 1994;32:379 –387.
11. Miller KL, Pauly JP. Nonlinear phase correction for navigated diffusion
imaging. Magn Reson Med 2003;50:343–353.
12. Butts K, Pauly J, de Crespigny A, Moseley M. Isotropic diffusionweighted and spiral-navigated interleaved EPI for routine imaging of
acute stroke. Magn Reson Med 1997;38:741–749.
13. Atkinson D, Porter DA, Hill DLG, Calamante F, Connelly A. Sampling
and reconstruction effects due to motion in diffusion-weighted interleaved echo-planar imaging. Magn Reson Med 2000;44:101–109.
14. Liu C, Bammer R, Kim D, Moseley ME. Self-navigated interleaved
spiral (SNAILS): application to high-resolution diffusion tensor imaging. Magn Reson Med 2004;52:1388 –1396.
15. Atkinson D, Counsell S, Hajnal JV, Batchelor PG, Hill DL, Larkman DJ.
Nonlinear phase correction of navigated multi-coil diffusion images.
Magn Reson Med 2006;56:1135–1139.
16. Pipe JG, Farthing VG, Forbes KP. Multishot diffusion-weighted FSE
using PROPELLER MRI. Magn Reson Med 2002;47:42–52.
17. Liu G, van Gelderon P, Duyn J, Moonen CTW. Single-shot diffusion
MRI of human brain on a conventional clinical instrument. Magn
Reson Med 1996;35:671– 677.
18. Wang F-N, Huang T-Y, Lin F-H, Chuang T-C, Chen N-K, Chung H-W,
Chen C-Y, Kwong KK. PROPELLER EPI: an MRI technique suitable for
475
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
diffusion tensor imaging at high field strength with reduced geometric
distortions. Magn Reson Med 2005;54:1232–1240.
Robson MD, Anderson AW, Gore JC. Diffusion-weighted multiple-shot
echo planar imaging of humans without navigation. Magn Reson Med
1997;38:82– 88.
Porter DA, Mueller E. Multi-shot diffusion-weighted EPI with readout
mosaic segmentation and 2D navigator correction. In: Proceedings of
the 12th Annual Meeting of ISMRM, Kyoto, Japan, 2004. (abstract 442).
Wirestam R, Greitz D, Thomsen C, Brockstedt S, Olsson MB, Stahlberg
F. Theoretical and experimental evaluation of phase-dispersion effects
caused by brain motion in diffusion and perfusion MR imaging. J Magn
Reson Imaging 1996;6:348 –355.
Nunes RG, Jezzard P, Clare S. Investigations on the efficiency of cardiac-gated methods for the acquisition of diffusion-weighted images. J
Magn Reson 2005;177:102–110.
Nguyen Q, Clemence M, Ordidge RJ. The use of intelligent re-acquisition to reduce scan time in MRI degraded by motion. In: Proceedings of
the 6th Annual Meeting of ISMRM, Sydney, Australia, 1998. (abstract
134).
Nguyen Q, Clemence M, Thornton J, Ordidge RJ. Isotropic diffusionweighted multishot imaging using automatic reacquisition. In: Proceedings of the 7th Annual Meeting of ISMRM, Philadelphia, Pennsylvania, USA, 1999. (abstract 559).
Porter DA, Heidemann RM. Multi-shot, diffusion-weighted imaging at
3T using readout-segmented EPI and GRAPPA. In: Proceedings of the
14th Annual Meeting of ISMRM, Seattle, Washington, USA, 2006.
(abstract 1046).
Porter DA. 2D-navigator-based re-acquisition for motion artefact suppression in multi-shot, diffusion-weighted imaging. In: Proceedings of
the 14th Annual Meeting of ISMRM, Seattle, Washington, USA, 2006.
(abstract 1047).
Holdsworth SJ, Skare S, Newbould RD, Guzmann R, Blevins NH, Bammer R. Readout-segmented EPI for rapid high resolution imaging at 3T.
Eur J Radiol 2008;65:36 – 46.
Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J,
Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47:1202–1210.
Jackson JI, Meyer CH, Nishimura DG, Macovski A. Selection of a
convolution function for Fourier inversion using gridding. IEEE Trans
Med Imaging 1991;10:473– 478.
Griswold MA. Advanced k-space techniques. In: Proceedings of the
2nd International Workshop on Parallel MRI, Zurich, Switzerland,
2004. p 16 –18.
Robson MD, Porter DA. Reconstruction as a source of artifact in nongated single-shot diffusion-weighted EPI. Magn Reson Imaging 2005;
23:899 –905.
Josephs O, Deichmann R, Turner R. Trajectory Measurement and Generalised Reconstruction in Rectilinear EPI. In: Proceedings of the 8th
Annual Meeting of ISMRM, Denver, Colorado, USA, 2000. (abstract
1517).
Pipe J. Motion correction with PROPELLER MRI: application to head
motion and free-breathing cardiac imaging. Magn Reson Med 1999;42:
963–969.