ISMRM_2006-2015_compressed

6020
Ferumoxytol-enhanced Plural Contrast Imaging of the Human Brain
Samantha J Holdsworth1
, Thomas Christen1
, Kristen Yeom1
, Jae Mo Park1
, Greg Zaharchuk1
, and Michael E Moseley1
1
Department of Radiology, Stanford University, Stanford, CA, United States
Purpose: To exploit the 3D multi-echo GRE sequence coupled with and without the injection
of ferumoxytol for plural-contrast clinical neuroimaging at 3T.
Introduction: Ferumoxytol (Feraheme, AMAG Pharmaceuticals, Inc., Cambridge, MA), an
FDA-approved ultra-small paramagnetic iron oxide (USPIO) compound, has been used as an
“off label” contrast agent to study inflammatory processes, vascular lesions, tumor, or stroke [1-
6]
. It is retained in the intravascular space early after injection and does not cross an intact
blood brain barrier [7-8]
. The high magnetic susceptibility of ferumoxytol considerably reduces
the T2* and the long intravascular half-life allows high-resolution/high-SNR acquisitions that
may be useful to study brain disorders[9]
. Particularly in pediatric brain, where patient
cooperation can be problematic, we are striving to generate and study multiple image contrasts
from a single imaging sequence. The 3D multi-echo gradient-recalled echo (ME-GRE)
technique is a particularly good candidate for this, as it allows for the simultaneous generation
of naturally co-registered images with various contrasts[10-11]
. Here we show preliminary data
scanned with a 5:44-minute 3D ME-GRE sequence acquired both pre- and post ferumoxytol,
with the subsequent generation of R2* maps, local field maps, Susceptibility-Weighted
Imaging (SWI), Time-of-Flight Magnetic Resonance Angiography (TOF), and Quantitative
Susceptibility Maps (QSM).
Methods: With IRB approval, pediatric patients were scanned on a 3T GE scanner (MR750,
GE Healthcare Systems, Waukesha, WI) equipped with an 8-channel head-coil. In addition to
the regular clinical pediatric brain protocol, a flow-compensated 3D GRAPPA-accelerated
ME-GRE sequence was scanned both before and after the intravenous injection of a single
dose of ferumoxytol (0.1 mL Fe/kg). The following parameters were used: resolution =
0.57x0.86x2.5mm3
, 66 z-partitions, acceleration factor = 2, 8 echoes ranging from TE = 4.3ms
– 37.5ms with 4.7ms increments, TR = 40.8ms, scan time = 5:44min. On completion of the
scan, the raw data from the scanner were automatically reconstructed using compiled threaded
MATLAB (MathWorks Inc., Natick, MA, USA) code, with all images sent to the hospital
database (PACS). First, the 8-channel coil data were combined with the complex sum-of-
squares. Weighted magnitude images (wMag) were calculated using the echo time of each
echo image as a weighting factor. R2* (1/T2*) maps were calculated from mono-exponential
fit of echoes. Field maps were generated by first performing a complex fit across echoes[12-14]
,
followed by phase unwrapping using a Laplacian algorithm[15]
and projection onto dipole
fields[16-18]
. QSM images were generated from this field map using the MEDI algorithm[12,16-18]
.
SWI images were created by generating a phase mask (using a 2D Hanning window), and
multiplying this mask 5 times by the weighted-magnitude image[19-20]
. TOF images were
produced by taking the Maximum Intensity Projection (MIP) over the first echo.
Results: Fig. 1 shows the multiple contrasts acquired on an 11yr old post-surgical male patient
using the 0.6x0.9x2mm3
3D ME-GRE sequence. Both pre- and post-ferumoxytol as well as the
difference images are shown. The high spatial resolution and high SNR of the 3D ME-GRE
images allows the visualization of fine vascular detail. The blooming effect seen on the field
map are blood products at the surgical resection site. The R2* difference maps are proportional
to the cerebral blood volume fraction according to the steady-state perfusion theory[21]
. The
spatial resolution achieved here is however much higher than the one usually obtained using
Dynamic Susceptibility Contrast approaches. Since the vasculature was suppressed on the post-
ferumoxytol TOF images (likely due to the severe T2* shortening effect), the TOF difference
images appear background suppressed – corresponding reasonably well with the regular TOF-MRA
(Fig. 2).
Conclusion: Here we show that the use of just one 5:44-minute 3D ME-GRE sequence and subsequent post-processing toolkit has
the potential to reveal complementary image features, by providing multiple contrast mechanisms such as R2* maps, Field maps,
SWI, QSM, and TOF-MRA. This sequence has the potential to be a surrogate for other single-sequence alternatives in clinical
practice. Further work is needed to determine the extent to which these contrasts improve our understanding of normal tissue
anatomy as well as changes in tissue in various pathological conditions. This might be particularly adapted to study vascular
malformations or lesions with heterogeneous tissue components such as brain tumors.
References: [1] Neuwelt EA et al. Neurosurgery 60(4), 601-611 (2007). [2] Hunt MA et al. AJNR 26(5), 1084-1088 (2005). [3] Neuwelt EA et al
Neuropathology and applied neurobiology 30(5), 456-471 (2004). [4] Taschner CA et al, AJR. 185(6), 1477-1486 (2005). [5] Nighoghossian N et al,
Stroke. 2007 38(2):303-7. [6] Saleh A et al. Brain. 2004 Jul;127(Pt 7):1670-7. [7] Simon GH et al. Investigative radiology 41(1), 45-51 (2006). [8]
Daldrup-Link H, Coussens LM: Oncoimmunology 1(4), 507-509 (2012). [9] Christen T et al. Magn Reson Med. 2012. [10] Luo J et al. NeuroImage 60:
1073–1082 (2012) [11] Deistung A et al. JMRI 29:1478–1484 (2009). [12] Liu T et al. MRM 2013;69(2):467-76. [13] Kressler B et al. IEEE TMI
29(2):273-81 (2010) [14] de Rochefort et al. MRM 60(4):1003-1009 (2008). [15] Schofield and Zhu, Opt. Lett. 28:1194-1196 (2003). [16] Liu T et al.
NMR Biomed 24(9):1129-36 (2011). [17] de Rochefort et al. MRM 63(1):194-206 (2010). [18] Liu J et al. Neuroimage 2012;59(3):2560-8. [19]
Reichenbach JR et al. Radiology 204:272-277 (1997). [20] Haacke EM et al MRM 52:612-618 (2004). [21] Tropres et al. MRM 45:397-408
(2001).Acknowledgements: The authors are grateful to Pascal Spincemaille from Cornell University for his helpful direction. Supported in part by the
National Institute of Health (NIH 1R01NS066506, NIH 2RO1NS047607, NCRR 5P41RR09784).
Fig. 1: Plural contrasts generated from one 3D ME-
GRE sequence, both pre- and post ferumoxytol.
Difference images are shown in the far right column.
Fig. 2: Regular
TOF-MRA (scan
time = 5 mins).

6755
Faster pediatric MRI
, Stefan Skare2
, Kristen Yeom3
1
Lucas Center for Imaging, Department of Radiology, Stanford University, Palo Alto, CA, United States, 2
Clinical Neuroscience, Karolinksa Institute, Stockholm, Sweden, 3
Lucile
Packard Children's Hospital, Department of Radiology, Stanford University, Palo Alto, CA, United States
Target Audience: Researchers and clinicians who are interested faster pediatric MRI.
Introduction: MRI provides excellent contrast between the different soft tissues of the body, which makes it especially useful in imaging
brain pathology. The main disadvantage of MRI is that is lengthy, taking up to one hour at our institution. Particularly in a pediatric setting,
long protocols increases the risk of motion artifacts in the acquired images, thus GA is often used – at the expense of patient throughput,
comfort, and cost. With the goal of shortening the overall scan time of pediatric MRI, here we present preliminary data acquired with the use
of four MR methods that have been built in-house and that can be retrospectively corrected for patient motion: a T1-weighted 3D Short-Axis
Propeller Echo Planar Imaging (SAP-EPI) sequence1
; a T2-weighted 3D SAP-EPI sequence2
, a novel Fluid Attenuated Inversion Recovery
(FLAIR) sequence using the Readout-Segmented (RS)-EPI trajectory3
; and a dual-echo Diffusion-Weighted-Imaging (DWI) sequence4
. The
first three of these methods are faster than their conventional counterparts; and the latter can also deliver R2 maps for free.
Materials & Methods: 10 pediatric
patients ranging from 1mo to 18 years old
were scanned with the above sequences on
a 3T GE system (Waukesha, WI) and an
8-channel head coil after informed
parental consent and assent were obtained.
The scan parameters were: dual-echo
DWI: FOV = 22cm, TR/TE1/TE2 =
4000/48/115ms, acquisition matrix = 1922
,
in-plane acceleration factor (R) = 3, signal
averages (NEX) = 3, slthck/gap = 4/0mm,
29 slices, 1 b=0, tetrahedral encoding with
b=800 s/mm2
, scan time = 1:12min. T1-w
3D SAP-EPI: matrix size = 1922
, R =
NEX = 3, TR/TE/FA = 46ms/9ms/50º, 64
z-partitions, slthk = 2mm, 7 blades of
width 64, scan time = 1:03min. T2-w 3D
SAP-EPI: matrix size = 2522
, R = NEX =
3, TR/TE/FA = 62ms/22ms/20º, 64 z-
partitions, slthk = 2 mm, 9 blades of width
48, scan time = 1:42min. FLAIR RS-EPI:
matrix size = 1922
, R=NEX=2, TR/TE/TI
= 10000/40/2250ms, slthk/gap = 4/0mm, 5
blades of width 64, scan time = 1:45min.
R2 maps were also calculated using the
dual-echo b=0 and b=800 s/mm2
images.
Results: Figure 1 shows images acquired with the proposed research sequences on a pediatric patient. The patient had prior resection of a
right temporal lobe glioblastoma. Aside from reactive white matter signal abnormality from tumor treatment (arrow), no soft tissue or
diffusion abnormality was seen to suggest residual/recurrent tumor.
Discussion & Conclusion: We have demonstrated preliminary data showing the promise of the use of four sequences that generate some of
the leading contrast mechanisms required for pediatric brain imaging. Together, the four sequences take 5:42min – faster than the sum of their
conventional alternatives scanned at our institution (totaling ~14min). Future work will assess the inherent motion-correction capability of
these sequences. We will then investigate whether these sequences have the diagnostic potential to replace the need for the longer scan
protocol acquired at our institution in the pediatric setting.
References: [1] Holdsworth, SJ et al. 17th
ISMRM, Hawaii, U.S.A 1239 (2009). [2] Holdsworth SJ, et al. 17th
ISMRM, Hawaii, U.S.A 756 (2009). [3] Porter D. 16th
ISMRM, Toronto, Ontario, Canada 3262 (2008). [4] Holdsworth, SJ et al. 20th
ISMRM, Melbourne, Australia, 649 (2012).
Fig. 1 – Proposed sequences acquired on a 4 year old male patient.

2390
Fluid attenuated inversion recovery (FLAIR) with readout-segmented (rs)-EPI
, Stefan Skare2
, Kristen Yeom3
1
Lucas Center for Imaging, Department of Radiology, Stanford University, Palo Alto, CA, United States, 2
Clinical Neuroscience, Karolinksa Institute, Stockholm,
Sweden, 3
Lucile Packard Children's Hospital, Department of Radiology, Stanford University, Palo Alto, CA, United States
Target Audience: Researchers and
clinicians who are interested in a fast FLAIR
sequence. Introduction: The fluid
attenuating inversion recovery (FLAIR) MRI
method
1
is an important technique for the
differentiation of brain lesions. Conventional
FLAIR uses the Fast-Spin-Echo (FSE) method
for image acquisition. However, the
combination of long inversion recovery (TI)
times and the inefficient sampling of the echo
times (TEs) result in prolonged scan times,
posing a challenge for motion-prone patients.
More rapid FLAIR imaging of the brain can be
achieved using the half-Fourier acquisition
single-shot turbo-spin (HASTE)-FLAIR and
Echo-Planar imaging (EPI)-FLAIR sequences
2
.
However these sequences have relatively poor
image quality and reduced ability to show
smaller lesions compared with FLAIR-FSE
2
.
Here, we show preliminary data using a readout-segmented (rs)-EPI
3,4
FLAIR implementation. Rs-EPI has reduced susceptibility-
related artifacts compared with EPI, is faster than FSE sequences, and is relatively robust to motion
5
.
Materials & Methods: All scans were performed on a 3T GE system (Milwaukee, WI, USA; G of 40 mT/m, slew rate of 150 mT/m/s)
using an 8-channel head coil. FLAIR datasets were acquired on a 6-year old pediatric patient after written formal consent was obtained
from the patient’s parents. The rs-EPI-FLAIR pulse sequence and k-space trajectory is shown in Fig. 1. The rs-EPI-FLAIR sequence
imaging parameters were: matrix size = 192
2
, segment width = 64, TE = 48ms, acceleration factor = 2, signal averages = 2, 32 slices,
scan time = 1:45min. Parallel imaging and Nyquist-ghost correction was performed on the center segment, with resulting calibration
parameters applied to each segment
5
. rs-EPI segments were then stitched together using gridding. For reference, a conventional
FLAIR-FSE was acquired (matrix size = 353 x 224, TEeff = 146 ms, 29 slices, scan time = 2:45min) and an EPI-FLAIR (matrix size =
192
2
, acceleration factor = 2, signal averages = 2, TE = 80ms, 32 slices, scan time = 21sec). All sequences used TR/TI = 10s/2.2s, FOV
= 22cm, and slthck = 4mm.
Results: Fig. 2 compares FLAIR-FSE, EPI-
FLAIR, and rs-EPI-FLAIR images. FLAIR-FSE
shows superior contrast with regards to white
matter signal abnormality than EPI and rs-EPI. rs-
EPI shows reduced blurring compared to EPI.
Discussion: This abstract shows that rs-EPI-
FLAIR produces images with improved effective
resolution and reduced blurring compared with
EPI-FLAIR. Compared with FLAIR-FSE, the white
matter contrast of rs-EPI (and EPI) was limited,
likely due to the shorter TE selected. The scan time
came at a 5-fold increase compared with FLAIR-EPI
and a ~1.5-fold decrease compared with FLAIR-
FSE. However, a further ~2-fold reduction in scan time for EPI and rs-EPI sequences can be achieved by using only one signal average.
Next we will deploy rs-EPI-FLAIR together with a navigator echo and test its motion-correction capability on a larger cohort of patients.
Conclusion: Here preliminary data are presented on the clinical application of FLAIR-rs-EPI. While the contrast within the white matter
of this implementation of FLAIR-rs-EPI needs improvement relative to the FLAIR-FSE, it has reduced susceptibility artifacts compared
with EPI-FLAIR. With a better selection of imaging parameters (such as TE), rs-EPI-FLAIR may be a useful rapid and motion-
correctable alternative to conventional FLAIR and EPI-FLAIR in the clinics.
References: [1] De Coene B, et al. AJNR 1;13(6):1555–64 (1992). [2] Filippi M, et al. A. AJR. 20(10):1931–8 (1999). [3] Holdsworth SJ, et al. EJR 65(1):36-46
(2008). [4] Porter DA, Heidemann RM. MRM 62(2):468–75 (2009). [5] Holdsworth SJ, et al. MRM 62:1629–40 (2009).
Figure 2: Comparing the FLAIR FSE, EPI-FLAIR, and rs-EPI-FLAIR images acquired
on a 6-year old patient. Note the right-edge artifact on the rs-EPI is an unexplained issue
with the reconstruction.
Figure 1: (A) Pulse sequence timing diagram for the rs-EPI-FLAIR sequence. The RF
pulses are shown in this order: inversion 180º, spectral-spatial 90º, 180º spin echo, and
refocusing180º. (B) Resulting segmented k-space. The central segment (bold black) is
used for the parallel imaging calibration.

Fig. 2: Comparison of (a) EPI and (b) RS-EPI simultaneous multislice
isoDWI data acquired with fat-suppressive PINS pulses.
6353
Simultaneous Multislice Readout-Segmented Diffusion-Weighted EPI with Blipped-Controlled Aliasing
, Rafael O'Halloran1
, Anh T Van1
, Eric Aboussouan1
, William A Grissom2
, Anuj Sharma2
, Murat Aksoy3
, Julian R Maclaren3
, Stefan Skare4
,
and Roland Bammer3
1
(Equal contribution): Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Palo Alto, CA, United States, 2
(Equal contribution):
Biomedical Engineering, Vanderbilt University, Nashville, Tennessee, United States, 3
Center for Quantitative Neuroimaging, Department of Radiology, Stanford
University, Palo Alto, CA, United States, 4
Clinical Neuroscience, Karolinska Institute, Stockholm, Sweden
TARGET AUDIENCE: For those interested in reducing distortion and reducing scan
time in diffusion-weighted imaging.
INTRODUCTION: Readout-segmented EPI (RS-EPI) is a promising candidate for
reducing distortion in diffusion-weighted (DW)-EPI while being robust to motion-induced
phase errors. However, the requirement of several adjacent segments (or 'blinds') in RS-
EPI can make the scan time can prohibitively long, particularly for thin slices which
require a large number of slices to achieve full brain coverage. A promising approach for
reducing scan time is to use multi-band acquisitions – which make diffusion imaging with
full brain coverage faster and more SNR-efficient by simultaneously acquiring multiple
slices with limited g-factor penalty [1]. In this work, we reduce the minimum TR in RS-
EPI with the use a simultaneous multislice acquisition using PINS multiband pulses [2]
coupled with blipped-controlled aliasing [1].
METHODS: Acquisition – A multiband RS-EPI scan was performed on a healthy
volunteer using a 3T GE system (GE 750) and a 32-channel head coil. Two slices
separated by 8 cm were excited simultaneously using PINS excitation and refocusing
pulses [1]. The PINS pulses are essentially a series of hard pulses spaced by 1/Δ in k-
space, that excite and refocus a set of slices separated by a distance Δ. Here the separation
of 8 cm was designed such that only 2 slices were excited in the z-FOV at once. Inherent
fat suppression was achieved by using positive slice select gradients and negative
refocusing slice gradients such that the bands of fat selected and refocused moved in
opposing directions in z and thus did not intersect (Fig 1). For RS-EPI the navigator echo
was acquired after a second 180° PINS refocusing pulse with negative polarity (Fig 1). Gz-
blips were applied during the imaging and navigator readouts to shift the slices by FOV/2
and better exploit the coil sensitivity [1]. The following scan parameters were used for multiband RS-EPI: Stejskal Tanner diffusion preparation with
xyz encoding (b = 1000 s/mm2
), one b=0, matrix size = 1282
, TR = 4s, 5 blinds of width 64, TE1 /TE2 = 63ms/110ms, FOV = 24cm, slthck/gap = 5
mm/0mm, and a scan time of 1:20min. For the slice-grappa calibration a b=0 scan (of the center blind only) was acquired by replacing the PINS 180°
pulses with standard sinc pulses. EPI multiband (and calibration) data were acquired at the same target resolution. Reconstruction – The center
imaging blind from the b=0 calibration and multiband image underwent FOV/2 ghost parameter estimation (image entropy-based approach) –
followed by application of ghost parameters to all acquired blinds. The
multiband imaging and navigator blinds were then unaliased using the
calibration scan with a slice-grappa approach [1]. For RS-EPI, the imaging
blinds were individually phase corrected (using the navigator blind) and
were gridded together to form the final image.
RESULTS: Fig. 2 shows multiband EPI and RS-EPI DWI data acquired
on a volunteer. As expected, RS-EPI DWI images show better geometric
fidelity than EPI. Note that the simultaneously excited slices appear in the
same image (Fig 2 left column) with one slice aliased to FOV/2 due to the
alternating blips in z. After the slice GRAPPA reconstruction (Fig 2 right
column) both slices appear unaliased effectively providing the
acceleration factor of 2.
DISCUSSION AND CONCLUSION: One of the biggest concerns about
the application of RS-EPI to DWI/DTI is the scan time, particularly for
high-resolution applications which require a large number of slices and
thus a long TR. One can get an extra ~2-fold acceleration of the imaging
sequence with the use of simultaneous multislice approaches, and here we
demonstrate that it is feasible to combine RS-EPI with this approach.
Further scan efficiency was achieved by shifting the fat slice in opposite
directions during the excitation and refocusing steps – removing the need
for an upfront chemical saturation pulse. Further acceleration in RS-EPI
can be achieved with the use of homodyne reconstruction in the kx-
direction [3]. Here we show that one can accelerate RS-EPI DWI scans
with the use of a fat-suppressive blipped-controlled multislice approach. References: [1] K Setsompop, et al. MRM 2012;67:1210–1224. MRM
68:441–451 (2012) [2] DG Norris et al, MRM, 66:1234-1240, 2011. [3] R Frost et al, MRM, 68:441–451 (2012). Acknowledgements: NIH (2R01 EB00271108-A1 ,
5RO1 EB008706, 5R01 EB01165402-02), the Center of Advanced MR Technology at Stanford (P41 EB015891), Lucas Foundation, Oak Foundation.
Fig. 1: Pulse timing diagram of the multi-band RS-EPI
trajectory. The multiband data is acquired using a PINS
90° and 180° pulse which selects two slices
simultaneously. In RS-EPI each imaging blind is
accompanied by a navigator blind in order to perform a
phase correction between diffusion-weighted blinds. Note
that the calibration scan used to perform the slice-grappa
reconstruction is acquired using a pins 90° and standard
180° sinc pulse which selects one slice only.

Fig 3: 69yr patient with vasospasm. The R2 map brings out the hypointense signal
associated with infarct (open arrow) as well as areas of hyperintense signal
correlating with intraventricular hemorrhage(arrowhead) and basal ganglia
calcification (closed arrow).
Fig. 2: Dual echo DWI and ADC images of a
66yr old embolic stroke patient.
Fig. 1: The dual-echo EPI sequence, with GRAPPA factor = interleaves = 3, FOV =
24cm, matrix = 1922
, slthk/gap = 5mm/1.5mm, 24 partial Fourier overscans, TE1/TE2
= 48/105ms, 1b0/4directions, b=1000s/mm2
, TR=3s, scan time 2:30min. With the set
of imaging parameters used here, this approach does not increase the scan time
compared to the single echo alternative since Echo 2 fills in the sequence dead-time.
6053
Dual-echo diffusion-weighted EPI for better sensitivity to acute stroke
, Stefan Skare2
, Kristen Yeom1
, Michael U. Antonucci1
, Jalal B Andre3
, Jarrett Rosenberg1
, Matus Straka1
, Nancy J Fischbein1
, Greg
Zaharchuk1
, and Roland Bammer1
1
Department of Radiology, Stanford University, Palo Alto, CA, United States, 2
Karolinska Institute, Clinical Neuroscience, Stockholm, Sweden, 3
Department of
Radiology, University of Washington, Seattle, WA, United States
Target audience: Practitioners interested in improving
diffusion lesion conspicuity in the setting of acute stroke.
Introduction: Many diffusion-restricting lesions also have
a prolonged T2 value compared to the surrounding tissue.
We hypothesize that one may improve lesion conspicuity in
acute stroke patients with the use of a longer TE than in
conventional practice by means of an accelerated dual-echo
diffusion-weighted (DW)-EPI approach (Fig. 1). Echo 1
provides a high SNR image used to calculate the apparent
diffusion coefficient (ADC), while Echo 2 can be used for
enhanced conspicuity. Furthermore, relaxivity (R2) maps
can be calculated from the dual echo images to potentially
reveal an additional source of image contrast. This study
investigated the applicability of such a dual-echo sequence
in the setting of acute stroke.
Methods:
Dual-echo
DWI data were acquired on 50 patients suspected of stroke using a 1.5T GE scanner
and 8-ch head coil. Three radiologists reviewed the echoes using the routine vendor-
supplied DWI as a reference. Images were graded on lesion conspicuity and diagnostic
confidence on the following Likert scale: 1–nondiagnostic, 2–poor, 3–acceptable, 4–
standard, 5–above average, 6–very good, 7–outstanding. R2 maps calculated from the
two echoes were evaluated for potential complementary information. Tests for
differences in ratings between Echo 1 and Echo 2 were done with a two-tailed
Wilcoxon signed-rank test.
Results: Echo 2 was unanimously favored over Echo 1 for the evaluation of acute
infarcts. Lesion conspicuity and diagnostic confidence were rated better for Echo 2 over
Echo 1 (mean values of 6.5/4.9 and 5.9/5.4, respectively p<0.0001). 72 more lesions
were found on Echo 2 across 34 patients diagnosed with acute stroke than on Echo
1. 93% of these were deemed as acute infarct on ADC, 4% were too small to assess, and
3% were non-restricting chronic lesions. Echo 2 was predicted to have changed the
overall radiological impression in 20% of cases; and to have impacted stroke workup in
16% of cases, and potentially influenced 32% of cases. As shown in Fig. 3, while the
DWI of Echo 2 has higher lesion sensitivity, the ADC of Echo
1 is the best candidate for confirming acute lesions. Echo 2
was also favored for ruling out stroke from regions of
heightened coil sensitivity (closed arrows). The R2 maps were
also useful for detecting ischemic infarct, subarachnoid
hemorrhage and basal ganglia calcification (Fig. 3).
Discussion: Longer TEs than those typically used can
increase the diagnostic sensitivity of DWI. Given that the
DWI from Echo 2 was more useful for lesion delineation and
detection, we recommend that the TE should be exploited to
draw attention to lesions, and that the accelerated dual-echo
EPI DWI approach is a good candidate.
Conclusion: Contradicting the common teaching to use short echo times to avoid T2-shine through, the long TE of Echo 2 gives rise
to DW images with superior conspicuity of diffusion lesions compared to DW images acquired at a shorter TE or conventional T2-
weighted imaging alone: Echo 1 provides high SNR ADC maps for specificity in acute stroke, and the information from both echoes is
a potential source of complementary information for the assessment of blood and mineralization products. In conclusion, using the
minimum TE to achieve maximum SNR and avoid T2-shine through may result in increased identification of stroke-related lesions on
DWI, and a dual-echo approach should be considered when protocoling DWI scans in stroke patients. Acknowledgements: NIH (2R01
EB00271108-A1 , 5RO1 EB008706, 5R01 EB01165402-02), the Center of Advanced MR Technology at Stanford (P41 EB015891), Lucas Foundation, Oak Foundation.

Fig. 2 - Patient data acquired on an 88yr old
male stroke patient. (a) Coil sensitivity-
corrected isotropic DWI (b=1000s/mm2
)
dual echo images showing increased lesion
conspicuity on the second echo (white
arrows). (b) Single slice from the same
dataset showing the isotropic ADC
calculated from the b=0 and DWI images
from echo 1 (top) and echo 2 (bottom),
respectively. The isoDWI echo 2 may be
useful for lesion conspicuity, while isoADC
echo 1 could be used for its higher SNR.
GRAPPA-accelerated dual-echo diffusion-weighted EPI with intensity correction
, Stefan Skare2
, Matus Straka1,3
, Manabu Inoue3
1
Clinical Neuroscience, Karolinska Institute, Stockholm, Sweden, 3
Stanford Stroke Center,
Stanford University Medical Center, Stanford University, Palo Alto, CA, United States
Introduction: A well-known problem with standard single-shot (ss)-EPI for diffusion-weighted imaging (DWI) is
geometric distortion and blurring. These artifacts can be reduced through the use of parallel imaging. The shortened
readout also affords a shorter echo time - which reduces T2-shine through, but may result in reduced lesion
conspicuity on DWI. Thus, in an attempt to both improve the image quality and retain the sensitivity of lesions
through longer echo times, here we implement a GRAPPA-accelerated1,2
, dual echo3
EPI DWI sequence on stroke
patients and present our preliminary data. For cases of absent coil sensitivity calibration scans, we also demonstrate
how the two echoes can be utilized to remove array coil-induced intensity modulations, which are often confused as
potential ischemically challenged regions.
Methods: Patient data was collected on 15 patients suspected of stroke using a 1.5T GE system and an 8-channel
head coil. The following scan parameters were used: Stejskal Tanner diffusion preparation with tetrahedral
encoding (b = 1000s/mm2
) and one b=0, matrix size = 192 x 192, GRAPPA acceleration factor R = 3 and 3
interleaves (used for the ghost- and GRAPPA-weights estimation as well as to boost SNR), NEX = 2, TR = 3s,
TE1/TE2 = 51ms/115ms, FOV = 24cm, slthck/gap = 5 mm/1.5mm, and a scan time of 2:15min. The first and second
echo were corrected for signal intensity variation across the images using the following:
for 1,2 Eq. 1
where Dci represents each corrected echo image, Di are the original echo images, CF is the Gaussian-filtered
contribution of, C, given by:
∑ where log log Eq. 2
and C represents contributions to the signal intensity from coil sensitivity, proton density, original magnetization,
and RF; and for the two echoes: t = TE, I are the b=0 images, D are the DWI images.
Results: Fig. 1-2 shows GRAPPA-accelerated dual-echo EPI patient data acquired on two patients with strokes of
the middle cerebral artery (MCA). The increased lesion conspicuity of echo 2 is apparent. Fig. 2b shows coil
sensitivity-corrected maps, which remove the hyperintensity of the signal - particularly in the posterior regions
where cortical regions are closer to individual coil elements. Fig. 2 shows patient data where the second echo on
isoDWI provided increased diagnostic confidence. Fig. 2b shows isotropic ADC maps calculated from echo 1 and 2,
suggesting that the first echo should always be used for its higher SNR.
Discussion: Here we show that GRAPPA-accelerated DWI can be made even more applicable in a clinical setting
with the acquisition of a second echo in the same TR. While the first echo can be used for high SNR ADC maps, the
second echo can be used for lesion detection (Fig. 1-2). The use of the second echo for lesion detection goes against
the common teaching that TE should be kept short to reduce T2-shine through. While this might be true for
differentiating between acute and subacute lesions, for general lesion conspicuity longer echo times may yield
greater diagnostic confidence. Interestingly, on long-TE DWIs, the combined effect of diffusion restriction and
prolonged T2 gives additional contrast that is not necessarily observed on long TE T2-w FSE scans or FLAIR alone.
The added information from quantitative ADC and T2 will further the ability to differentiate between acute and
subacute lesions. In addition, the two echoes can be used to remove the contribution of coil sensitivity which can
lead to misdiagnosis (since hyperintensity around the cortex on DWI can indicate pathology such as stroke,
encephalitis, Creutzfeldt-Jakob disease, and epilepsy). This bias field removal may also improve automated
segmentation procedures such as used for diffusion-perfusion-mismatch calculation4
. One could argue that a calibration scan could be acquired for coil-
sensitivity correction - however this comes at an additional scan time cost. With the set of imaging parameters that we routinely use for our stroke protocol at
our institution, the dual-echo DW-EPI approach does not increase the scan time since the second echo fills in the dead-time of the sequence.
Conclusion: Here we show that by observing DWI images acquired at two echo times, complementary information can be gleaned, at no additional scan time
cost. In addition, the two echoes can be used to remove the coil sensitivity contribution to the DWI images, which may provide clinical confidence and will
improve the performance of automated segmentation procedures used to suggest treatment outcome for stroke patients.
References: [1] Griswold, M. et al. MRM 2002;47:1202-1210. [2] Qu, P. et al. JMR 2005;174(1):60-67. [3] Feinberg, D. et al. MRM 1994:31:461. [4] Straka, M. et al. JMRI 2010;32:1024. Acknowledgements: This work
was supported in part by the NIH (5R01EB002711, 5R01EB008706, 3R01EB008706, 5R01EB006526, 5R21EB006860, 2P41RR009784), the Center of Advanced MR Technology at Stanford (P41RR09784), Lucas
Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4). An extra special thanks to Patricia Lassus, Murat Aksoy, and Nancy Fischbein for their insights.
Fig. 1: Data collected on a 91 yr old male stroke
patient (a) Original dual echo images, and (b) the
dual echo images with contribution from coil
sensitivity, proton density, initial magnetization
and B1 removed (as in (c)). The corrected images
rule out any suspicion of posterior lesions (white
arrows). Note the increased contrast of the stroke
lesions on echo 2. (d) Relaxivity maps calculated
from the average contribution of the T2-w and
DWI dual-echo images.
649Proc. Intl. Soc. Mag. Reson. Med. 20 (2012)

Comparison between EPI and RS-EPI at high acceleration factors
, Anh T Van1
, Stefan Skare2
1
'Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Palo Alto, CA, United States, 2
Clinical Neuroscience, Karolinska Institute,
Stockholm, Sweden
Introduction: Readout-Segmented EPI (RS-EPI) [1] has been proposed in a number of studies (including several of our own)
as a variant of EPI to reduce distortion in diffusion-weighted (DW) neuroimaging [1-3]. RS-EPI segments k-space into
individual EPI ‘blinds’ along the readout direction, and requires an additional navigator blind (for phase correction of the off-
center DW blinds). With the increasing number of coils coming available with advanced phased-array technology that increases
the capacity of EPI to achieve higher acceleration factors (Fig. 1), and advanced distortion-correction methodology (such as the
Reversed Gradient Polarity method (RGPM) [4,5]), we suspect that the utility of RS-EPI is perhaps more tailored to ultra-high
resolution DWI. An initial qualitative demonstration of this is shown here, with EPI and RS-EPI data acquired with the highest
acceptable GRAPPA [6,7]-acceleration factor that we have been using with our 32-channel coil.
Methods: Experiments were
conducted on a 3T GE system
(Gmax=50mT/m, Smax=200 mT/m/ms)
and a 32-channel head coil channel
(Nova Medical, Wilmington, MA,
USA). In order to find a maximum
GRAPPA acceleration factor, R,
achievable with our set-up, scan-time
matched EPI DWI datasets were
acquired first on both a QA agar
phantom and a human volunteer using
acceleration factors of R = 1-6. With
the selection of R = 5 as an acceptable
choice of acceleration factor for EPI
(Fig. 2), a distortion-matched RS-EPI
dataset was acquired using R=2, as well as R=5 dataset. The following
common parameters for EPI and RS-EPI were: twice-refocused diffusion
preparation with x,y,z diffusion encoding, b = 1000s/mm2
, one b = 0, a target
resolution of 240 x 240, slice thickness = 5mm, TE/TR = 70ms/4000ms, FOV
= 24 cm, partial Fourier encoding with 24 overscans. RS-EPI used 5 blinds of
width 54. All scans were scan time matched (by using a larger NEX for EPI).
For each dataset, a b = 0 image was acquired with an opposite phase-encoding
gradient to enable subsequent distortion correction using the RPGM method.
Results: A comparison between the distortion- and scan-time matched EPI (R = 5) and RS-EPI (R = 2) datasets are shown in Fig. 3a-b. Despite the longer readout for
EPI, the extent of image blurring appears to be similar between the two datasets thanks to the higher acceleration factor and hence larger step through phase-encoding k-
space. However, because RS-EPI only uses the central (54 x 240) segment of k-space to estimate the GRAPPA weights, there appear to be residual GRAPPA noise in
some slices (white arrows, Fig. 3b). As one moves to the highest acceleration factor capable with our system (achievable with EPI), the RS-EPI datasets are slightly less
distorted, but become significantly noisier. Furthermore, since RS-EPI is essentially a multi-shot technique, the quality of
the diffusion-weighted images relies on the robustness of the phase error correction. At very high acceleration factor, the
low SNR of images reconstructed from individual blinds, especially the edge blinds, might negatively impact the phase
correction procedure, resulting in poor quality images. Fig. 4 compares the original b = 0 images with those corrected for
distortion using the RPGM method.
Discussion & Conclusion: With the advent of more advanced phased-array head coils, an increasing maximum slew rate
achievable by gradient coils, the image quality of standard EPI is becoming much improved. For standard clinical imaging
EPI is likely to be the method of choice for its speed and high SNR, while RS-EPI may still have application to ultra-high
resolution diffusion imaging at high field strengths [3] where scan time is not as critical. As shown in Fig. 4, with
acceleration factors of R = 5 achievable with EPI, the residual distortion provides a good starting point for distortion
correction methods (such as RGPM). Furthermore, the gain that one gets using RS-EPI with R = 5 in terms of reduced
distortion is rather small and, in addition, much poorer image quality results due to the smaller central strip used for the
GRAPPA estimation. In fact, as this abstract has
made apparent, a more robust GRAPPA
estimation procedure may be required (such as the
use of adjacent blinds), as GRAPPA noise in the
RS-EPI images is visible with R = 2, and
considerable with R = 5. A separate calibration
scan could remedy this, but would have an
additional scan time cost (in addition to the extra
blinds required to fill k-space). RS-EPI has a
number of additional practical limitations,
including gaps in k-space (between blinds) in the
presence of large motion; eddy current effects
between blinds (due to the different kx-dephasing
gradient) on poorly calibrated systems; and the
requirement for an additional navigator reduces
the number of slices that can be acquired per TR.
References - [1] Porter D, MRM, 62:468, 2009. [2] Holdsworth, S
et al. MRM, 62:1629, 2009 . [3] Heidemann et al. MRM, 64:9-14, 2010. [4] Andersson JL. Neuroimage 2003;20(2):870-888.
[5] Skare S ISMRM 2010. [6] Griswold MA. et al. MRM 2002;47:1202-1210. [7] Qu P. et al. JMR 2005;174(1):60-67.
Fig. 1 - EPI and RS-EPI k-space
traversal. A schematic representation
of the k-space trajectory for EPI and
RS-EPI for (a) smaller acceleration
factors, and (b) larger acceleration
factors achievable through the use of
coils. The utility of RS-EPI is
significantly reduced with increasing R
since it results in a smaller scan time.
Fig. 3 - Isotropic DWI images acquired using (a) EPI with R = 5, (b) RS-EPI
with R = 2 (distortion-matched with (a)), and (c) RS-EPI with R = 5. The white
arrows depict regions where GRAPPA noise affects the image quality in RS-
EPI. The red arrows depict regions of slightly reduced distortion achievable
with RS-EPI R = 5.
Fig. 2 - DWI datasets acquired at 3T
using a 32-channel head coil. (a)
GRAPPA fit error calculated for R =
2-6 b=0 datasets on a QA agar
phantom. From R = 5-6 the
GRAPPA fit error begins to increase
considerably. (b) b=0 images
acquired on a volunteer using no
acceleration (R=1) and with
acceleration. The R = 2-5 images are
reconstructed with GRAPPA. The R
= 5 image retains good image quality
and thus is the acceleration factor we
chose for our EPI versus RS-EPI
comparison.
Fig. 4 - Comparison of original and distortion-
corrected b = 0 images acquired with EPI R =
2, RS-EPI R=2, and RS-EPI R=5. The
acquisition matrix = 240x240.
4212Proc. Intl. Soc. Mag. Reson. Med. 20 (2012)

6619
Fast Susceptibility Weighted Imaging (SWI) using Readout-Segmented (RS)-EPI
S. J. Holdsworth1
, R. O'Halloran1
, S. Skare2
, and R. Bammer1
1
Introduction: Susceptibility-weighted imaging (SWI) is an MRI technique that has been used to provide
improved conspicuity of venous blood vessels and other sources of susceptibility effects [1,2]. The most
commonly used SWI acquisition uses a 3D gradient echo (GRE) sequence, however due to the inefficient
coverage of k-space per TR, 3D GRE suffers from a long scan time. In addition, even subtle motion in 3D
GRE can considerably hamper the quality of final processed SWI image. A 3D EPI trajectory has been used
as a faster alternative for SWI [3]. Unless 3D EPI is acquired with multiple interleaves (which consequently
would also make it prone to motion artifacts), the images can suffer from significant blurring and geometric
distortion. 3D Short-Axis Propeller (SAP)-EPI [4] has been proposed as a method to minimize distortion and
blurring compared to 3D EPI for SWI, whilst keeping the scan time at a level appropriate for regular clinical
use. A disadvantage of 3D SAP-EPI is the blurring that results from the combination of several EPI segments
acquired at different angle. Readout-Segmented (RS)-EPI [5] is a similar technique to SAP-EPI, except that
rather than acquiring several overlapping angular segments, adjacent overlapping segments are acquired.
Since the segments are all acquired at the same angle, this results in reduced blurring in the final gridded
image. Here we implement 3D RS-EPI as an alternative to 3D GRE and 3D SAP-EPI for SWI.
Methods: The 3D RS-EPI k-space trajectory is shown in Fig. 1. Experiments were conducted on a healthy
volunteer using a 3T GE system and an 8-ch head coil. The following scan parameters were used: matrix size
= 288 x 288, 7 blinds of width 64, R = 3, NEX = 3, TR/TE/FA = 55ms/20ms/15º, FOV = 23 x 23 x 12.8cm3
,
64 z-partitions, a 2 mm slice thickness, brick frame rate = 3.5s, and a scan time of 1:15min. For comparison a
3D SAP-EPI scan was acquired with equivalent scan parameters (1:15min scan time). A high resolution flow-compensated 3D GRE sequence was acquired
for comparison using a matrix size = 512 x 256, rectangular FOV = 0.75, TR/TE/FA = 37ms/20ms/20º, z-partitions = 32, 2mm slice thickness, and a scan time
= 4:40mins. All SWI images were produced by generating a phase mask using a 2D Hanning window for each individual coil, a multiplication of the phase
mask by the magnitude coil by 5 times, followed by the sum of squares over coils. A minimum intensity projection (MinIP) was then taken over a 14mm thick
stack of partitions.
Results: A comparison between the SWI images acquired with
3D GRE and 3D RS-EPI is shown in Fig. 2. Although the
resolution and SNR is highest for the 3D GRE, the 3D RS-EPI
images show that it is possible to acquire good quality SWI
images with twice the brain coverage in about a third of the scan
time. Fig. 3 shows a comparison between the SWI MinIPs for
both 3D RS-EPI and 3D SAP-EPI acquired at the same target
resolution and scan time. Because of the unidirectional distortion
in 3D RS-EPI, it demonstrates better vessel conspicuity than 3D
SAP-EPI.
Discussion & Conclusion: Here we have shown that 3D RS-
EPI is a fast alternative technique to 3D GRE for SWI (Fig. 2).
The acquisition of 3D RS-EPI SWI images in 1:15min makes
this technique applicable for routine use in the clinics. Parallel
imaging with R = 3 was used in order to minimize distortion,
whilst keeping each blind consistent (i.e. inter-blind motion is
negligible). One can also reduce the scan time for 3D GRE for
SWI through the use of parallel imaging, but our experiments
have shown that this can come at a significant detriment to the
SNR (data not shown). In addition, this study demonstrated
superior resolution and conspicuity when using 3D RS-EPI over
3D SAP-EPI (Fig. 3).
Even subtle motion can be a major problem for standard SWI
imaging, as it can corrupt the phase image used for the SWI
processing. Due to the consistency of data within each brick, both 3D RS-EPI and 3D SAP-EPI image
phase are significantly less sensitive to motion compared with both GRE and interleaved EPI.
Furthermore, the small temporal footprint of the bricks (3.5s) for 3D RS-EPI makes it much easier than
GRE to catch moderate motion and reacquire data if necessary. In addition, for jerky movement, as long
as the brick frame rate in 3D RS-EPI is fast enough and there is enough overlap between bricks (to
account for rotations of individual bricks), it may be possible to correct for motion between bricks in k-
space in 3D. Future work will explore this claim. Further studies are warranted to perform a detailed
comparative assessment of both 3D RS-EPI and 3D SAP-EPI methodologies to assess whether the
(already demonstrated) motion-robust self-navigated 3D SAP-EPI is truly needed when the better
resolved 3D RS-EPI can handle motion to an acceptable level as demonstrated in this study. Overall, the
minimum gain in small vessel conspicuity of 3D GRE does not appear to justify the almost 4-fold
longer scan time at half the coverage in cervico-caudal direction.
References: [1] Reichenbach JR. Radiology 1997;204:272-77. [2] Hacke EM. MRM 2004;52:612-18. [3] Patel MR. Stroke 1996;27:2321–2324.
[4] Holdsworth SJ. ISMRM 2009; 756. [5] Porter D. ISMRM 2008;3262. Acknowledgements: NIH (5R01EB002711, 5R01EB008706,
3R01EB008706, 5R01EB006526, 5R21EB006860, 2P41RR009784), the Center of Advanced MR Technology at Stanford (P41RR09784), Lucas
Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4).
Fig. 1: 3D RS-EPI k-space trajectory. One segment
(or blind) is acquired for every z-partition – resulting
in one ‘brick’. Multiple adjacent overlapping bricks
are acquired (overlap not shown for clarity).
Fig. 3: Two slices from a 3D GRE and 3D RS-EPI sequence acquired with a matrix size of 512 x 256 and
288 x 288, respectively, and a partition thickness of 2mm and a FOV of 23cm. A 14mm thick MinIP is also
shown. As depicted schematically on the far right, 3D RS-EPI acquired full brain coverage (twice the
coverage compared with 3D GRE) in a significantly reduced scan time.
Fig. 3: (a) 3D RS-EPI, (b) 3D SAP-EPI SWI MinIP images
acquired with the same scan time (1:15min). RS-EPI shows
less blurring and better vessel conspicuity than SAP-EPI.

6571
Comparison of two alternative approaches for diffusion-weighted Readout-Segmented (RS)-EPI
S. J. Holdsworth1
, S. Skare2
, M. Aksoy1
, R. O'Halloran1
, and R. Bammer1
1
Introduction: Readout-segmented (RS)-EPI is a technique that has been used for reducing distortion
in diffusion-weighted imaging (DWI) [1], whereby multiple EPI segments are used to fill up k-space
(Fig. 1). Two alternatives have been described for filling k-space in RS-EPI: one recently
demonstrated to reduce scan time by filling k-space with full readout segments (Fig. 1a) [2], and the
other which minimizes the echo time (TE) through the use of partial readout segments (Fig. 1b) [3].
In first example, only three segments are acquired (requiring a full echo), with partial Fourier
reconstruction performed in the kx direction (labeled ‘RS-EPI-X’). In the second example, five
partial readout segments are used to fill up k-space, with partial Fourier encoding performed in ky
(labeled ‘RS-EPI-Y’).
Here we explore whether one gains from reducing the number of segments at the expense of a longer
TE (RS-EPI-X), or from having a larger number of segments at the minimum TE (RS-EPI-Y). We
performed an SNR comparison for the two approaches, comparing the SNR for both single- [4] and
twice refocused [5] diffusion preparation, the latter which is commonly used to reduce the effects of
eddy currents in diffusion imaging.
Materials & Methods: GRAPPA-accelerated DW RS-EPI-X and RS-EPI-Y images were acquired
on a healthy volunteer on a 1.5T whole-body GE system using an 8-channel head coil. Both single-
refocused (1 x 180) and twice- refocused (2 x 180) DW preparation schemes were tested, together with two matrix sizes: (192 x 192, blind width = 48, number of
overscans = 24) and 288 x 288 (blind width = 64, number of overscans = 32). The TE and maximum number of slices are reported in Table 1. Other parameters were:
acceleration factor R = 3, NEX = 3, slice thickness = 5 mm, FOV = 24 cm, b = 1000 s/mm2
, and TR = 3s. For SNR measurements, noise maps generated from three
repeated b = 0 scans, and the SNR normalized for scan time efficiency ( / scan time/sliceη = SNR ) was calculated over 20 slices. For the post-processing stage, the
ghost calibration [6] and GRAPPA weights [7-8] were calculated
from the central segment of first b = 0 scan and applied to each
volume, followed by ramp sampling correction, POCS reconstruction
[9-10], and sum-of-squares over coils.
Results: Table 1 shows the normalized SNR ratio ( Yη / Xη ).There
was no significant difference in normalized SNR for all cases except for the matrix size of 288 x 288 with the 2 x 180 diffusion preparation scheme. Human brain data
showing scan time matched b = 0 and isotropic DW images for the matrix size of 288 x 288 and two diffusion preparation schemes are shown in Fig. 2. The effect of
the SNR reduction for RS-EPI-X for 2 x 180 is particularly pronounced in isotropic DWI images. Note that the longer TE of the RS-EPI-X scheme resulted in b = 0
images with greater T2-weighting.
Discussion: Here we explored the SNR difference between two
alternative RS-EPI techniques. While fewer segments are required for
RS-EPI-X, the combination of the longer echo train and resulting
reduction in the number of slices that can be acquired per TR reduces
its scan time efficiency more than originally expected. The scan time
efficiency was similar for all but one case tested here: the case of RS-
EPI-Y with a 2 x 180 and 288 x 288 matrix size was a factor of 1.6
times more efficient – clearly evident in the isotropic DW images of
Fig. 2. This is due to the excessively long TE brought about by the
longer echo train for 2 x 180 RS-EPI-X. While it is reasonable to
assume that the need to only acquire 3 segments (instead of 5) in RS-
EPI-X (in this particular example) implies that one can scan faster, for
most clinical purposes one needs to average several times to keep the
SNR for DW RS-EPI at a reasonable level, so the scan time
advantage of RS-EPI-X can be misleading. At our institution we use 2
x 180 DW preparation for the acquisition of reliable FA maps, and
often acquire these with matrix sizes of 288 x 288 – thus RS-EPI-Y
would be the best choice of implementation. Furthermore, as one
moves to higher matrix sizes one can expect that RS-EPI-Y will
become even more efficient, as the TE does not increase with matrix
size.
References: [1] Porter D et al. ISMRM 2008;3262. [2] Frost R et al. ISMRM 2010; 1625. [3] Holdsworth SJ et al. ISMRM 2008;4. [4] Stejskal EO. J. Chem. Phys. 1965;43(10):3597–3603.
[5] Reese TG et al. MRM 2003;49:177–82. [6] Nordell A et al. ISMRM 2007:1833. [7] Griswold M et al. MRM 2002;47:1202-1210. [8] Qu P. et al JMR 2005;174(1):60-67. [9] Haacke EM et
al. JMR 1991;92:126–45.[10] Liang ZP et. al. MRM 1992;4:67–185.
Acknowledgements: This work was supported in part by the NIH (5R01EB002711, 5R01EB008706, 3R01EB008706, 5R01EB006526, 5R21EB006860, 2P41RR009784), the Center of
Advanced MR Technology at Stanford (P41RR09784), Lucas Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4).
Blind width x
final resolution
RS-EPI-X
(1 x 180 / 2 x 180)
RS-EPI-Y
(1 x 180 / 2 x 180)
/Y Xη η
(1 x 180 / 2 x 180)TE (ms) Max. # slices TE (ms) Max. # slices
48 x 192 67/90 24/20 55/70 29/25 1.0/1.1
64 x 288 83/128 17/13 55/70 24/20 1.1/1.6
Table 1: Echo time (TE) and maximum number of slices achievable in a TR = 3s for RS-EPI-X and RS-EPI-Y.
The SNR ratio (RS-EPI-Y/RS-EPI-X) is also shown.
Fig. 1. The two alternative approaches for RS-EPI (a) fully sampled
segments in the ky direction, i.e. a full echo acquired with partial Fourier
encoding in kx (b) partially sampled segments in the ky direction, i.e.
minimizing the TE, which therefore requires partial Fourier
reconstruction in the ky direction.
Fig. 2. Scan-time matched RS-EPI-X and RS-EPI-Y b = 0 and isotropic DW images of a healthy
volunteer, acquired with 1 x 180 and 2 x 180 at a matrix size of 288 x 288.

6512
Diffusion Weighted Imaging of Spinal Tumors with Reduced Field of View EPI
S. J. Holdsworth1
, R. O'Halloran1
, K. Yeom1
, M. Aksoy1
, S. Skare2
, and R. Bammer1
1
Introduction: At our institution, there is a great need for diffusion-weighted (DW) and diffusion-tensor (DT) imaging of the spinal cord for the assessment of cord
lesions, including tumors, cysts, post traumatic cord contusions, as well as congenital anomalies. While DWI can provide insight into the nature of cord lesions,
such as ischemia, infectious or toxic-metabolic processes, neoplasms, and dermoids/epidermoids, the relationship between these lesions and spinal tracts are
difficult to define. DTI could further probe biology of the underlying lesion, elucidate integrity and location of the spinal tracts, reveal otherwise occult lesions, and
potentially aid in surgical navigation. However, due to the geometric distortion that arises from the slow phase-encoding bandwidth in echo-planar imaging (EPI),
our radiologists have largely given up on spinal cord diffusion imaging using conventional single-shot DW-EPI. The zonally oblique multislice (ZOOM)-EPI
technique [1-3] is an approach which uses a tilted refocusing pulse to reduce the phase-encoding FOV, thus reducing geometric distortion and image blurring. Here
we demonstrate the feasibility of ZOOM-EPI for assessing spinal tumors at 1.5T and 3T. To demonstrate the potential for further improvements in this technique
we also present distortion-corrected images acquired on a healthy volunteer at 3T.
Methods: DTI datasets were acquired on a 1.5T and 3T whole-
body GE system. Both magnets used a 4-channel spine coil.
ZOOM-EPI DTI datasets were acquired on two patients: a 5 year
old male with a suspected thoracic cord neoplasm (3T, FOV =
22cm x 5.5cm, matrix size = 224 x 56) and a 35 year old female
with cervicomedullary neoplasm (1.5T, FOV = 18cm x 4.5cm,
matrix size = 192 x 48). Imaging parameters common to both
patients were: TR = 3s, seven slices with 3mm thickness and no
gap, zoom-angle θ = 10º, TE/TR = 73ms/3s, partial Fourier
encoding (18 overscans), 5 b = 0, 35 isotropically distributed DW
directions with b = 500 s/mm2
, and a scan time of 2mins. Both
patients underwent surgical resection of the tumor, and the tumor
grade was evaluated histopathologically. In addition, volunteer
cervical and thoracic DTI data were acquired with the same
parameters as above (3T, FOV = 22cm x 5.5cm, matrix size = 224
x 56), except that only four b = 0 images were acquired, and two
of these were acquired with an opposite phase-encoding gradient
to enable subsequent distortion correction using the reversed
gradient polarity (RPGM) method [4].
Results: The ZOOM-EPI b = 0 s/mm2
, isotropic DWI, color fractional anisotropy (FA), and tractography images for the two patients
are shown in Figs.1-2. Fig. 1 shows a well-encapsulated tumor, which was found surgically to lack local tumor infiltration and also
was confirmed pathologically to be a low-grade tumor (grade I pilocytic astrocytoma). Fig. 2 shows a more aggressive neoplasm with
an infiltrative tumor biology where neoplastic cells infiltrate and traverse alongside spinal tracts. At surgery, the lesion was adherent
and difficult to completely excise due to its infiltrative nature; pathologic assessment revealed infiltrating grade II astroctyoma,
confirming our pre-operative diagnostic suspicion based on ZOOM-EPI data. Fig. 3 demonstrates distortion-corrected ZOOM-EPI
data in a volunteer, showing that it is possible to make further improvements to ZOOM-EPI with post-processing.
Discussion: This work shows the application of ZOOM-EPI in a useful clinical setting. Based on the utility of these first patient data
acquired at our initiation, ZOOM-EPI DTI data enhanced diagnostic capacity for pathologic tumor grade, and further defined the
relationship between spinal tracts and the underlying lesion, an important potential application in surgical management.
There was initially some reservation about the multi-slice capability of ZOOM-EPI given that the tilted slice approach may saturate
neighboring slices (and therefore significantly reduce the SNR). However, since the useful anatomical region typically resides in the
center of the slice, it is possible to get away with small zoom-angles (10º), and thus reduced saturation effects. Our volunteer
experiments have shown that multi-slice ZOOM-EPI only comes at a very small expense of SNR. Other reduced FOV approaches
such as spatial-spectral selection [5] do not suffer from this saturation effect, however these approaches have a limited number of
slices that can be acquired.
At present, we have limited the FOV to a maximum of 18cm at 3T in order to keep
distortions at a reasonable level, but we are encouraged by volunteer data that larger
rectangular FOV can be achieved in combination with distortion correction (with the use
of an oppositely-acquired b = 0 image). Future work will explore a reasonable set of
imaging parameters one should use clinically in combination with distortion correction.
Summary: This work demonstrates that spinal cord DTI can be extremely useful for the
clinical work up of patients, and that ZOOM-EPI is a useful acquisition method to keep
distortions within a reasonable limit. Both patient DTI data acquired at 1.5T and 3T
revealed important information. Future work will explore whether additional post-
processing to reduce distortion can aid the clinical workup of patients.
References: [1] Mansfield P. Phys. E: Sci. Instrum. 1988 21:275. [2] Symms M. ISMRM 2000;160. [3]
Wheeler-Kingshott, MRM 2002 47:24. [4] Andersson JL. Neuroimage 2003;20(2):870-888. [5] Saritas E. MRM
2008;60:468. Acknowledgements: This work was supported in part by the NIH (5R01EB002711,
5R01EB008706, 3R01EB008706, 5R01EB006526, 5R21EB006860, 2P41RR009784), the Center of Advanced
MR Technology at Stanford (P41RR09784), Lucas Foundation, Oak Foundation, and the Swedish Research
Council (K2007-53P-20322-01-4).
Figure 1: ZOOM-EPI images of a 35 year old patient with a low grade
tumor. (a) b = 0, (b) isoDWI, (c) color FA, (d-e) tractography showing
that the tumor is well-encapsulated. This was confirmed surgically.
Figure 2: ZOOM-EPI images of a 5 year old pediatric patient with an
aggressive neoplasm. (a) b = 0, (b) isoDWI, (c) color FA, (d-e) tractography
depicting infiltrating tracts that were consistent with surgical findings.
Figure 3: Thoracic b = 0
images (NEX = 2) (top
row) No distortion
correction. (bottom row)
distortion-corrected
images produced by
using the displacement
field calculated from the
positive (+ve) and
negative (-ve) images.
The far right is the
average of the positive
and negative images.

Reduced-FOV Diffusion Imaging with ZOnal Oblique Multislice (ZOOM) combined with Readout-Segmented (RS)-EPI
S. J. Holdsworth1
, S. Skare1
, R. L. O'Hallaran1
, and R. Bammer1
1
Radiology, Stanford University, Palo Alto, CA, United States
Introduction: Diffusion-weighted imaging (DWI) using echo-planar imaging
(EPI) has been limited by geometric distortion and blurring, particularly in
regions with large off-resonance effects such as in the spinal-cord and in regions
of the brain residing near tissue/air interfaces. Geometric distortion in EPI is
proportional to the FOV in the phase encoding direction (FOVpe), as well as the
echo-spacing between adjacent echoes in the EPI train (Tro). Parallel imaging,
which is frequently used for distortion reduction in EPI scans, is difficult to use
for certain geometric arrangements of coils (such as some spine array coils) and
for small FOVs (such as in orbital scans). To reduce FOVpe and avoid aliasing,
saturation pulses can be used, however the suppression efficiency is generally
limited leading to partial aliasing. Spatially selective RF pulses [1,2] can be used,
but have limitations in the number of slices that can be prescribed. The zonal
oblique multislice EPI (ZOOM-EPI) technique [3-5] is another approach, which
uses a tilted refocusing pulse as shown in Fig. 1. To reduce distortion further, Tro
can be reduced by covering k-space with a series of consecutive segments or
‘blinds’, known as RS-EPI [6,7] (Fig. 2c). In this work, we implement both the
ZOOM pulse together with the RS-EPI trajectory (ZOOM-RSEPI) to get the
benefits of both methods for reducing distortion. Fig. 2 illustrates the theoretical
reduction in distortion as one goes from a full FOV EPI, to ZOOM-EPI, and finally to ZOOM-RSEPI.
Methods: All diffusion images were acquired on a 3T whole-body GE DVMR750 system. ZOOM-EPI and ZOOM-RSEPI DW images were first acquired on a
phantom (8-channel head coil, single-shot, FOV = 26cm x 7.8cm, matrix size = 180 x 54 (square pixels)). Second, thoracic spine DTI images were acquired on a
volunteer using a 4-channel spine coil, followed by a scan of the orbits using a single-channel birdcage coil. The following common parameters were used:
rectangular FOV = 30 x 10cm, Δz = 4 mm (zoom-angle θ = 5º, slthck90º = 4mm, slthck180º = 8mm), TR = 3s, a matrix size = 200 x 60 (square pixels), partial Fourier
(18 overscans), 24 isotropically distributed DW directions with b = 500 s/mm2
, and a scan time of 9:48mins. ZOOM-RSEPI used TEmin = 55 ms, 7 blinds of width
= 32 and ZOOM-EPI used TEmin = 75 ms, and 7 NEX. For comparison, full FOV images were acquired on the phantom and volunteer. For the orbital scan, the
parameters as above were used with Δz = 5 mm, b = 600 s/mm2
, 60 diffusion directions and 4 b = 0 (ZOOM-EPI), 8 diffusion directions with 1 b = 0 (ZOOM-
RSEPI), and a scan time of 3:12min.
Results: The EPI and RS-EPI b = 0 s/mm2
phantom images acquired
with and without the use of rectangular FOV and ZOOM pulse is
shown in Fig. 3. As shown, geometric distortion can be reduced
significantly with ZOOM-RSEPI. This method also reduces the
‘jagged’ appearance of the spinal cord as shown in the b = 0 s/mm2
images (Fig. 4). Fig. 5 compares a ZOOM-EPI and ZOOM-RSEPI
scan of the orbits. The eyes, the optic nerve, and the ocular muscles
show reduced distortion with ZOOM-RSEPI.
Discussion: This work shows that RS-EPI in combination with the
ZOOM pulse in order to spatially select a region of interest may be useful for diffusion imaging of regions with large off-resonance effects. While the rectangular
FOV used in this work (30cm x 10cm) reduced distortion by 30% (compared to a full FOV acquisition), RS-EPI further reduced the distortion by 33%. A
disadvantage of RS-EPI is the reduced SNR efficiency compared with EPI. However the resulting RS-EPI DTI data shown in Fig. 4 shows high SNR 3T images
acquired in a reasonable scan time for DTI, and the jagged appearance that often hampers the quality of EPI DW images is reduced significantly. The orbital scan
in Fig. 5 more accurately depicts the shape of the optical nerve, thus may be a useful method for DTI or fiber-tracking of the optical nerve for the early diagnosis of
pathology such as multiple sclerosis.
References: [1] Kiefer C. ESMRMB 1999;302 [2] Saritas E. MRM 2008;60:468. [3] Mansfield P. Phys. E: Sci. Instrum. 1988 21:275. [4] Symms M. ISMRM 2000;160. [5] Wheeler-Kingshott, MRM 2002 47:24 [6] Porter D.
ISMRM 2004;442. [7] Holdsworth S. ISMRM 2009:6247. Acknowledgements: This work was supported in part by the NIH (1R01 EB008706, 1R01 EB008706S1, 5R01 EB002711, 1R01 EB006526, 1R21 EB006860), the
Center of Advanced MR Technology at Stanford (P41RR09784), Lucas Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4).
Figure 1: The zoom pulse.
Following a 90º slice-selective
pulse, a 180º pulse is applied
obliquely at an angle θ. The
resulting exciting parallelogram is
the desired rectangular volume [3].
Figure 2: Reduction of distortion as one moves from (a) full-FOV EPI to (b) a rectangular FOV
with EPI (zoom-EPI), and (c) a rectangular FOV with RS-EPI (ZOOM-RSEPI). For the parameters
used in this work, this 'distortion meter' accompanying the trajectories is drawn to scale.
Figure 5: Orbital
isoDWI images (b =
600 s/mm2
) acquired
using ZOOM-EPI (60
diffusion directions)
and ZOOM-RSEPI (8
diffusion directions, 7
blinds), both in a
scan time of 3:12min.
Figure 3: Single-shot EPI and single-shot RS-
EPI images (b = 0s/mm2
) acquired on a
phantom using using: (Left column) a full FOV
of 26 cm and matrix size = 180 x 180; (Right
column) a rectangular FOV = 26 x 7.8 cm and
matrix size = 180 x 54 (square pixels).
Figure 4: Comparison between the b = 0 s/mm2
images of a thoracic spine using full FOV EPI (30 x 30 cm, matrix size = 200 x 200), ZOOM-
EPI and ZOOM-RSEPI (30 x 10 cm, matrix size = 200 x 60 (square pixels)). For ZOOM-RSEPI, the isotropic DWI (isoDWI, b = 500 s/mm2
),
fractional anisotropy (FA), and 1st
eigenvector (colormap) are also shown. Note that there is less ‘disc bulging’ into the spinal canal and less
blurring on the ZOOM-RSEPI scans than on the ZOOM-EPI scans.

Figure 3: Comparison between the ss-EPI DWI and
ss-RS-EPI DWI 1min sequence acquired at 3T on a
pediatric patient under suspended respiration.
Imaging parameters were: FOV = 28cm, Δz = 5mm,
matrix-size = 128x128, TR = 2s, two b = 500s/mm2
(A/P direction), 7 blinds of width 32 (RS-EPI), and 7
NEX (EPI).
Diffusion-Weighted Imaging of the Abdomen with Readout-Segmented (RS)-EPI
S. J. Holdsworth1
, S. Skare1
, S. S. Vasanawala1
, and R. Bammer1
1
Introduction:
Diffusion-weighted imaging (DWI) in the abdomen has proven useful for various
pathologies, including liver lesion characterization [1-4] and simple vessel
suppression, diagnosis of diffuse renal disease [5-8], and detection of metastatic
spread to lymph nodes [9,10]. However, image distortions arising from the use of
EPI has shown to be problematic. We have recently applied DW ‘Short-Axis
Propeller’ (SAP)-EPI to the abdomen on adults to reduce geometric distortions via its
faster k-space traversal [11]. In this work we explore the use of another short-axis
readout technique, Readout-Segmented (RS)-EPI [12], for imaging the abdomen. As
shown in Figure 1, the use of several adjacent segments in RS-EPI results in reduced
distortion compared with EPI.
Materials & Methods:
Breath-hold single-shot (ss)-EPI and ss-RS-EPI diffusion-weighted images
were acquired on an adult volunteer using a 3T whole-body GE DVMR750
system using an 8-channel cardiac-array coil. Both sequences used a matrix
size of 192 x 192, FOV = 34cm, TE = minimum (RS-EPI: 56 ms, EPI: 72
ms), partial Fourier imaging (24 overscans), slthck/gap = 8 mm/1.5mm, TR =
2s, one b = 500 s/mm2
(S/I direction), in a scan time of 30sec. Seven blinds
of size 64 x 192 (freq.×phase) were used for RS-EPI, and 7 NEX were used
for EPI to keep the scan time equivalent. By using RS-EPI over EPI, the
distortion was reduced by 50% (due to the bandwidth increase in the phase-
encode direction). Both sequences were then also acquired on a 6-month old
pediatric patient under general aesthesia, after obtaining IRB approval and
consent from the patient's parent. Imaging parameters (as different from
above) were as follows: matrix size of 128 x 128, FOV = 28cm, slthck/gap =
5 mm/0mm, b = 500 s/mm2
(A/P direction, applied twice), 7 blinds of size 32
x 128 (RS-EPI), and a total scan time of 1 min. In this case, the distortion
reduction was 45%. The reconstruction of the RS-EPI data was performed as
in Ref. [13], with one exception: the triangular window used for phase
correction [14] was increased to the full k-space radius in order to reduce
phase errors (and address the larger extent of motion that occurs in body
imaging).
Results:
A comparison between the b = 0 s/mm2
and isotropic b = 500 s/mm2
EPI and RS-EPI images of the abdomen for the volunteer and patient is
shown in Figs. 2 and 3. At an equivalent matrix size and scan time, RS-EPI appears sharper and less distorted, at the expense of a lower SNR.
Discussion & Conclusion:
While EPI-based DWI of the abdomen has proven useful for the diagnosis of various
pathologies, image distortions arising from off-resonance effects (especially in the presence
of bowel gas) and large FOVs can significantly hamper the image quality. This work shows
that RS-EPI can be useful for DWI of the abdomen by reducing geometric distortion and
blurring (as shown in Figs. 2-3). Disadvantages of RS-EPI are the increased scan time
compared with EPI – which is tied to the extra number of blinds required to cover k-space –
as well as the increased risk of phase-artifacts that can occur between blinds. Further
experiments will explore these effects under free-breathing and respiratory triggering.
References:
[1] Namimoto T, Radiology 1997;204:739. [2] Kim T, AJR Am J Roentgenol 1999;173:393. [3]
Ichikawa T, Abdom Imaging 1999;24:456. [4] Taouli B, Radiology 2003;226:71. [5] Ries M, JMRI
2001;14:42. [6] Chan JH, Clin Imaging 2001;25:110. [7] Namimoto T, JMRI 1999;9:832. [8] Fukuda Y,
JMRI 2000;11:156. [9] Takahara T, Radiat Med 2004;22:275. [10] Koh D, AJR Am J Roentgenol
2007;188:1622. [11] Skare S, ESMRMB 2009:49. [12] Porter D. ISMRM 2004;442. [13] Holdsworth S,
MRM 2009:62 Early view. [14] Pipe J. MRM 1999;42(5):963.
Acknowledgements:
This work was supported in part by the NIH (1R01EB008706, 1R01EB008706S1, 5R01EB002711,
1R01EB006526, 1R21EB006860), the Center of Advanced MR Technology at Stanford (P41RR09784),
Lucas Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4).
Figure 1: K-space traversal of
RS-EPI [12]. K-space is
acquired with a series of
adjacent EPI segments or
‘blinds’. Note that each blind is
accompanied by an extra
central segment (or navigator
blind) in order to perform
phase correction on the DW
blinds.
Figure 2: Comparison between ss-EPI and ss-RS-EPI DWI 30sec
breath-hold images on a volunteer acquired at 3T. Imaging parameters
were: FOV = 34cm, Δz = 8mm, matrix-size = 192 x 192, TR = 2s, one
b = 500s/mm2
(S/I direction), TEmin = 72ms/56ms (EPI/RS-EPI), 7
blinds of width 64 (RS-EPI), and 7 NEX (EPI).

Clinical Application of Readout-Segmented (RS)-EPI for Diffusion-Weighted Imaging in Pediatric Brain
S. J. Holdsworth1
, K. Yeom1
, S. Skare1
, P. D. Barnes1
, and R. Bammer1
1
Introduction: Readout-segmented (RS)-EPI [1] has been suggested as an alternative approach to EPI for high
resolution diffusion-weighted imaging (DWI) with reduced distortions. Here we implemented GRAPPA-
accelerated RS-EPI DWI on 35 pediatric patients at 3T. We compared these images with standard accelerated
(ASSET) EPI DWI used routinely for clinical studies at our pediatric hospital.
Methods: RS-EPI and EPI images were acquired on 35 pediatric patients using a 3T whole-body system (GE
DVMR750) and an 8-channel head coil. The following parameters were used: FOV=20cm, slthck=4mm, TR=3s,
one b=0 and three diffusion directions with b=1000 s/mm2
(xyz encoding). RS-EPI used a twice-refocused
diffusion preparation with a matrix size = 1922
, 7 segments (width=64, overlap factor=57%), acceleration factor
R=3, NEX=3, and a scan time of 4:12min. GRAPPA and ghost calibration were performed on the multi-shot
data, thus no separate calibration scan was acquired. Data were reconstructed as described elsewhere [2]. The
routine ASSET-accelerated EPI sequence used for comparison: matrix size=1282
, R=2, one b=0 and three
diffusion directions with b=1000 s/mm2
(xyz encoding), and a scan time of 50s. A pediatric neuroradiologist
evaluated the DW images, scoring them in terms of resolution, distortion level, SNR, lesion conspicuity, and
diagnostic confidence as follows: 1 – non-diagnostic, 2 – poor, 3 – acceptable, 4 – standard, 5 – above average, 6
– good, 7 – outstanding. First the images were scored independently, followed by a reevaluation of the RS-EPI
images with the datasets viewed together. Finally, an overall preference was selected.
Results: Fig. 1 shows the average scores calculated across 35 patients for EPI, RS-EPI, and EPI vs RS-EPI. The
RS-EPI dataset was preferred overall in all except for two patients due to the presence of phase artifacts on RS-
EPI arising from pulsatile brain motion. In 12 patients, the EPI scans suffered from mild-to-severe ‘worm-like’
artifacts also arising from brain motion (though not accounted for when considering the final preference). RS-EPI
identified a lesion not identified by EPI in one patient (small subdural empyema, Fig. 2); more accurately defined
the extent and structure of lesions, such as a cystic encephalomalacia (Fig. 3a) and a clival chordoma (Fig. 3b);
had improved additional lesion localization in one patient (Leigh's disease, Fig. 4); and correctly identified a false
positive lesion seen on EPI on another patient (Moya Moya disease, Fig. 5). RS-EPI also demonstrated exquisite
anatomic detail at the cortical-subcortical levels, brainstem, temporal and inferior frontal lobes, skull base, orbits,
naso-ethmoid, and the cranial nerves – all of which were more difficult to assess on EPI. Overall, the RS-EPI had
significantly improved diagnostic confidence.
Discussion & Conclusion: Averaged over 34 patients, RS-EPI out-performed the product ASSET EPI sequence
(Fig. 1). RS-EPI was chosen as the overall preference for all but two patients due to the presence of phase
artifacts on DWI. Note that these artifacts were later removed by increasing the triangular window used for phase
correction to the full k-space radius (data not shown). In conclusion, RS-EPI may be a useful alternative to EPI
for DWI for evaluating lesions such as hypoxic-ischemic brain injury, diffuse axonal injury, tumors,
dermoid/epidermoid, and skull base/orbital pathology. While some of the imaging parameters of the two
sequences were not identical, this study shows the importance of both resolution and decreased distortions in the
clinics, which can be accomplished by a combination of parallel imaging and alternative k-space trajectories such
as RS-EPI. Aside from SNR, increasing the number of averages for EPI (to match the scan time of the RS-EPI) is
not expected to change the outcome of this study as it is primarily the resolution and distortion improvements
that led to increased lesion conspicuity and diagnostic confidence.
References: [1] Porter D.
ISMRM 2008;3262. [2]
Holdsworth S. MRM 2009:62
(early view).
Acknowledgements: This
work was supported in part by
the NIH (1R01 EB008706,
1R01 EB008706S1, 5R01
EB002711, 1R01 EB006526,
1R21 EB006860), the Center
of Advanced MR Technology
at Stanford (P41RR09784),
Lucas Foundation, Oak
Foundation, and the Swedish
Research Council (K2007-
53P-20322-01-4). Special
thanks to Serman Lim, Alfred
Barikdar, Allan White, Young
Chang, Harold Estrada, Liz
Ellison, and Abbie Bird and
for their assistance with the
patient studies.
Fig 1. Comparison between the routine ASSET-accelerated EPI sequence and our
implementation of RS-EPI in terms of 5 categories averaged over 35 patients.

5835
3D SAP-EPI motion-corrected fast susceptibility weighted imaging
S. J. Holdsworth1
, S. Skare1
, K. Marty1
, M. Straka1
, and R. Bammer1
1
Lucas MRS/I Center, Stanford University, Stanford, CA, United States
Introduction: Susceptibility-weighted imaging (SWI) has been utilized as a useful contrast mechanism in MRI that
accentuates the paramagnetic properties of blood products (1,2). With the use of both magnitude and phase images, SWI
can provide improved conspicuity of venous blood vessels and other sources of susceptibility effects (2). Typically, the
SWI acquisition uses a high-resolution, three-dimensional gradient echo (GRE) sequence. However, the GRE acquisition
used for SWI suffers from a long scan time (~5 mins at 3T), which decreases patient through-put and increases the chances
of motion artifacts. A 3D GRE-EPI trajectory has been proposed as a faster alternative. However, unless the data are
acquired with several interleaves, the images may suffer from considerable blurring and geometric distortion artifacts. The
problem with using multiple interleaves is that, like standard GRE, it makes 3D GRE-EPI vulnerable to motion. Here, a 3D
short-axis readout propeller (SAP)-EPI trajectory (3) is suggested as an alternative approach to 3D GRE and 3D GRE-EPI.
SAP-EPI can achieve higher resolution than EPI with significantly reduced distortions (4). As a result, fewer interleaves
can be used (GRAPPA-acceleration factor R ≤ 4), making the use of parallel imaging (PI) applicable. With PI, each
interleave can be acquired with the full image FOV, making one ‘blade’ consistent (that is, inter-blade motion is
negligible), while the acquisition of one ‘brick’ can be performed with full brain coverage in a few seconds. The speed of
3D SAP-EPI makes the risk of intra-brick motion (ghosting) small – leaving only the inter-brick 3D motion to be corrected
(3). Here we show human 3D SAP-EPI SWI images, as well as demonstrate initial motion-corrected SWI SAP-EPI human
data from controlled motion experiments. 3D SAP-EPI is also compared with interleaved 3D EPI as well as with standard
3D GRE.
Methods: The 3D SAP-EPI k-space trajectory is shown in Fig. 1. Experiments were conducted on a
healthy volunteer using a 3T whole-body GE Excite system and an eight-channel head coil. The
following scan parameters were used for the SAP-EPI and interleaved EPI sequence: matrix size = 256
x 256, TR/TE/FA = 55ms/20ms/20º, FOV = 24 x 24 x 12.8cm3
, 64 z-partitions, slthk = 2 mm. The
SAP-EPI sequence used 8 blades of width 64, R = NEX = 4, a brick frame rate of 3.5 s, and a scan time
of 1:48 min. The EPI sequence used 32 interleaves for an equivalent scan time. Two 3D SAP-EPI
datasets were acquired, the second with a through-plane rotation of ~10º. Blade data were mixed, such
that every second blade was chosen from the rotated dataset. The R bricks per brick angle originating
were 3D motion corrected (in the image domain). All bricks together were then motion corrected prior
to gridding (in k-space). A high resolution flow-compensated GRE sequence was acquired for
comparison (matrix size = 512 x 256, rectangular FOV = 0.75, TR/TE/FA = 37ms/20ms/20º, z-
partitions = 32, slthk = 2mm, scan time = 5mins). All data were processed by generating a phase mask using a 2D Hanning window. The phase mask was multiplied by
the magnitude image 5 times to produce the final SWI image.
Results: A comparison between the original magnitude images
acquired with 3D GRE, interleaved EPI, and SAP-EPI is shown in
Fig. 2. Although the resolution and SNR is highest for the GRE
scans, both the interleaved EPI and SAP-EPI scans demonstrate
darker vessels in a number of regions. In addition, interleaved EPI
and SAP-EPI have a considerably reduced scan time and a better
extent of brain coverage (64 partitions in 1:52mins for interleaved
EPI and SAP-EPI, versus 32 partitions in 5mins for GRE). Although
the scan time of the interleaved EPI scan is equivalent to the SAP-
EPI and the vessels are slightly more conspicuous, it has reduced
SNR and suffers from ghosting artifacts – even for a cooperative
volunteer. Motion artifacts are problematic for both GRE and EPI, as
even small patient motion could result in an unusable image. Fig. 3
shows a side-by-side comparison of GRE and SAP-EPI images.
While small vessels are more easily depicted in the GRE image,
SAP-EPI renders thicker and more prominent larger vessels –
probably due to the additional T2*-dephasing that occurs during the
EPI readout. Also depicted are motion corrupted SAP-EPI SWI data
from the mixed-blade dataset which have been corrected for motion.
This example demonstrates an additional advantage of SAP-EPI over
GRE and interleaved EPI.
Discussion: Perhaps one of the greatest hindrances to the adoption of SWI in the clinics is the long scan time associated with standard GRE. However, with parallel
imaging, the scan time of GRE can be reduced by up to a factor of 4 (8-channel head coil) – although this can result in a significant SNR penalty. EPI can be used to
significantly speed up the acquisition, but with an SNR which makes it more compatible with PI. However, PI-enhanced EPI still suffers from distortion artifacts. On
the other hand, multi-shot EPI used without PI (R > 4), like standard GRE can suffer from severe ghosting in the presence of motion. SAP-EPI helps to alleviate some of
these problems. The ‘short-axis’ blades with a smaller echo-spacing (compared to EPI) result in reduced distortion (roughly proportional to the width of the blade). As a
result, fewer interleaves can be used – which enable the use of PI (here R = 4) to give a consistent motion-free blade (that is, a blade acquired at full FOV) in a reduced
overall scan time compared with GRE. In addition, any motion that does occur between blades can be corrected for via 3D rigid-body correction. In the event that the
brick frame rate (of 3.5 s in this case) results in inter-brick motion, this brick could be re-acquired (3).
Summary: Here we have presented SWI images acquired with an efficient propeller-based EPI readout method, which has an inherent ability to allow motion
correction. While 3D SAP-EPI still suffers from some geometric distortion, its significantly shorter scan time and relatively high SNR suggest that 3D SAP-EPI may be
a useful alternative to GRE for use in susceptibility-weighted imaging, particularly in uncooperative patients. References: [1] Reichenbach JR et al. Radiology 1997;204:272-77. [2] Hacke
EM et al. MRM 2004;52:612-18. [3] Holdsworth SJ et al. ISMRM 2008:1352. [4] Skare S et al. MRM 2006;55:1298-1307. [5] Nordell A et al. ISMRM 2007:1833. [6] Griswold MA et al. MRM 2002;47:1202-1210. [7] Qu P
et al. JMR 2005;174(1):60-67. [8] Skare S et al. MRM 2007;57:881–890. Acknowledgements: This work was supported in part by the NIH (2R01EB002711, 1R01EB008706, 1R21EB006860), the Center of Advanced MR
Technology at Stanford (P41RR09784), Lucas Foundation, Oak Foundation, and the Swedish Research Council (K2007-53P-20322-01-4). We would also like to thank Bronwen Holdsworth for her continuous help.
Fig. 2: 3D GRE, 3D interleaved EPI (32 shots), and 3D SAP-EPI (R =
NEX = 4, 8 blades) original magnitude images.
Fig. 3: (A) 3D GRE, (B) 3D SAP-EPI SWI (top row) and SWI minIP images (bottom row). Also included in (B) are
motionless, motion-corrupted, and 3D motion corrected SAP-EPI images.
Fig. 1: 3D SAP-EPI k-space traversal (3). One
blade is acquired for every z-partition –
resulting in one ‘brick’, which is then rotated.
Referenceless Nyquist-ghost calibration
parameters (5) are estimated on the middle slice
in hybrid-space to correct for each brick,
followed by a 1D FFT along kz. GRAPPA
weights (6,7) are estimated on each brick (8).

6397
T1-weighted 3D SAP-EPI for use in pediatric imaging without general anesthesia
S. J. Holdsworth1
, S. Skare1
, K. Yeom1
, and R. Bammer1
1
Introduction: 3D Short-Axis readout Propeller EPI (SAP-EPI) was recently introduced as an alternative to the
3D T1-w Spoiled Gradient Echo (SPGR) technique commonly used for routine clinical studies [1]. In the
presence of motion, the data consistency for encoding in 3D-SPGR is violated and gives rise to ghosting
artifacts. In 3D SAP-EPI, the acquisition of ‘bricks’ (each consisting of an EPI blade that goes through the
center of k-space) gives full brain coverage with a frame rate of ~3s (Fig. 1). Since each brick can be acquired at
full FOV, the option of performing 3D motion and distortion correction between bricks is available. 3D SAP-
EPI can have particular impact on pediatric imaging, since gross patient motion is difficult to avoid, and as a
consequence, many patients are sedated or anesthetized. Whilst motion-compensated methods are available for
T2w/FLAIR/DWI, a T1w method is not currently available and would be tremendously useful. In this abstract,
we demonstrate the utility of 3D SAP-EPI applied to a moving pediatric patient.
Methods: The 3D SAP-EPI k-space trajectory
is shown in Fig. 1. Images were acquired on a
1.5T GE Excite (Waukesha, WI, USA:
50mT/m, 150mT/m/s) and an 8-channel head coil. First of all, to verify the utility of the
sequence controlled motion experiments were performed on a healthy adult volunteer. 3D
SAP-EPI data were acquired with the following scan parameters: FOV = 25 cm3
; a matrix
size of 192 x 192 x 128; a voxel size of 1.3 x 1.3 x 2.0 mm3
, 8 blades of size 48(kRO) x
192(kPE) x 128(kPEz), a GRAPPA-acceleration factor R = 3; NEX = 3; full Fourier;
TR/TE/FA = 55ms/17ms/60º; a brick frame rate of 7 s. The scan time was 2:48mins (one
would expect a ~3mins for a conventional 3D fast SPGR of a similar scan prescription). The
volunteer was asked to move in the through- and in-plane direction and the scan was then
repeated. Blades from the first and second dataset were then mixed. After obtaining IRB
approval and consent from the patient’s parents, 3D SAP-EPI were acquired on a 6yr old
male with an optic glioma. Imaging parameters were: FOV = 24 x 24 x 12.8 cm3
; a matrix
size of 192 x 192 x 64; and a voxel size of 1.3 x 1.3 x 2 mm3
. Four blades of size 64 x 192 x
64 were acquired with a 180º sweep (thus the edges of k-space were slightly undersampled);
a brick frame rate of 3.5s; and a total scan time of 2:07mins. Three repetitions of each blade
angle were made, in order to increase the chance of acquiring a brick without inter-brick
motion. The brick with the best GRAPPA fit and Nyquist ghost parameters was used in the final reconstruction. For the post-processing stage, the blades
underwent referenceless Nyquist-ghost correction [2], and GRAPPA weights [3-5] estimation and application on a per brick basis. 3D motion and distortion
correction was then applied using the combination of all blades for estimating the ΔB0 field [6], followed by gridding of the blades together [7].
Results: Data obtained from the mixed-blade dataset is shown in Fig. 2. The 3D motion- and distortion-corrected image shown in Fig. 2c shows the successful
correction of the motion-corrupted data in Fig. 2b. Patient data are shown in Fig. 3. The top row shows the routine T1-w fast SPGR image corrupted by
motion. The rows below this show motion corrupted 3D SAP-EPI data which also have been corrected for motion, as well as combined motion and distortion
correction. As indicated by the white arrow, a double image is evident in the posterior region
of the motion corrected image, due to residual distortion. A marked improvement can be
observed in this area for the combined 3D distortion and motion corrected images.
Discussion & Conclusion: In our experience, approximately 20% of pediatric patients must
either be rescanned or sedated, due to severely motion corrupted images. 3D SAP-EPI has
yielded images with high grey-white matter contrast and data can be acquired in a similar
scan time than fast 3D SPGR. Together with its motion correction capability, 3D SAP-EPI
could be a useful alternative to fast 2D and 3D SPGR routinely used for pediatric brain
imaging. Fig. 3 is an example of a successfully corrected dataset acquired on a 6yr old
moving patient. By collecting several bricks per blade angle, one can discard the corrupted
bricks in the event of substantial intra-volume motion – and 3D motion correction can then
be performed using the remaining bricks. While the brick frame rate of 3.5s cannot rule out
intra-volume motion, the use of 4 blades and 3 repetitions of the acquisition enabled two of
the bricks that were corrupted by inter-brick motion to be discarded. To increase the brick
frame rate, future work would be to implement GRAPPA in the z-direction also, or to use a
multi-slab approach. Deciding which bricks to discard in a non-supervised manner could be
investigated with the use of k-space entropy or by the GRAPPA fit error. Here, the use of
thin SAP-EPI blades combined with parallel imaging has allowed reduced distortions. Any
residual distortion can be partly corrected for, using 3D distortion correction with the
combination of all blades/bricks, without the penalty of extra calibration time. Future work
would be to acquire images at higher field strengths to test the distortion correction method,
and to put forward a good method for automatic brick elimination based on motion.
References: [1] Holdsworth SJ et al. ISMRM 2008:1352. [2] Nordell A et al. ISMRM 2007:1833. [3] Griswold MA et
al. MRM 2002;47:1202-1210. [4] Qu P et al. JMR 2005;174(1):60-67. [5] Skare S et al. MRM 2007;57:881–890. [6]
Skare S et al. ISMRM 2008:417. [7] Jackson JI. IEEE Trans Med Imag 1991;10:473-478. Acknowledgements: This
work was supported in part by the NIH (2R01EB002711, 1R01EB008706, 1R21EB006860), the Center of
Advanced MR Technology at Stanford (P41RR09784), Lucas Foundation, and the Swedish Research Council
(K2007-53P-20322-01-4). We would like to thank Bronwen Holdsworth, Tom Brosnan, Allan White, Serman Lim, Michael Beers, Alfred Barikdar, Young Chang, and Liz Ellison for their assistance.
Figure 2. Human brain scans from a controlled motion 3D SAP-EPI
experiment.
Figure 3. Human brain scans acquired with fast 3D SPGR (top row) and
3D SAP-EPI (remaining rows). Parameters were: a matrix size of 192 x
192, blade width = 64, 4 blades selected from 3 repetitions, R = NEX = 3,
TR/TE = 56 ms/17ms, a FOV = 24 x 24 x 12.8 cm
3
, 64 partitions, a voxel
size of 1.3 x 1.3 x 2 mm, and scan time = 2:07mins.
Figure 1. Traversal of k-space in 3D SAP-EPI
[1]. Bricks are swept by 180º in the x-y plane.

6761
On the application of phase correction and use of k-space entropy in partial Fourier diffusion-weighted EPI
S. J. Holdsworth1
, S. Skare1
, and R. Bammer1
1
Lucas MRS/I Center, Stanford University, Stanford, CA, United States
Introduction: It is well-known that diffusion-weighted (DW) imaging is very sensitive to the effects of brain motion, even in single-shot (ss)-EPI [1-4]. While the extent of rigid body motion can be
minimized through patient compliance and by securing the patient’s head, pulsatile brain motion is ubiquitous and can be significant. Pulsatile brain motion that occurs during the application of
the DW gradients can result in the dispersion of k-space, corresponding to signal dropout and shading in the image domain. Severe brain motion may yield a k-space completely corrupted by
brain motion [4].
Typically, partial Fourier (PF) encoding in the phase-encoding direction is used to reduce the echo time in DW-ssEPI. Here, the number of ‘overscans’ is used to denote how many extra lines of
k-space are acquired past the k-space center. If k-space is dispersed in the case of pulsatile brain motion, the number of overscans acquired may not be enough to encode some of the dispersed
signal and considerable information may be lost. In addition, the lack of phase information provided by the small central strip of k-space used for PF reconstruction may result in artifacts in the
final image. This abstract shows that phase correction applied prior to partial Fourier reconstruction in ss-EPI is helpful for recovering signal lost in cases where k-space is corrupted by brain
motion.
Using the same k-space from several repetitions of a DW-ssEPI scheme, we
explore the use of k-space entropy [5] as a metric to identify k-space corrupted
by non-linear brain motion; the use of peripheral cardiac gating and non-gating;
phase correction applied before both homodyne and POCS reconstruction; as
well as the number of overscans that should be used to avoid significant artifacts
due to pulsatile brain motion.
Materials & Methods: A healthy volunteer was scanned on a 3T whole-body GE
EXCITE system (Waukesha, WI, USA, 40 mT/m, SLR = 150 mT/m/s) with an 8-
channel head coil. Data were acquired by repeating an EPI diffusion scheme 150
times along the S/I direction (the direction most sensitive to pulsatile motion [6]).
This scheme was repeated both with and without peripheral cardiac gating. A
target resolution of 128 x 128 was used, a TR = 3 s (or 3 RR intervals and
minimum trigger delay for the gated acquisition), R = 3, b = 1000 s/mm2
, and 21
slices with a thickness of 5mm. Full Fourier data were acquired – and this data was used to test various overscans. The entropy of k-space was used to determine the correlation between
entropy and the extent of non-linear motion due to brain pulsation. Corrupted k-space identified by the entropy measure was then reconstructed using the following number of overscans: 8, 16,
and 32. Each dataset were phase corrected using a triangular windowing approach [7] modified for partial Fourier data (Fig. 1). The phase correction approach corrects for any low spatially
varying linear and non-linear motion with the use of a low resolution phase-map extracted from the center strip of k-space [7]. This approach was applied before both POCS [9] and homodyne
[10] reconstruction. The images were compared with the same data reconstructed without phase correction.
Results: Fig. 2 shows a plot of the k-space entropy for 150 DW-EPI repetitions calculated from one slice
acquired at the base of the brain. As shown, non-linear motion causes a substantial dispersion of k-space
data – which is paralleled with an increase in the entropy of k-space, and correspondingly large signal voids
in the center of the image. While peripherally-gated DW-ssEPI sequences are robust against pulsatile brain
motion, non-gated sequences yield corrupted k-space with 15% prevalence for slices located at the base of
the brain (where pulsatile motion is greatest). Fig. 3 shows slices with high- and low- k-space entropy (taken
from the highest and lowest peak in Fig. 2, respectively). Both datasets are reconstructed with POCS and
homodyne (using 8 overscans), as well as reconstructed without (top row) and with (bottom row) phase
correction. While there is little difference in image quality between the two types of PF reconstruction
methods in the low k-space entropy case, the dataset with high k-space entropy yields significant artifacts in
the image domain. For the latter, both homodyne and POCS yield a large signal void in the brain stem, with
additional ‘worm-like’ artifacts for the homodyne reconstruction. For both reconstruction techniques, it is
clear that performing the phase correction approach before Partial-Fourier reconstruction recovers
significant signal in the brain stem for both PF methods, and has fewer worm-like artifacts for the homodyne
reconstruction.
Important to note is the utility of using a larger number of overscans for avoiding brain motion artifacts [4], as
shown in Fig. 4. Here, the same dataset is used, except k-space is trimmed to 8, 16, and 32 overscans,
respectively, prior to PF reconstruction. In the case of severe non-linear motion, the increasing benefit of
more overscans is evident. In addition, even in the case of 16 overscans, phase correction prior to
homodyne reconstruction recovers signal in the brain stem. However, as the number of overscans
approaches 32, the benefit of performing phase correction before PF is not clear.
Discussion: This work demonstrates k-space entropy as a robust metric to identify data corrupted by motion.
Out of 150 repetitions of a diffusion scheme, ~15% of slices in the base of the brain (b = 1000 s/mm2
, S/I
direction) revealed elevated k-space entropy which closely correlated with the extent of signal dropout and
image artifacts in the image domain. The entropy metric tended to be deterministic – either low in the case of no
motion, or considerably elevated (see Fig. 2).
Perhaps the most important message is the following: to help to avoid artifacts due to brain motion, it is useful
to perform phase correction on PF data before PF reconstruction. It was shown that POCS out-performs
homodyne reconstruction for data corrupted by non-linear motion. Both methods help to recover lost signal,
however POCS results in fewer worm-like
artifacts. For both methods, consistent
with [4], a larger overscan factor will yield
a more robust estimation of the image
phase and fewer motion-related image
artifacts. Used in conjunction with phase
correction prior to PF reconstruction, we
recommend using a minimum number of
overscans of 16 to help avoid these image
artifacts. Whilst the cases shown in the
figures are the most extreme cases of
non-linear motion we saw in our experiments, in some cases motion may be so severe (particularly when there is significant
motion in the through-plane direction) that k-space may have to be reacquired.
References: [1] Norris DG. JMRI 2001;13:486–495. [2] Wedeen VJ. MRM 1994;32:116-120. [3] Butts K. MRM 1996;35:763-770. [4] Storey P.
MRM;57(3):614-619. [5] Shannon CE. Weaver W. Uni. Illinois Press; 1963. [6] Wirestam R. JMRI 1996;6(2):348-355. [7] Pipe J. MRM 2002;47(1):42-52. [8]
Holdsworth SJ. ISMRM 2008;4. [9] Liang ZP. Rev MRM 1992;4:67-185. [10] Noll DC. IEEE Trans. Med. Imag. 1991;10(2):154. Acknowledgements: This
work was supported in part by the NIH (2R01EB002711, 1R01EB008706, 1R21EB006860), the Center of Advanced MR Technology at Stanford
(P41RR09784), Lucas Foundation, and the Swedish Research Council (K2007-53P-20322-01-4).
Figure 2: Plot showing the entropy of k-space calculated for 150 DW-ssEPI acquired in the S/I
diffusion-encoding direction (b = 1000 s/mm2
, black = gated; blue = not gated). The red line
shows the threshold above which k-space is significantly dispersed by non-linear motion (high k-
space entropy), causing large signal dropouts in the image domain. The threshold (red line) was
determined by using the mean of the entropy over the 150 repetitions + one standard deviation,
and indicates the line above which these corrupted blinds have severe signal dropouts and shading
in the image domain.
Figure 3: DW-EPI datasets with a) low k-space entropy, and b) high k-space entropy. A
matrix in-plane resolution of 128 x 128 was used, slthck = 5mm, b = 1000 s/mm2
(S/I
direction). Both datasets were acquired with PF in the p/e-direction, using 8 overscans and
both homodyne and POCS reconstruction. The data were also reconstructed both without
(top row) and with (bottom row) phase correction. The long red arrows indicate the ‘worm-
like’ artifacts prevalent in homodyne-reconstructed datasets corrupted by brain motion, as
well as the severe signal dropouts in the brainstem in both PF methods. The small arrows
indicate areas of significant improvement in the image quality due to the phase correction.
Figure 4: DW-EPI (b = 1000 s/mm2
, S/I direction) datasets acquired at an in-
plane resolution of 128 x 128 showing: A) Low k-space entropy for reference
(full Fourier data). B) Severely motion corrupted data reconstructed with
various overscans both without phase correction (top row), as well as with
phase correction performed prior to homodyne reconstruction. The red arrows
indicate areas with significantly improved image quality.
Figure 1: Triangular phase correction process [7] applied to the partial Fourier diffusion-weighted ssEPI data. Note
that this process is similar to [8], however is faster with the use of zerofilling (rather than POCS).

ISMRM_2006-2015_compressed

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (18)

Similar to ISMRM_2006-2015_compressed

Similar to ISMRM_2006-2015_compressed (20)

ISMRM_2006-2015_compressed