Gwendolyn Van Steenkiste1, Ben Jeurissen1, Steven Baete2,3, Arnold J den Dekker1,4, Dirk H.J. Poot5,6, Fernando Boada2,3, and Jan Sijbers1
1iMinds-Vision Lab, University of Antwerp, Antwerp, Belgium, 2Center for Advanced Imaging Innovation and Research (CAI2R), NYU School of Medicine, New York, NY, United States, 3Center for Biomedical Imaging, Department of Radiology, NYU School of Medicine, New York, NY, United States, 4Delft Center for Systems and Control, Delft University of Technology, Delft, Netherlands, 5Imaging Science and Technology, Delft University of Technology, Delft, Netherlands, 6Biomedical Imaging Group Rotterdam, Erasmus Medical Center Rotterdam, Rotterdam, Rotterdam, Netherlands
Synopsis
Achieving
a high spatial resolution with DTI is challenging due to the inherent trade-off
between resolution, acquisition time and signal-to-noise ratio (SNR). We propose a strategy to improve this trade-off by
combining super-resolution DTI (SR-DTI) and simultaneous multi-slice (SMS)
acquisition. With SMS-SR-DTI, high resolution DTI
parameters can be recovered from thick slice images which have a high SNR. By
acquiring the images with SMS, the overall acquisition time remains clinically
feasible. As such, high resolution in vivo DTI becomes feasible in a clinical
setting. This opens up exciting possibilities for diffusion MRI research.Introduction
High resolution in vivo diffusion tensor imaging (DTI)
within clinically feasible scan times is a challenging task. Multiple diffusion
weighted (DW) images are needed to estimate the diffusion parameters and
inherently, DW images have a low signal-to-noise ratio (SNR). Improving the SNR
of the DW images by increasing the number of averages would require an
infeasible long acquisition time. As a result, to ensure sufficient SNR, DW
images are often acquired with a low spatial resolution, commonly 2x2x2mm
3
- 3x3x3mm
3, leading to large partial volume effects. One way
to improve the trade-off between the spatial resolution, acquisition time and
SNR is acquiring the data with an MRI scanner with stronger gradients (Human
Connectome Project (HCP) scanner [1]) which allows a shorter TE and
consequently a higher SNR. Complementary, recent developments in acquisition
techniques, such as simultaneous multi-slice (SMS) [2], provide a significant
reduction of the acquisition time. With the introduction of super-resolution
reconstruction techniques for DTI (SR-DTI) [3, 4], high resolution DTI has
become more feasible on MRI scanners with a regular gradient set. In this work,
we propose to combine SR-DTI with SMS acquisitions (SMS-SR-DTI). This enables
high resolution in vivo DTI within a clinically feasible scan time on a common
MRI scanner.
Method
Four diffusion weighted (DW) data sets with voxel size
1.25x1.25x2.5 mm
3 were acquired on a 3T clinical scanner with a common
gradient set (80 mT/m, Prisma, Siemens, Erlangen Germany), using SMS
echo-planar imaging (EPI) and an SMS factor of 3. Each data set was acquired
with a different slice orientation, which was rotated around the phase encoding
axis with incremental steps of 45°. Each of the four data sets consisted of 12
DW images (b=1000 s/mm
2) and 2 non-DW images (b=0 s/mm
2).
One of the non-DW images was collected with reversed phase encoding blips. The diffusion gradient directions were sampled
differently for each subset, leading to a total of 48 unique diffusion gradient
directions. The overall
scan time was 7min24. The acquired DW images were first corrected for EPI
distortions by using reversed phase encoding correction [5, 6]. From the
corrected DW images, high resolution DTI parameters were estimated on a grid
with voxel size 1.25x1.25x1.25 mm
3 using SR-DTI [3]. The motion was
modeled within the SR-DTI model by an affine transformation, which was
estimated using an iterative model based registration method [7]. To illustrate
the high resolution of the resulting DTI parameters, DTI parameters were also
estimated from high resolution HCP data and from low resolution data, both
acquired in the same acquisition time as the SMS-SR-DTI data. From a
pre-processed HCP data set (300 mT/m) with voxel size 1.25x1.25x1.25 mm
3,
1 non-DW and 40 DW (b=1000 s/mm
2) images were selected. The total acquisition
time was 7min36. For the low resolution data, 2 non-DWI (one with
reversed phase encoding) and 31 DW (b=1000 s/mm
2) images with voxel
size 2x2x2 mm
3 were acquired. The acquisition time was 7min31. The
images were corrected for EPI distortions prior to estimating the DTI
parameters. From each DTI parameter set the directionally encoded color
fractional anisotropy (DEC FA) was calculated.
Results
Figure 1 illustrates a transversal, coronal and sagittal
slice of the DEC FA map for the low resolution data (fig 1a), the HCP data (fig
1b) and the SMS-SR-DTI data (fig 1c). Note that as each data set is acquired at
a different scanner and from a different healthy volunteer, no direct
quantitative comparison can be made between the different data sets. The
SMS-SR-DTI map shows finer structures compared to the low resolution map. The
SMS-SR-DTI map and the HCP map show a similar level of detail. This is for example clearly visible in the
cerebellum (right column in figure 2). The left column of figure 2 shows a
coronal zoom of the posterior region of corona radiate (pcr). In both the SMS-SR-DTI
and HCP maps, the pcr, tapetum and superior longitudinal fasciculus (slf) can
be delineated while in the low resolution map, the tapetum is not discernible from
the pcr and slf due to large partial volume effects.
Conclusion
SR-DTI was combined with an SMS acquisition to estimate
DTI parameters with a high resolution from DW images acquired in a clinically
feasible scan time. For a fixed acquisition time and with data acquired on an
MRI scanner with weaker gradients, SMS-SR-DTI accomplishes a resolution
visually similar to the HCP data. The use of SMS-SR-DTI opens up exciting
possibilities for diffusion MRI in research and clinical routine.
Acknowledgements
This work was financially supported by the Interuniversity Attraction
Poles Program (P7/11) initiated by the Belgian Science Policy Office. B. J. is a post-doctoral research fellow
supported by the Research Foundation Flanders. Data were provided [in
part] by the Human Connectome Project, WU-Minn Consortium (Principal
Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the
16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience
Research; and by the McDonnell Center for Systems Neuroscience at Washington
University.References
[1] Ugurbil
et al. “Pushing spatial and temporal resolution for functional and diffusion
MRI in the Human Connectome Project”, NeuroImage 2013; 80:80-104
[2] Setsompop, et
al. "Improving diffusion MRI using simultaneous multi-slice echo planar
imaging." Neuroimage 2012; 63.1:569-580.
[3] Van
Steenkiste et al. “Super-resolution reconstruction of diffusion parameters from
diffusion-weighted images with different slice orientations.” Magn Reson Med.
doi: 10.1002/mrm.25597
[4] Fogtmann et al. “A
unified approach to diffusion direction sensitive slice registration and 3-D
reconstruction from moving fetal brain anatomy.” IEEE T Med Imaging 2014; 33(2):272-289
[5] Andersson et al. “How to correct susceptibility distortions in spin-echo
echo-planar images: application to diffusion tensor imaging”. NeuroImage 2003,
20(2):870-888
[6] Smith et al. “Advances in functional and structural MR image analysis and
implementation as FSL.” NeuroImage 2004; 23(S1):208-219
[7] Bai et al. “Model-based
registration to correct for motion between acquisition in diffusion MR imaging”,
I S Biomed Imaging 2008; 5:947-950