Oliver Maier1, Stefan M Spann1, Lea Bogensperger1,2, and Rudolf Stollberger1,3
1Institute of Medical Engineering, Technical University Graz, Graz, Austria, 2Institute for Computer Graphics and Vision, Technical University Graz, Graz, Austria, 3Biotechmed, Graz, Austria
Synopsis
Multi-Shell DTI suffers from low SNR for high
b-value data and prolonged scan time. The Gaussian noise assumption
is typically violated
due to multi-coil imaging
and magnitude forming thus
requiring special
treatment to avoid biases in the DTI estimates. To this end, we
propose a model-based reconstruction technique
to exploit the Gaussian noise in the raw k-space data and enable
acceleration of
the DTI measurement.
We show the acceleration potential and quantitative accuracy of the
proposed method for mono- and bi-exponential fitting approaches on
freely available DTI
data and full brain DTI measurements of one
healthy volunteer.
Introduction
Diffusion
MRI with multiple diffusion encodings (directions, b-values) requires
accelerated imaging techniques for the practical applicability and
suffers from poor SNR especially for higher b-values. Proper
treatment of the data is required to avoid biases in subsequent
analysis. Model-based reconstruction has been shown to allow
acceleration of data acquisition beyond conventional
parallel
imaging techniques1,2,3,4. The reconstruction and fitting is
directly performed on the raw measurement data and thus avoids
complications from non-Gaussian noise models, in particular for
multi-array coils. The validity of the Gaussian noise assumptions is
notably important for fitting high b-value data to models such as
Kurtosis or bi-exponential based diffusion tensor imaging (DTI). To this end, we propose a
novel model-based reconstruction method for DTI to reduce the
acquisition time of multi-shell acquisition and overcome non-Gaussian
noise based biases in the estimates of the diffusion tensor. Further, we employ a simultaneous
multi slice (SMS) diffusion sequence to
reduce acquisition time.
This combination yields high resolution (2.3 mm3)
multi-shell DTI measurements of the whole brain within
an acquisition time of 6 min.Methods
For
a given diffusion gradient direction $$$\mathbf{g_m}$$$ and strength $$$b_i$$$ the measured
k-space signal can
be expressed as:
$$y_{i,m}(S_0, \mathbf{D})=\mathcal{F}\,CS_0e^{-b_i\mathbf{g_m}^T\mathbf{D}\mathbf{g_m}}$$
where $$$\mathcal{F}$$$ is the Fourier operator including k-space sampling and $$$\mathcal{C}$$$ are the
complex coil sensitivity profiles estimated from the data5. $$$S_0$$$ corresponds to no diffusion
weighting and $$$\mathbf{D}$$$ is the fully symmetric diffusion tensor, assuming a Gaussian diffusion model. The symmetry property of $$$\mathbf{D}$$$ can be
exploited by fitting the elements of the Cholesky decomposition $$$\mathbf{D}^{}=\mathbf{L}^{}\mathbf{L}^T$$$ . This implicitly poses a positive (semi-) definite constraint on $$$\mathbf{D}$$$,
justified by the physical properties of DTI. The following
optimization task describes the fitting procedure:
$$\underset{x=(S_0,\mathbf{L})}{\min}\,\,\frac{1}{2}\|\sum_{i=1}^{N_b}\sum_{m=1}^{N_g}\mathcal{F}\,CS_0 e^{-i\phi_{i,m}}e^{-b_i \mathbf{g_m}^T\mathbf{L}\mathbf{L}^T\mathbf{g_m}}-y_{i,m}\|_2^2+\lambda{TGV}(x)$$
The
minimization of this non-linear problem is implemented using a
combination of an iteratively-regularized
Gauss-Newton and a primal-dual algorithm6, as recently proposed for
quantitative MRI4, solving
for $$$S_0$$$ and the six non-zero entries of $$$\mathbf{L}$$$. An additional spatial
prior, Total Generalized Variation ($$$TGV$$$)7,8, directly acting on the tensor
elements is applied, which has been shown to be superior compared to
individually regularizing the DWI images8,10,11. Phase
variations $$$\phi_{i,m}$$$ are estimated prior
to the optimization using a CG-SENSE based reconstruction of each
diffusion weighted acquisition. The
implementation of the bi-exponential model12 follows analogously.
Our
proposed method is compared to
adaptive
soft coefficient matching13 using the implementation in DiPy14. For the
reference approach the following steps were performed: reconstruction
of the DWI using a CG-SENSE based algorithm followed by
adaptive
soft coefficient matching13 and finally estimation
of the diffusion tensor by fitting the DWI series using the supplied tensor
fitting routines of DiPy.
The
reference data set consists of magnitude images provided by Hansen et
al.15.
Prospective
accelerated measurements were
performed on
a 3T
MR system (Prisma, Siemens Healthcare,
Germany) using a single-shot
EPI sequence16,17 with 33 diffusion directions15. The
acquisition parameters are given in Table 1.Results and Discussion
Figure
1 shows
that the values
produced by the proposed
model-based reconstruction are highly comparable
to
the reference for
fully sampled, high SNR data, preserving details without blurring of the tensor estimates. Reconstructions from prospective
accelerated in-vivo data show improved visual quality using the proposed
method as compared to the reference (Figure 2 and 3) which
tends to overestimate tensor values in the central brain region.
This
may be
due to
residual coil sensitivity errors and g-factor amplification and becomes even more pronounce with increasing
multi-band acceleration factors (Figure 3). Model-based reconstruction compensates these errors and shows reconstructions with only minimal loss of quantitative and visual quality over increasing acceleration (Figure 3). In
addition, fine
structures are better preserved with
the proposed model-based approach (red arrows Figure 3). This
excellent detail preservation is particularly
visible in the area around the corpus callosum and
in
the sagittal view of the axial diffusivity. The
delineation between CSF and corpus callosum can be difficult in the
reference reconstruction as opposed to the proposed method.
Increasing
errors in the reference reconstruction with increasing acceleration
may stem in part from violations of the Gaussian noise assumption in the DWI images. In contrast, model-based reconstruction exploits the raw k-space data which generally has a Gaussian noise distribution.
Thus the proposed model-based approach utilizes the correct data
similarity measure in form of the L2-norm.
To fully explore the acquired multi-shell data we employed a bi-exponential model. Even though bi-exponential fitting typically demands many b-values the tensor reconstructions show visually good estimates for fast and slow diffusion (Figure 4). Contrast for grey and white matter tracts also improves compared to mono-exponential fitting. Clear boundaries between grey and white matter are also visible in the water fraction maps. Principal diffusion direction maps of the tensor estimates for slow and fast
exchanging compartments reveal connections not visible in the mono exponential counterpart.Conclusion
In
this study we combined an SMS acquisition with in-plane acceleration and a model-based reconstruction approach to reduce the acquisition time
for a multi-shell DTI acquisition of the whole brain down
to
6 minutes. The model-based method
overcomes the inherent transformation of Gaussian noise in multi-coil
magnitude images, typically used for DTI, and thus avoids possible bias in
the tensor estimates.Acknowledgements
Oliver Maier is a Recipient of a DOC Fellowship (24966) of the Austrian Academy of Sciences at the Institute for Medical Engineering at TUGraz.
We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research.
References
1.
Sumpf TJ, Uecker M, Boretius S, Frahm J. Model-based nonlinear
inverse reconstruction for T2 mapping using highly undersampled
spin-echo MRI. J. Magn. Reson. Imaging 2011;34:420–428. doi:
10.1002/jmri.22634.
2.
Wang X, Roeloffs V, Klosowski J, Tan Z, Voit D, Uecker M, Frahm J.
Model-based T 1 mapping with sparsity constraints using single-shot
inversion-recovery radial FLASH. Magn. Reson. Med. 2018;79:730–740.
doi: 10.1002/mrm.26726.
3.
Roeloffs V, Wang X, Sumpf TJ, Untenberger M, Voit D, Frahm J.
Model-based reconstruction for T1 mapping using single-shot
inversion-recovery radial FLASH. Int. J. Imaging Syst. Technol.
2016;26:254–263. doi: 10.1002/ima.22196.
4.
Maier O, Schoormans J, Schloegl M, Strijkers GJ, Lesch A, Benkert T,
Block T, Coolen BF, Bredies K, Stollberger R. Rapid T 1
quantification from high resolution 3D data with model‐based
reconstruction. Magn. Reson. Med. 2019;81:2072–2089. doi:
10.1002/mrm.27502.
5.
Uecker M, Hohage T, Block KT, Frahm J. Image reconstruction by
regularized nonlinear inversion - Joint estimation of coil
sensitivities and image content. Magn. Reson. Med. 2008;60:674–682.
doi: 10.1002/mrm.21691.
6. Chambolle
A, Pock T. A First-Order Primal-Dual Algorithm for Convex Problems
with Applications to Imaging. J. Math. Imaging Vis. [Internet]
2011;40:120–145. doi: 10.1007/s10851-010-0251-1.
7.
Bredies K, Kunisch K, Pock T. Total Generalized Variation. SIAM J.
Imaging Sci. 2010;3:492–526. doi: 10.1137/090769521.
8.
Knoll F, Bredies K, Pock T, Stollberger R. Second order total
generalized variation (TGV) for MRI. Magn. Reson. Med.
2011;65:480–491. doi: 10.1002/mrm.22595.
9.
Knoll F, Raya JG, Halloran RO, Baete S, Sigmund E, Bammer R, Block T,
Otazo R, Sodickson DK. A model-based reconstruction for undersampled
radial spin-echo DTI with variational penalties on the diffusion
tensor. NMR Biomed. 2015;28:353–366. doi: 10.1002/nbm.3258.
10.
Dong Z, Dai E, Wang F, Zhang Z, Ma X, Yuan C, Guo H. Model‐based
reconstruction for simultaneous multislice and parallel imaging
accelerated multishot diffusion tensor imaging. Med. Phys.
2018;45:3196–3204. doi: 10.1002/mp.12974.
11.
Welsh CL, DiBella EVR, Adluru G, Hsu EW. Model-based reconstruction
of undersampled diffusion tensor k-space data. Magn. Reson. Med.
2013;70:429–440. doi: 10.1002/mrm.24486.
12.
Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud J, Laval-Jeantet
M. Separation of diffusion and perfusion in intravoxel incoherent
motion MR imaging. Radiology 1988;168:497-505.
13.
Coupé P, Manjón JV, Robles M, Collins DL. Adaptive
multiresolution non-local means filter for three-dimensional magnetic
resonance image denoising. IET Image Process. 2012;6:558. doi:
10.1049/iet-ipr.2011.0161.
14.
Garyfallidis E, Brett M, Amirbekian B, Rokem A, van der Walt S,
Descoteaux M, Nimmo-Smith I and Dipy Contributors (2014). DIPY,
a library for
the analysis of diffusion MRI data.
Frontiers
in Neuroinformatics, vol.8, no.8.
15.
Hansen B, Jespersen SN. Data for evaluation of fast kurtosis
strategies, b-value optimization and exploration of diffusion MRI
contrast. Sci. Data 2016;3:160072. doi: 10.1038/sdata.2016.72.
16.
Moeller S, Yacoub E, Olman CA, Auerbach E, Strupp J, Harel N, Uğurbil
K. Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration
using partial parallel imaging with application to high spatial and
temporal whole-brain fMRI. Magn. Reson. Med. 2010;63:1144–1153.
doi: 10.1002/mrm.22361.
17.
Setsompop K, Gagoski BA, Polimeni JR, Witzel T, Wedeen VJ, Wald LL.
Blipped-controlled aliasing in parallel imaging for simultaneous
multislice echo planar imaging with reduced g-factor penalty. Magn.
Reson. Med. 2012;67:1210–1224. doi: 10.1002/mrm.23097.