Steen Moeller^{1}, Sudhir Ramanna^{1}, Edward Auerbach^{1}, Pramod Pisharady^{1}, Christophe Lenglet^{1}, Mehmet Akcakaya^{1,2}, and Kamil Ugurbil^{1}

A method is proposed for self-navigation of DWI 3D multislab multiband SE-EPI, to enable whole brain high-resolution imaging, with optimal
imaging TR for higher SNR efficiency. Data for high b-value (b=3k s/mm^{2}) and 1mm^{3}
resolution is presented.

The use of SMS/MB SE-EPI increases SNR per unit time for diffusion weighted MRI (dMRI), achieving near optimal TR’s for moderate resolutions (e.g. >1 mm isotropic) for whole brain coverage. For higher resolution imaging, however, the 2D accelerated approach is less SNR efficient than an optimized 3D approach because it leads to TR’s that are much longer than T1. The 3D approach enables imaging with the optimal TR (~1.2*T1 [1,2]), but requires multiple repetitions to fully encode the volume. For either a 2D or 3D segmented dMRI SE-EPI acquisition, the need for correcting for physiologically induced phase variations between segments, requires computational approaches for 2D [3,4] and/or additional measurements, such as 2D navigators for 3D [1,5] acquired after diffusion weighting. Such additional measurements reduce the efficiency of 3D approaches by 30-50% [1] and are SNR dependent.

In this work, we propose a self-navigated approach to correct
modulations in k_{z}-segmented 3D EPI, which enables an efficient
real-time processing for integration into existing 3D reconstruction frameworks,
and is suitable for high b-values/low SNR protocols. Use of self-navigation for removing
macroscopic sensitivity to B0 induced phase variations from physiology [5] eliminates
the need for 2D navigators, increasing efficiency. Additionally, we demonstrate a combination with SMS/MB using simultaneously excited multiple slabs (multislab-multiband) for large FOV coverage at optimal TRs. We
use a SQUASHER-type encoding for introducing quadratic phase across the slab
[6] - spreading the signal in k_{z} - to estimate accurately the signals for
self-navigation through k_{z}. Additionally, with SQUASHER, the peak power is
reduced enabling high bandwidth multiband which imparts
significant advantages with respect to Fourier/sinc
encoding.

Imaging: Diffusion‐weighted
data were acquired on healthy volunteers using a 32‐channel receiver head-coil
on 3T Prisma system (Siemens) equipped with 80 mT/m gradients with a slew rate
of 200 T/m/s, using the following : **MB1**: Excitation/Refocusing = HS2R12/HS2R14, duration 7680us,
1mm^{3}, TE/TR of 92.2/1610ms with 12slice/slab, 10 slabs, FOV 210x210x**120** mm^{3}, iPAT=2, Volume acquisition
time (VAT)=26s (TR=9.5s for an equivalent 2D SMS/MB coverage with [MBxiPAT=2x2])
**MB2**: Excitation/Refocusing = HS2R10/HS2R12,
duration 7680us, 1mm^{3}, TE/TR of 92.4/1500ms with 10slice/slab, 16
slabs, FOV 210x210x**160** mm^{3}, iPAT=2, VAT=
21s (TR=12.5s for an equivalent 2D SMS/MB coverage with [MBxiPAT=2x2])). For
2D, the total acceleration is limited to less than 2x2 due to g-factors. For
the prescribed whole brain FOV using MB1 the 3D sequence has a 300% longer VAT
compared to a 2D SMS sequence, but the 3D sequence has 16 more averages. For the larger
FOV the 3D MB2 acquisition has 70% longer VAT, and 14 more averages than the 2D
acquisition. The excitation profiles were designed with 1 slice overlap, and acquisition is with 2 slice oversampling.

Self-navigated segmentation correction: For each k_{z}-plane
a relative phase reference map is calculated
from an (uncorrupted) b=0 acquisition. A k_{z}-dependent phase is
calculated for b>0, and updated with
a low-pass filtered difference relative to the reference phase, see flowdiagram
in figure 1.

Image Analysis: Final images were generated with SENSE-1 combination [8]. These were subsequently processed with TOPUP, EDDY and bedpostX in FSL [9] and visualized with FSLeyes and Connectome Workbench [10].

The quadratic phase in SQUASHER spreads the signal in the k_{z}
direction (**Figure 2A**). b=0 images using
SQUASHER and standard encoding with the Siemens default SE-EPI are shown in **Figure 1B**. The zoomed regions depict sharper
band profiles with the higher bandwidth pulses in SQUASHER. The extracted
physiological induced phase variations $$$\Phi_b(x,y,k_z)$$$ and the reference slab
phase-variation $$$\Phi_{b=0}(x,y,k_z)$$$ are shown in **Figure 3A,B** for a representative slab. Representative image slices from a slab without
and with the proposed self-navigation are depicted in **Figure 3C,D** for b=1500 s/mm^{2}.

The reconstructed SQUASHER 3D SE-EPI images, with an axial
orientation acquisition, before and after correction for slab discontinuity for
b=0, 1500, 3000 s/mm^{2} images are shown in **Figure 4** (row 1, 2 and 3 respectively, with sagittal (left) and
coronal (right) orientation), including profile correction after the
weighted-slab combined signal. The average correction for different b-values is
plotted in **Figure 4B**, showing that a
b-value dependent correction is preferred [5].

The extracted FA maps and fiber orientations results shown in
**Figure 5** have a high degree of similarity
between the MB1 and MB2 data without any evidence of the common slab-boundary
issues in the FA maps.

[1], On the signal-to-noise ratio efficiency and slab-banding artifacts in three-dimensional multislab diffusion-weighted echo-planar imaging. Magn Reson Med. 2015 Feb;73(2):718-25. doi: 10.1002/mrm.25182. Engström M, Mårtensson M, Avventi E, Skare S.

[2] High spatial resolution diffusion weighted imaging on clinical 3 T MRI scanners using multislab spiral acquisitions. Holtrop JL, Sutton BP. J Med Imaging (Bellingham). 2016 Apr;3(2):023501. doi: 10.1117/1.JMI.3.2.023501.

[3] A robust multi-shot scan strategy for high-resolution diffusion weighted MRI enabled by multiplexed sensitivity-encoding (MUSE). Chen NK, Guidon A, Chang HC, Song AW. Neuroimage. 2013 May 15;72:41-7. doi: 10.1016/j.neuroimage.2013.01.038

[4] Multi-shot sensitivity-encoded diffusion data recovery using structured low-rank matrix completion (MUSSELS). Magn Reson Med. 2017 Aug;78(2):494-507. doi: 10.1002/mrm.26382. Mani M, Jacob M, Kelley D, Magnotta V.

[5]. Reducing slab boundary artifacts in three-dimensional multislab diffusion MRI using nonlinear inversion for slab profile encoding (NPEN). Magn Reson Med. 2016 Oct;76(4):1183-95. doi: 10.1002/mrm.26027. Wu W, Koopmans PJ, Frost R, Miller KL.

[6]. SQUASHER: Slice Quadratic Phase with HSn Encoding and Reconstruction. Moeller, Steen Wu, Xiaoping Harel, Noam Garwood, Mike Akcakaya, Mehmet , ISMRM 2017, p 1522

[7] Frequency-Modulated radiofrequency pulses in Spin-Echo and Stimualted-Echo Experiments, Kunz, Dietmar, Magn Reson Med , 1987, (4)129-136

[8] Effects of image reconstruction on fiber orientation mapping from multichannel diffusion MRI: reducing the noise floor using SENSE. Sotiropoulos SN, Moeller S, Jbabdi S, Xu J, Andersson JL, Auerbach EJ, Yacoub E, Feinberg D, Setsompop K, Wald LL, Behrens TE, Ugurbil K, Lenglet C. Magn Reson Med. 2013 Dec;70(6):1682-9. doi: 10.1002/mrm.24623

[9] M. Jenkinson, C. Beckmann, T. Behrens, M. Woolrich, S. Smith. Fsl. Neuroimage, 62 (2012), pp. 782-790 [10] D.S. Marcus, J. Harwell, T. Olsen, M. Hodge, M.F. Glasser, F. Prior, M. Jenkinson, T. Laumann, S.W. Curtiss, D.C. Van Essen. Informatics and data mining: tools and strategies for the human connectome project Front. Neuroinformatics, 5 (4) (2011), pp. 1-12

Figure 1: Flow diagram of the reconstruction pipeline. The
self-navigation is the phase update in steps, 3 and 4, with a hamming window filtered
phase (see also figure 3b), where C_{n} are slab wise sensitivity
profiles. The 1D whole volume correction applied at step 9, uses a smooth
profile function for reference to update from [1]. Parallel Imaging
reconstruction in k_{y} is with GRAPPA using FLEET for ACS, and slice-GRAPPA with
slice-blocking for SMS/MB unaliasing.

Figure 2 :A; The signal in the hybrid space [x,y,k_{z}] for SQUASHER and
the Siemens optimized sequence
B: Image reconstruction for each slab and concatenated
without removing oversampled slices. In the insert, the difference between a
high bandwidth SQUASHER sequence and the Siemens optimized SE-EPI show the
higher spatial specificity in areas with high ΔB0.

Figure 3: Physiological induced phase
variation for b=1500 s/mm^{2 }A: High-resolution reference phase pr. k_{z} plane in a slab in
[x,y,k_{z}] space. B: Estimated low resolution phase variation pr. k_{z}-plane
estimated from the data itself. C: Reconstruction of slices in a slab without
accounting for the physiological induced phase variation. D: Reconstruction of
slices in a slab after accounting for the physiological induced phase variation

Figure 4: Removal
of slab boundary effects. A: the progression of axial 3D slab images from Raw
3D images with over sampling, to profile averaged slabs, to 1D profile
corrected volumes in both sagittal (left) and coronal view (right). The three
rows are for b=0,1500,3000 s/mm^{2 }respectively. B/ The average 1D profile for
b=0,1500,3000 s/mm^{2} respectively. Each volume is corrected with it’s own profile which
is close to the average for a given b-value.

Figure 5: Diffusion data were acquired with two shells (b=1500 and
b=3000 s/mm^{2}), with 36 diffusion encoding directions, and repeated with AP/PA phase encoding.Color coded fiber orientations modulated with FA (upper panels) and FA maps (middle panels) from DTI model fitting
to MB1 (left panels) and MB2 (right panels) data. Lower panels show color coded
multiple fiber orientations at high resolution, resolved by ball and stick
partial volume model, at a region of interest highlighted in red in the middle
panel. The background in lower panels is the sum of the estimated anisotropic fiber
volume fractions.