Jana Hutter1, Paddy J Slator2, Anthony N Price3, Ana Dos Santos Gomes4, Laura McCabe4, Maria Murgasova Kuklisova5, Paul Aljabar1, Mary Rutherford4, and Joseph V Hajnal3
1Biomedical Engineering, King's College London, London, United Kingdom, 2Centre for Medical Image Computing, University College London, London, United Kingdom, 3Biomedical Engineering Department, King's College London, London, United Kingdom, 4Centre for the Developing Brain, King's College London, London, United Kingdom, 5Perinatal Imaging, King's College London, London, United Kingdom
Synopsis
Conventional diffusion MRI acquisitions acquire all slices per volume with the same diffusion weighting. This can have two drawbacks: Excessive heating of gradient hardware caused by multiple repeats of the same combination of large drive currents, and low signal at high b-values, which impairs motion correction based on image registration. For placental diffusion MRI, where large slice stack are needed for spatial coverage and anatomical structure can be lost after diffusion, both of these factors can be extreme. We propose intra-volume interleaving of different diffusion weightings, ordered to facilitate image registration for motion correction and minimise gradient heating.
Background
The
potential of diffusion MRI (dMRI) to probe microstructural complexity
continues to grow with advanced biophysical modelling techniques (1).
Emerging
applications such
as placental dMRI offer fascinating new insights, but also
significant challenges, since these advanced analyses require ever
more data at high diffusion sensitivity levels (bval) and high
angular resolution (HARDI). Furthermore, while frequently
employed 2D-single-shot-echo-planar imaging (ssEPI) freezes
intra-slice motion, it does not resolve inter-slice movement, so
that, especially in non-brain applications, post-processing
techniques such as slice-to-volume-reconstruction with registration
are required to ensure 3D-continuity.(2,3) In this
work we address two problems that are common in dMRI and are
accentuated in the placenta application. First, the signal content of
high b-value images is strongly attenuated, potentially close to the
noise floor. Such images may lack anatomical features, hindering
image registration approaches. Second, the required gradients place a
high load on the gradient system and require time to cool, which can
prolong scan time. A large number of slices, as is required for
placental coverage, exacerbates this problem by demanding many EPI
shots with unchanging gradient load, which can cause excessive
equipment heating. To address both issues, we explore free
interleaving of diffusion weighting for each slice within each
acquired volume.Methods
The
proposed free diffusion sampling technique breaks the traditional
one-volume-one-diffusion-sampling paradigm (Fig.1a) and allows the
acquisition of every slice with a different diffusion weighting
(Fig.1b).
This
novel capability, including all required modifications to gradient
duty cycle calculations, reconstruction and slice ordering was
implemented into the ssEPI sequence on a Philips 3T-Achieva scanner
(R3.2 software).
An ideal
design maximally interleaves low and high b-values for the two above
mentioned goals, but ensures that complete volumes are obtained even
if the scan is interrupted or abandoned. Therefore, a bloc design
with length L was chosen, where L consecutive diffusion samplings
were interleaved so that all slices are acquired with all diffusion
weightings (i.e. b-value shells) after L volumes. To distribute
thermal load the direction of sensitisation can be varied as well as
the b value for each slice. The number of required blocks depends on
the total number of diffusion samples.
To ensure
optimal registration properties, the low-b value data was spread out
not only maximally in time to densely sample motion patterns but also
in space to ensure spatial proximity of every high-b-slice to a
low-b-slice. This was realized by maximizing the inter-slice –
inter-shot distances. For example, for the frequently used even-odd
slice ordering (0-2-4-6….1-3-5-7…), this requires that low
b-value-slices are acquired every L shots such that the step stride
wraps between slice groups. See Fig.1c for an example with 35
slices, L=5.
The
required post-processing algorithm was developed in-house using
IRTK(4).
To ensure optimal registration, a single low-b volume was
acquired in a 12s-breath-hold
in addition to the diffusion dataset. The complete registration
algorithm is shown in Fig.2. Here, the threshold between high-b and
low-b was chosen to be b20. Linear interpolation was used between
transformations.
To
illustrate the technique, 4 shell HARDI placental dMRI data with a
total of 28 directions (4b0,6b1000,6b2000 and 12b4000), isotropic
resolution 2.2mm3, 65 slices, L=5, SENSE1.8 was acquired with a
32ch-cardiac coil. The protocol was repeated with three
different acquisition orders (I) conventional slice-stacks with
sequentially increasing b-value, (II) optimally ordered
volume-interleaving to balance gradient demand and (III) optimally
ordered full slice-interleaving.Results
Management
of thermal demands (Fig.3) allowed the scan time to be reduced from
10:59s (sequential) to 7:23 (volume-interleaved)
to 5:28 (slice-interleaved). Image registration failed for full
volumes of high b-value images, but was feasible when temporally
adjacent low-b data was available. Figure 4ab illustrates data
from conventional and fully interleaved acquisition for 3 consecutive
volumes, and Fig. 4c shows registration results with the breathhold
target-volume (left) and aligned b400 data for conventional (middle)
and the fully interleaved acquisition with the proposed approach
(right). The improved alignment is clear. The plotted translations in
AP-direction resulting from the low-b registration (Fig.5a overlaid
with respiratory belt measurements) nicely depict the breathing
cycle, whereas the results from conventional whole-volume approach
only accurately depict motion for the b0 volumes (b) but fail for
b1000 (c).Discussion
The fully
flexible framework presented proved effective in both decreasing
acquisition time and enhancing subsequent processing for placental dMRI, but clearly is widely generalizable to any diffusion study that
includes high b-values. Future developments could include dynamic field map calculation based on sparse but frequently acquired b0-slices, which can provide significant improvement in the presence of
motion or varying B0-fields(e.g. as a result of intestinal gas near
the placenta).Acknowledgements
MRC strategic funds (MR/K006355/1), GSTT BRC and NIH-funded Placenta imaging Project: project number 1U01HD087202-01.References
(1)Zhang H, Schneider T, Wheeler-Kingshott CA, Alexander DC. NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain. Neuroimage. 2012 Jul 16;61(4):1000-16. doi: 10.1016/j.neuroimage.2012.03.072. Epub 2012 Mar 30
(2) Kuklisova-Murgasova M, Quaghebeur G, Rutherford MA, Hajnal JV,
Schnabel JA. Reconstruction of fetal brain MRI with intensity
matching and complete outlier removal. Med Image Anal. 2012
Dec;16(8):1550-64. doi: 10.1016/j.media.2012.07.004. Epub 2012 Aug 9.
(3) Ben-Amitay S,
Jones DK, Assaf Y. Motion correction and registration of high b-value
diffusion weighted images. Magn Reson Med. 2012 Jun;67(6):1694-702.
doi: 10.1002/mrm.23186. Epub 2011 Dec 19.
(4) D. Rueckert, L. I. Sonoda, C. Hayes, D. L. G. Hill, M. O. Leach, and
D. J. Hawkes. Non-rigid registration using free-form deformations:
Application to breast MR images. IEEE Transactions on Medical
Imaging, 18(8):712-721, 1999