Yuning Gu1, Huiyun Gao2, Kihwan Kim1, Ciro Ramos-Estebanez3, Yunmei Wang2, and Xin Yu1,4,5
1Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Case Cardiovascular Research Institute, Case Western Reserve University, Cleveland, OH, United States, 3Department of Neurology, Case Western Reserve University, Cleveland, OH, United States, 4Department of Radiology, Case Western Reserve University, Cleveland, OH, United States, 5Department of Physiology and Biophysics, Case Western Reserve University, Cleveland, OH, United States
Synopsis
A golden-means-based 3D radial sampling and k-space weighted image reconstruction method
was developed for dynamic tracking of intravenously injected 17O-water
in mouse brain. The method enabled image
reconstruction with adaptive temporal resolution of 3 to 15 s to capture the
regional differences in 17O-water uptake and washout kinetics in
post-stroke mice, with an isotropic voxel size of 1.77 µL on a 9.4T scanner.
Introduction
Dynamic 17O-MRI enables imaging of
cerebral perfusion and metabolism. However,
current 17O-MRI methods offer limited spatial and temporal
resolution due to the low sensitivity and fast T2 decay of 17O
signal. Imaging strategy using stack-of-stars
sampling and k-space view-sharing has
enabled dynamic tracking of intravenously injected 17O-water in
mouse brain with 7.6-s temporal resolution and 11.25-μL spatial resolution.1 However, further improving the spatial
resolution using this approach entails increasing the phase-encoding steps,
leading to reduced temporal resolution. In
the current study, we employed golden-means-based 3D radial sampling to enable
flexible reconstruction of dynamic 17O-MRI data with isotropic
spatial resolution and adaptive temporal resolution tailored to the dynamics of
the acquired signal. The efficacy of
this method was demonstrated in quantifying 17O-water uptake and
washout kinetics in post-stroke mouse brain.Methods
3D Radial Sampling: All
MRI studies were performed on a Bruker 9.4T scanner. 17O data were acquired with a 2-cm
surface coil. Data acquisition used center-out,
golden-means-based 3D radial sampling proposed by Chan et al (Fig. 1a).2 Specifically, for the mth spoke,
its azimuth angle and projection on the z-axis were $$$2\pi m\varphi_1$$$ and $$$(-1)^mk_{max}mod(m\varphi_2,1)$$$, respectively, with $$$\varphi_1$$$ (=0.6823) and $$$\varphi_2$$$ (=0.4656) being the 2D golden means derived from the
Fibonacci matrix. This sampling scheme
achieved a nearly uniform coverage of k-space with an arbitrary number of
spokes. For each spoke, 16 data points
were acquired in 2 ms with 24x24x24 mm3 FOV, yielding a nominal
isotropic resolution of 0.46 μL. TR and
TE were 10 ms and 75 μs, respectively.
Image Reconstruction: Image
reconstruction used a 3D k-space
weighted image contrast (KWIC) strategy for simultaneous high temporal and
spatial resolution.3 For a specific time frame, the k-space was partitioned
into a set of concentric shells. Each shell
consisted of radial spokes acquired around the time frame of interest, with the
number of spokes being a Fibonacci number to ensure the most uniform coverage. The
minimal number of spokes needed to fulfill the Nyquist criterion in each shell is
shown in Fig 1b. With the center k-space partition updated every 60 spokes, this
sampling and reconstruction scheme gave rise to an effective temporal
resolution of 0.6 s. Further, temporal resolution can be traded to gain SNR by
increasing the number of spokes in each shell.
Simulation: Simulation studies were performed on
a modified Shepp-Logan phantom to optimize the data reconstruction strategy. The
phantom consisted of three compartments, with three segments in the time courses
of 17O signal changes in each compartment representing the baseline,
the injection, and the washout phases of an 17O-water injection
experiment (Fig. 2a). Gaussian noise was
added to the k-space data at a level
similar to in vivo SNR. Image reconstruction used varied temporal
resolution. A 3-s window size was used to capture the rapid signal increase
during the injection phase, and it transitioned from 3 to 15 s during the
washout phase to achieve the best trade-off between SNR and accurate
delineation of signal dynamics (Fig. 2b). Parameter estimation using the
adaptive reconstruction strategy was compared with that using fixed window size
of 3 s and 15 s.
Point Spread Function: Point
spread function (PSF) using the proposed acquisition and reconstruction
strategy was measured on a point-source phantom. A 1-mm microhematocrit capillary tube filled
with 17O-water (40% enrichment) was placed in parallel to the
z-direction of the magnet. The full
width half maximum (FWHM) of the PSF was used to calculate the actual voxel
size.
In Vivo Study: Adult
male C57BL/6 mice (n=8) underwent 60-min middle cerebral artery occlusion
(MCAO) surgery. MRI scan was performed 2 hours post MCAO with 0.1 µL of 17O-water
injected via tail vein. 17O-MRI acquisition was performed
continuously for 25 min covering baseline (8 min), injection (~10 s), and washout
(17 min) phases. Peak and steady-state 17O
signal, and the washout rate of 17O-water was estimated. Results
Point Spread Function: Fig. 3a
shows the cross-sectional image of a microhematocrit capillary tube. Its 1D
profile was considered representative of the PSF reconstructed using the
proposed method. The FWHM of the PSF was 1.21 mm, corresponding to a true voxel
size of 1.77 µL.
Simulation Study: Simulation
results are shown in Fig. 2c-e. The 17O signal changes in each
compartment was representative of 17O-water kinetics in normal (#1),
infarct core (#2), and penumbra tissue (#3) (Fig. 2a). The window size for adaptive reconstruction
was derived from the derivative of the center k-space data (Fig. 2b). Comparing
to reconstruction with fixed window sizes, adaptive reconstruction led to more
accurate parameter estimation with reduced deviation from the ground truth (Fig.
2c-e).
In Vivo Study: Fig. 4 compares
17O dynamics in MCAO-affected and contralateral regions in post-stroke
mouse brain. Slower transition from
wash-in phase to wash-out phase was observed in the stroke-affected region
(Fig. 4a). Peak 17O-water uptake and washout rate were reduced by
20% and 30%, respectively, in stroke-affected hemisphere, suggesting a
perfusion deficit. Discussion and Conclusion
3D golden-means-based radial sampling, in
combination with KWIC reconstruction, enables reconstruction of 17O-MRI
data with adaptive window size for accurate tracking of the dynamics of 17O-water
with high isotropic spatial resolution in post-stroke mouse brain. Acknowledgements
This work was supported by a grant from the
National Institute of Health (R01 EB23704).References
1. Liu Y, Zhang Y, Wu C, Zhu J, Wang C,
Tomko N, Linetsky MD, Salomon RG, Ramos-Estebanez C, Wang Y, Yu X.
High-resolution dynamic oxygen-17 MR imaging of mouse brain with
golden-ratio–based radial sampling and k-space–weighted image reconstruction. Magn
Reson Med. 2018;79:256–263.
2. Chan RW, Ramsay EA, Cunningham CH,
Plewes DB. Temporal Stability of Adaptive 3D Radial MRI Using Multidimensional
Golden Means. Magn Reson Med. 2009;363:354–363.
3. Song HK, Dougherty L. k-Space weighted
image contrast (KWIC) for contrast manipulation in projection reconstruction
MRI. Magn Reson Med. 2000;44:825–832.