Wenwen Jiang1, Frank Ong2, Kevin M Johnson3, Scott K Nagle4, Thomas Hope5, Michael Lustig2, and Peder E.Z Larson5
1Bioengineering, UC Berkeley/UCSF, Berkeley, CA, United States, 2Electrical Engineering and Computer Science, UC Berkeley, Berkeley, CA, United States, 3Medical Physics, University of Wisconsin, Madison, Madison, WI, United States, 4Radiology, University of Wisconsin, Madison, Madison, WI, United States, 5Radiology and Biomedical Imaging, UCSF, San francisco, CA, United States
Synopsis
Structural pulmonary imaging with MRI has many potential applications
including lung nodule detection and interstitial lung disease assessments, but
is limited by the challenges of short T2*, low proton density, and respiratory
and cardiac motion. We propose a combination of an optimized 3D UTE acquisition
with advanced reconstruction methods, including motion correction, parallel
imaging, and compressed sensing, aiming to make MRI become a clinical option
for pulmonary imaging.Target audience
Thoracic radiologists, engineers working
on reconstruction and motion correction
Purpose
Structural
pulmonary imaging with MRI has potential to reduce radiation dose from repeated CT
scans, particularly in pediatric diseases requiring longitudinal follow-up,
such as cystic fibrosis. However, structural lung MRI has seen limited clinical success, due to the combined factors of short
T2*, low proton density, and respiratory and cardiac motion. Recently, 3D UTE
sequences have been proposed with optimized SNR efficiency from the data
acquisition1 and respiratory gating based on a respiratory belt. However,
respiratory belt signals are often not robust or may include significant signal
drift. This leads to suboptimal respiratory gating and produces motion blur in
the resulting images. Moreover, traditional respiratory gating often results in
low imaging efficiency.
The purpose
of this study is to optimize the reconstruction of 3D UTE MR pulmonary imaging
using 1) a self-gating technique to measure respiratory motion from MRI data
itself (DC based self-gating) 2,3, 2) a soft-gating technique to
penalize the motion state inconsistency 4, and 3) a motion-resolved
technique to provide images of all respiratory motion states,5 and combine them
with L1-ESPIRiT, 6 an autocalibrating parallel imaging and compressed
sensing method.
We evaluated the proposed methods across different applications:
detecting pulmonary nodules, cystic fibrosis and demonstrated the initial
feasibility of the proposed method.
Methods
Optimized 3D
UTE SPGR sequence with slab excitation and a bit-reversed pseudo-random view
ordering from Johnson et al. 1 was implemented on a 3T (GE
Healthcare) scanner at two sites. Free-breathing scans on a healthy volunteer as
well as on 6 patients were performed across sites, with prescriptions similar
to the following: field of view (FOV) = 32×32×30 cm3, TE/TR = 70 μs / 2.932 ms, flip angle = 4º, 1.25 mm isotropic resolution and total acquisition time is 5 min
14 s.
We
extracted
the respiratory motion from the acquired data by applying a low-pass filter on k-space
center signal over time ($$$\|k_0\|$$$).
For
the soft-gating technique, we applied exponential decay weighting,4
to penalize motion state inconsistency, resulting in an effective undersampling
ratio of 2.5:
$$w = \left\{\begin{array}{ll}
e^{-\alpha(d - Thresh)} & \textrm{otherwise} \\
1 & d \le Thresh
\end{array}
\right. \\
\text{ d is the respiratory motion estimation, $\alpha$ is scaling factor}\\
\text{ w is the soft-gating weights. } $$
For the
motion-resolved reconstruction,
5 we divided the data equally into 5
motion states and enforced temporal smoothness between motion states by adding
total variation norm along respiration dimension on the objective function.
Additionally, we incorporated both methods with L1-ESPIRiT, a parallel imaging
and compressed sensing method to better exploit the inherent incoherency from
the bit-reversed UTE acquisition. Reconstruction and motion correction models are shown in
Figure 1. Due to current memory limits on our workstations, the motion-resolved
images were reconstructed only on the volunteer with reduced isotropic
resolution of 1.6 mm. We compared the soft-gating and motion-resolved
L1-ESPIRiT
6 reconstruction with the traditional reconstructions, gated/non-gated
gridding with density compensation, by evaluating qualitative SNR and image
sharpness. All the reconstructions were carried out using the open source
Berkeley Advanced Reconstruction Toolbox (BART: https://mikgroup.github.io/bart/).
7Results
Figure 2 shows the comparison of traditional reconstructions (left and
middle) with our soft-gating reconstruction (right). Red arrows point out the
comparison where vessels and fine structures were blurred out by the
respiratory motion when the traditional reconstruction was used, while the
soft-gating technique was able to resolve the fine structures and diaphragm.
Yellow arrows show the streaking artifacts due to gating induced undersampling
when parallel imaging and compressed sensing were not applied. Figure 3 shows
the motion-resolved images of all the motion states (from left to right) from
the same dataset, and they also show greatly improved sharpness over the
traditional reconstructions. Soft-gating with L1-ESPIRiT has qualitatively
improved SNR than each motion state of motion-resolved result, while
motion-resolved results provide all the motion states for comprehensive
evaluations.
Figure 4 shows the comparison
of traditional reconstructions with the soft-gating reconstruction on a
cystic fibrosis patient. Figure 5 shows clinical examples of a 4 mm pulmonary nodule
(as green circle indicates) and cystic fibrosis (as the red circle indicates).
Conclusion
In our study, DC based self-gating was reliable in extracting
motion signal from MR data. Both soft-gating and motion resolved techniques
addressed the motion issues for pulmonary imaging effectively. Given the
optimized 3D UTE sequence combined with the optimized reconstruction, MRI
could become a clinical standard for pulmonary imaging. In the spirit of
reproducible research, we will provide the corresponding reconstruction code
online in BART.
7Acknowledgements
No acknowledgement found.References
[1] Johnson KM, et al.
MRM 2013; 70(5):1241–1250. [2] Larson AC, et al. MRM 2004
Jan; 51(1): 93–102. [3] Brau AC, Brittain JH. MRM 2006 Feb;55(2):263-70. [4]
Cheng JY, et al. JMRI 2014; 2014. DOI: 10.1002/jmri.24785. [5] Li F, et al. MRM
2015; DOI: 10.1002/mrm.25665. [6] Uecker M, et al. MRM
2014; 71:990–1001; [7] BART: 10.5281/zenodo.31907.