Free-breathing non-contrast enhanced 3D radial respiratory-motion resolved pancreatic MRI at 3T using sparse iterative reconstruction
Jessica AM Bastiaansen1, Jerome Yerly1,2, Jean-Baptiste Ledoux2, Ruud B van Heeswijk1,2, Davide Piccini3, and Matthias Stuber1,2

1Department of Radiology, University hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 2Center for Biomedical Imaging, Lausanne, Switzerland, 3Advanced Clinical Imaging Technology, Siemens Healthcare, Lausanne, Switzerland

### Synopsis

Pancreatic MRI is commonly performed during breath-held or navigator-gated acquisitions. The long breath-holds needed for high spatial resolution are not always feasible in patients and residual respiratory motion may still occur. Additionally, in some implementations, the navigator leads to a local signal void that may obscure parts of the anatomy of interest. Here we used a free-breathing self-navigated 3D radial gradient-recalled-echo (GRE) imaging sequence, and compared the 1D motion correction as performed on the scanner versus a motion-resolved 4D sparse iterative reconstruction. We show that non-contrast enhanced pancreatic MRI can be performed at 3T during free-breathing, while motion-resolved sparse reconstruction can efficiently minimize the adverse effects of respiratory motion.

### Purpose

Currently, MR examinations of the pancreas are performed during multiple breath-holds or navigator-gated acquisitions, and rarely in 3D. However, improved spatial resolution is needed for an in-depth anatomical examination of the organ [1]. The downside of contemporary techniques is that the long breath-holds needed to obtain high spatial resolution are not always feasible in patients and residual respiratory motion may still occur. Additionally, in some implementations, the diaphragmatic navigator may lead to a local signal void that obscures parts of the anatomy of interest. To overcome these challenges, free-breathing acquisition schemes can be used in combination with retrospective motion correction. However, the presence of signals from static structures, such as abdominal fat that cannot always be homogeneously suppressed, causes artifacts in the motion compensated data. In this context we explored the use of a free-breathing self-navigated 3D golden angle radial gradient-recalled-echo (GRE) imaging sequence using either fat saturation or water excitation, and compared a 1D motion correction [2] as performed on the scanner versus a motion-resolved 4D sparse iterative reconstruction [3].

### Materials and Methods

Pancreatic MRI was performed in healthy volunteers (n=3) on a 3T clinical scanner (PRISMA, Siemens Healthcare, Erlangen, Germany). Data were acquired using a prototype ECG-triggered respiratory-self-navigated free-breathing 3D radial GRE imaging sequence [2] preceded by an adiabatic T2 preparation module (T2Prep) to improve blood tissue contrast [4]. Two different lipid-nulling strategies were implemented to minimize image artifacts originating from background tissue after motion correction: 1) water excitation (WE) or 2) spectral pre-saturation of the fat signal (FS). Imaging parameters were: field-of-view (220 mm)3, matrix size 2083, TET2-Prep = 40 ms, RF excitation angle 18°, (WE) TE/TR = 2.5 ms/5.6 ms, (FS) TE/TR = 2.0 ms/4.6 ms, with 24 radial readouts per segment for a total of ~15k k-space lines. Both FS and WE datasets were 1D motion-corrected (1D-corr) using a superior-inferior (SI) projection acquired every 24 k-space lines as described in [2] (Fig 1B). The same WE data were also subjected to a 4D motion-resolved (4D-resol) sparse iterative reconstruction [3] (Fig 1C). For 4D-resol, independent-component analysis (ICA) was performed on the k-space center amplitudes to sort the data into 4 different respiratory phases [5], ranging from end-expiration to end-inspiration. The respiratory-resolved images of the dimension 192×192×192×4 were obtained by solving:

$$\underset{m}{\text{arg min}} \parallel F \cdot C \cdot m - s \parallel _2^2 + \lambda _1 \parallel D_1m\parallel _1$$

where F represents the non-uniform fast Fourier transform (NUFFT) operator, C the coil sensitivity maps, m the 4D image set to reconstruct, s the radial k-space data, D1 the finite difference operators applied along the respiratory dimension, and λ1 = 0.05-0.08 as a regularization parameter, which was empirically selected. The reconstructed images using the algorithm for 1D-corr and 4D-resol were compared by visual inspection.

### Results and Discussion

To efficiently minimize the effects of respiratory motion, non-contrast enhanced abdominal MRI was performed and data subjected to two types of motion compensation (Fig. 1). Volunteer data were successfully acquired during free-breathing and reconstructed using both 1D-corr and 4D-resol reconstruction with an isotropic voxel size of 1.1 mm3. The 3D acquisitions which were performed using WE showed an improved fat signal suppression in the abdomen as well as in the anterior and posterior chest when compared to FS (green arrows, Fig. 2, A2-C2). Comparing the 4D-resol versus 1D-corr data, images show improved anatomical detail of the veins and arteries located at the head of the pancreas (yellow arrows, Figure 2, A1-C1, A3-B3), as well as an improved delineation of the splenic vein (orange arrows, Fig. A2-C2). The improved depiction of anatomical details in 4D-resol images is most likely due to the complex respiratory-induced 3D motion of the organs in the abdomen, which the motion-resolved reconstruction takes into account (Fig. 3, and Fig. 4 as animated GIF online). A major advantage of the 4D-resol approach includes that no motion model is needed for motion correction as opposed to 1D-corr, which only accounts for SI motion, but neither for the other dimensions nor for more complex motion components.

### Conclusion

Non-contrast enhanced, fat-suppressed 3D isotropic pancreatic MRI can be performed at 3T during free-breathing, while motion-resolved sparse reconstruction efficiently minimizes adverse effects of respiratory motion.

### Acknowledgements

The authors would like to thank the Center for Biomedical Imaging, Nanotera, and NYU for the use of the NUFFT package provided from www.cai2r.net.

### References

[1] Sandrasegaran, K., et al., State-of-the-art pancreatic MRI. AJR Am J Roentgenol, 2010. 195(1): p. 42-53.

[2] Piccini, D., et al., Respiratory self-navigation for whole-heart bright-blood coronary MRI: methods for robust isolation and automatic segmentation of the blood pool. Magn Reson Med, 2012. 68(2): p. 571-9.

[3] Feng, L., et al., XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing. Magn Reson Med, 2015.

[4] Nezafat, R., et al., B1-insensitive T2 preparation for improved coronary magnetic resonance angiography at 3 T. Magn Reson Med, 2006. 55(4): p. 858-64.

[5] Bonanno, G., et al., Self-navigation with compressed sensing for 2D translational motion correction in free-breathing coronary MRI: a feasibility study. PLoS One, 2014. 9(8): p. e105523.

### Figures

Two different methods were used to compensate for abdominal respiratory motion (A). 1) A 1D motion correction based on the Fourier transformed SI k-space readout of each segment (B). 2) A 4D respiratory-motion resolved sparse iterative reconstruction, which uses data binning in 4 respiratory phases based on ICA analysis of the k-space center amplitudes (C).

Typical pancreatic MR images acquired in this study. Data were either motion-resolved using a 4D-resol technique (A), or 1D corrected using SI projections (B&C). A clear lipid signal decrease (green arrowheads) can be observed comparing a water excitation pulse (WE, B) with a conventional fat saturation (FS, C). Note the increased anatomical detail of veins and arteries (yellow arrows) and improved delineation of the splenic vein (orange arrows) in the motion resolved images.

Visualization of abdominal motion in 4D-resol (A-C) compared with 1D-corr (D-F) displayed in axial, coronal, and sagittal planes. Numbers indicate the different bins used to resolve the respiratory motion using the 4D-resol algorithm. Dashed and solid lines depict the kidney (yellow) or liver (red) edges in bin 4 and bin 1 to emphasize on the motion across the 4 temporal bins.

Animated GIF of a 3D motion-resolved dataset obtained with 4D-resol. Data are displayed in axial, coronal, and sagittal planes.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0619