Beachball: A fast and efficient 3D k-space trajectory with time optimal gradients
Pavan Poojar1 and Sairam Geethanath1

1Medical Imaging Research Centre, Dayananda Sagar College of Engineering, Bangalore, India

Synopsis

3D Variable density spiral (VDS)–“beachball” was designed and generated to cover 3D k-space efficiently. Beachball is obtained by rotating along one axis of a 2D VDS for various number of shots. Beachball was demonstrated in-silico by analyzing point spread function and restrospective analysis of a in-vitro water phantom. Reconstruction was done using NUFFT. Optimal gradient waveforms were generated for beachball by using convex optimization formulations based on the constraints of maximum gradient amplitude and slew rate. This trajectory provides for smoother coverage and densely sampled at the centre of k-space and allows implementation of silent MR.

Purpose

Rapid MR imaging typically requires highly undersampled k-space with efficient 3D coverage, incoherent artifacts and time optimal gradient waveforms. The current work involves design and retrospective demonstration of the 3D variable density spiral (VDS) – “beachball” trajectory which accomplishes the above requirements and delivers an optimal balance among the requirements listed above.

Methods

The beachball k-space trajectory, so called due to the shape of k-space coverage that is achieved, was generated by rotating 2D VDS around one axis (x-axis) over 3600, shown in figure 1. The designed 2D VDS parameters are shown in figure 2. In silico simulation: A 3D point spread function of size 128x128x128 was simulated with a signal intensity value of 100 and size 5x5x5 voxel at the centre. 3D NUFFT reconstruction was performed for fully sampled k-space with the beachball trajectory using the irt toolbox [1]. PSF was obtained for 837 shots as shown in figure 3. 3D PSF was also obtained for 20 shots with each shot of 1251 points resulting in undersampled k-space data. All simulation was performed using MATLAB, Mathworks., USA. In vitro: Images were acquired from 1.5 T Siemen’s Avanto scanner for a spherical water phantom by using 3D Magnetization Prepared Rapid Acquisition GRE (MP-RAGE) sequence with TR/TE=1650/2.51 ms, matrix size 128x128x128, slice thickness =2.08 mm. Reconstruction was performed in a similar manner to that of the in silico data. Gradient design: Optimal gradient waveforms were obtained for single and 8 shots by using a convex optimization tool (cvx) [2] solving $$$\parallel k-A\times g \parallel$$$, where $$$\parallel . \parallel$$$ represents the norm operator, A is the integration matrix developed based on the trapezoidal rule, and g is the gradient under the constraints of maximum gradient amplitude 33 mT/m and maximum slew rate 100 mT/m/ms.

Results

Figure 1(a) shows the 2D VDS for 1 shot, which is in x-y plane and has 1251 points. Figure 1(b) and 1(c) depict 20 shots in which first and last point was sampled 20 times. In silico : Figure 3 shows the PSF using NUFFT reconstruction in all the three planes for 837 shots for the centre slices. Incoherent artifacts can be observed in all the three planes, which can be reduced using compressed sensing. Figure 4 shows the point spread function, which is reconstructed using 3D NUFFT and shown in all the three planes. In vitro: Figure 4(e) depicts the water phantom and corresponding reconstruction using 3D NUFFT was shown in figure 4(f)-(h) for xy, xz and xz planes respectively. Gradient design: VDS, which is rotated at an angle of 45 degrees along x-axis is depicted in shown in figure 5(a) and corresponding gradient waveform is shown is figure 5(b). Figure 5(a) shows 3D VDS for 8 shots and its equivalent gradient waveforms for 8 shots where gradient along x-axis was not changed and are shown in figure 5(b).

Discussion and Conclusion

Maximum extent of k-space can be reached in a optimal time as it is VDS and is superior to stack of spirals which has constant space between the cycles [3]. First and last points of the beachball trajectory are sampled Ns times as shown in figure 1(b). 3D PSF, which is shown in figure 3 in all the direction, has incoherent artifacts in both in silico simulation and in vitro. This reconstruction was performed by using 20 shots. Optimal gradient waveforms were generated based on the constraints above-mentioned constraints shown in figure 5 for 8 shots. In figure 5 (b), gradients in x-direction were constant for all the 8 shots as VDS is rotated along x-axis. Y. Shu et.al., [4] developed a 3D k-space trajectory - “spherical shell trajectory” which is densely sampled at the pole region as opposed to the centre region which beachball accomplishes. This is relevant to acquisitions exploiting structure in k-space. This trajectory provides for smoother coverage and densely samples at the centre of k-space and allows implementation of silent MR. In addition, this sequence can be used to rapidly image 3D volumes as each shot reaches out to different corners of k-space in an optimal manner. Future work involves prospective implementation of beachball trajectory for 3D ASL.

Acknowledgements

No acknowledgement found.

References

[1] Jeffrey A. Fessler “Nonuniform Fast Fourier Transforms Using Min-Max Interpolation”, IEEE T-SP, 51(2):560-74, Feb. 2003 [2] Micheal Grant and Stephen Boyd, Disciplined convex programming,2014. [3] Holden H. Wu and Dwight G. Nishimura “3D Magnetization-Prepared Imaging Using a Stack-of-Rings Trajectory”, Magn Reson Med. 2010 May ; 63(5): 1210–1218. doi:10.1002/mrm.22288 [4] Y. Shu, et.al., “3D RINGLET: Spherical Shells Trajectory for Self-Navigated 3D MRI”, Proc. Intl. Soc. Mag. Reson. Med. 13 (2005).

Figures

Figure 1. 3D variable density spiral a) 1 shot. b) 20 shots. c) 837 shots.

Figure 2: 2D VDS parameters.

Figure 3. PSF using NuFFT reconstruction for 837 shots. a) PSF. b), c) and d) are PSFs for x-y, x-z and y-z directions respectively.

Figure 4. NuFFT for simulation and phantom data of size 128x128x128. a) mask. b), c) and d) are PSF using NuFFT reconstructed image for xy, xz and yz plane respectively. e) Water phantom image (centre slice). f), g) and h) are NuFFT reconstructed image for xy, xz and yz plane respectively.

Figure 5. a) VDS k-space trajectory (1251 points) which is tilted by 45 degrees. b) Gradient waveforms for all the three axes c) 3D k-space trajectory -8 shots and each shot has 1251 points. b) Gradient waveforms for all the 8 shots.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3187