Synopsis

Acceleration techniques like SENSE have enabled significant decreases in scan time, but the $$$G$$$-factor penalty on SNR limits attainable accelerations.
Wave-CAIPI lowers this $$$G$$$-factor by spreading aliasing and taking advantage of the coil sensitivity distributions. A high performance single-axis insert gradient is utilised to increase the attainable wave amplitudes and thereby increasing the effectivity of Wave-CAIPI.

Simulations and 7T phantom & in-vivo acquisitions show significant improvements in $$$G$$$-factor when using Wave-CAIPI, especially when utilizing the higher performance of the insert gradient.
Not only are the $$$G$$$-factors lower but also show increased spreading and lack hard edges, allowing for even faster acquisitions.

Introduction

In recent years, acceleration techniques like SENSE and GRAPPA have enabled significant decreases in scan time, but even with the advance of ultra-high field scanners and their increased SNR, the penalty on SNR from $$$\sqrt{R}$$$ and $$$G$$$-factor limit attainable accelerations.
Improved methods like CAIPI, BPE and Wave-CAIPI¹ have been developed that try to lower this $$$G$$$-factor by spreading aliasing over the FOV and taking advantage of the coil sensitivity distributions. Previous work has predicted that Wave-CAIPI can benefit from using a high-performance insert gradient², even when it only contains a single axis³.

Purpose

To reduce the $$$G$$$-factor penalty on SNR in highly accelerated scans by utilising Wave-CAIPI on a high performance single-axis head insert gradient, thereby allowing for even faster acquisitions without compromising on quality.

Methods

To study the effectivity of Wave-CAIPI with increased wave amplitudes on a single axis, simulations are performed and then validated in both phantom and in-vivo brain acquisitions on a 7T MRI scanner (Philips, Best, The Netherlands) with 32ch head coil (Nova Medical, Wilmington, USA).

The chosen sequence is designed for SWI and consists of a whole-brain 3D GRE with a resolution of 0.8mm³ over a FOV of 256x192x256mm, with $$$T_{R}\mathrm{/}T_{E}\mathrm{/}α\mathrm{/}BW = 43\mathrm{[ms]/}20\mathrm{[ms]/}14°\mathrm{/}285\mathrm{[Hz]}$$$. The acquisition duration is reduced to 2min42sec by accelerating 16-fold, 4 times in both Y-& Z-axes, using SENSE and two types of Wave-CAIPI.

Wave-CAIPI is created by introducing a FOV/2 shift and 10 periods of sinusoidal waves on both Y-& Z-axes. The Z-axis wave is given a $$$90°$$$ phase offset to produce a cosine during the readout. The whole-body gradient specifications limit the wave amplitude to $$$9$$$ [mT/m] and that amplitude is therefore played out on the Y-axis.
The insert gradient produces the Z-axis wave and since previous work has shown that Wave-CAIPI efficiency improves with higher wave amplitudes⁴, the higher performance of the insert gradient is used to double the amplitude of the Z-axis wave to $$$18$$$ [mT/m] in the simulations and phantom acquisitions. To remain under auditory limits, a factor of 1.7x and thus $$$15.3$$$ [mT/m] waves are utilised in the in-vivo acquisitions. A secondary 1x reference acquisition is acquired for direct comparison with the higher performance waves.

Reconstruction is performed using an in-house developed NUFFT-based pipeline implemented in MATLAB complemented by MRecon (GyroTools, Zurich, Switzerland) and BART⁵. The real-time scanner behaviour and all resulting gradient waveforms are reproduced and convolved with a GIRF⁶ captured previously using a field camera⁷ (Skope, Zurich, Switzerland). For scans that just utilise the whole-body gradients the resulting trajectory suffices, but the insert gradient produces a B0 offset when not positioned perfectly on iso-center and therefore requires a PSF scan to measure and correct that offset. Sensitivity maps are calculated using ESPIRiT⁸ on a low-resolution reference acquisition. 18 iterations of CG-SENSE⁹ with $$$L_{2}$$$-regularisation produce the final reconstruction.

Simulations are created from a generated 3D phantom and acquired sensitivity maps. The phantom is multiplied by the sensitivities to produce ideal coil images. Noise is added to create a 14dB SNR and simulated K-space is calculated from those images using a NUFFT on trajectories taken from the reconstruction pipeline above.

Finally, G-factor maps are created for all reconstructions using the pseudo multiple replica method¹⁰ that performed the entire reconstruction 50 times with additional simulated noise.

Results

The $$$G$$$-factor analysis results for the simulations, phantom and in-vivo acquisitions that are depicted in figure 2 to 4 respectively, show a consistent pattern that is quantified in the table of figure 5.

The Wave-CAIPI acquisition shows very significant reduction in $$$G$$$-factor over the SENSE acquisition. The enhanced (1.7x or 2x) Wave-CAIPI acquisition consistently improves over the regular Wave-CAIPI acquisition and shows even lower $$$G$$$-factor values.
The vendors SENSE implementation shows a similar pattern in $$$G$$$-factor to our SENSE implementation, but achieves slightly different results in which our implementation performs better on the phantom acquisition and the vendor SENSE shows favourable in-vivo $$$G$$$-factors.
Exemplar slices show slight folding artefacts in the SENSE acquisitions, which are not visible in the Wave-CAIPI acquisitions. However, the phantom Wave-CAIPI reconstructions show a ghost in the readout dimension that increases with increasing wave amplitude.

Discussion & Conclusion

Wave-CAIPI shows significant and consistent improvements in all acquisition types, not only are the absolute $$$G$$$-factor values themselves lower, they also show geometrical spreading and therefore lack the hard edges that frequently cause visible artefacts.

The enhanced performance of the insert gradient enhanced the effect of Wave-CAIPI by further lowering the $$$G$$$-factor and increasing the spreading. The increase in wave amplitude slightly increased an artefact in the phantom acquisition, probably due to the higher sensitivity for imperfections in the trajectory and coil sensitivities produced by the pipeline that is tailored to in-vivo data. The chosen wave frequency limited the in-vivo capability of the gradient insert due to auditory limits, but that is reducible by changing frequency and could be completely eliminated by going above 20khz.

Conclusion

Wave-CAIPI significantly reduces the $$$G$$$-factor penalty on accelerated scans. While the high-performance single-axis insert head gradient shows clear improvements of the performance of Wave-CAIPI in the studied sequence, faster acquisitions are now feasonable (with increased R and shorter readouts, such as EPI) and have even larger potential performance benefits.

Acknowledgements

No acknowledgement found.