Synopsis

Keywords: Vessels, RF Pulse Design & Fields

The incomplete or time-inefficient suppression of fat signals is an unsolved issue in 3D cardiovascular MR imaging (CMR). We present a flexible framework named OptiPulse to design spectrally-selective radiofrequency (RF) pulses using numerical optimization of Bloch equation simulations. The RF pulse performance was assessed using a composite loss function for both fat suppression and power requirement. OptiPulse was used to design a B-splines-interpolated pulse for use in a free-running 3D whole-heart pulse sequence. Its fat suppression performance was ascertained both in vitro and in vivo, where it resulted in more homogeneous fat suppression when compared to other WE pulses (P<0.01).

Background

In radial 3D imaging of the heart at 1.5T, fat suppression methods are necessary for the correct delineation of the anatomy, but they are often technically challenging while they also need to be time-efficient to limit the scan duration.

Recently, short water-excitation (WE) radiofrequency (RF) pulses have been reported as an efficient solution for fat suppression in high-resolution imaging of the heart^1-3. These WE RF pulses were designed using analytical methods and may require manual tuning of the parameters. For this reason, their optimization is constrained to low-dimensional (<5) parameter space, which may limit the search to a small domain of solutions while a more global optimum may be overlooked.

We therefore propose a flexible and efficient framework of optimizing B-splines-interpolated⁴ WE RF pulses, named OptiPulse. The goal of this study was thus to design an RF pulse with OptiPulse, to compare its fat suppression to a standard non-suppressive pulse, a 1-1 binomial WE pulse and the LIBRE 1-1 WE pulse³ in a phantom, and to validate it in vivo in the heart.

Methods

The optimization of B-splines-interpolated⁴ RF pulses was performed with 30 parameters (15 phase, 15 amplitude). The frequency response in the transverse plane was computed using Bloch equations simulation of 400 isochromats ranging from -1kHz to 1kHz (Fig.1). Next, the performance of the pulse was evaluated with a tailored loss function L:

$$ L(P) = \frac{[||M_{xy} ||_2]_f}{[||M_{xy}||_2]_w} + \frac{\lambda_{1}}{[||M_{xy}||_2]_w} + \lambda_{2}[(||M_{xy}||_2)^{2}]_f, $$

Where $$$\lambda_{1}=0.01$$$ and $$$\lambda_{2}=400$$$ are the regularization coefficients, $$$||M_{xy}||_2$$$ is the L2 norm of the transverse magnetization and $$$[…]_w$$$ and $$$[…]_f$$$ are the averages over the water and fat bands. Finally, the loss function was minimized with the Covariance Matrix Adaptation Evolution Strategy CMA-ES⁵, a black-box optimizer.

A WE RF pulse with a duration of 2.56ms was optimized and implemented on a 1.5T clinical scanner (MAGNETOM Sola, Siemens Healthcare) as part of a free-running 3D radial bSSFP research sequence⁶ (α=70°, TR=5.4ms). The fat suppression of the optimized RF pulse was compared to that of a 0.3ms non-fat-suppressing pulse, to that of a 2.6ms 1-1 WE and to that of a 2.6ms LIBRE 1-1 pulse in an agar-oil phantom by measuring the contrast-to-noise ratio (CNR) between the blood and fat phantom compartment.

Finally, free-running 4D bSSFP images with 3D translational correction of respiratory motion using fNAV⁷ were reconstructed in the heart of 8 healthy volunteers with the 1-1 WE pulse, the LIBRE pulse, and the optimized RF pulse (voxel size (1.1mm)³, duration 11min). The contrast ratios (CR=S_fat/S_blood) between both chest fat/epicardial fat and the left-ventricular blood pool were calculated since the noise can no longer be reliably measured, while the traceable length of the right coronary artery (RCA) and the left anterior descending artery (LAD) was measured⁸ for the three water-selective acquisitions. All metrics were tested against those obtained with the optimized pulse using a paired Student’s t-tests including a Bonferroni correction for multiple comparisons.

Results

The numerical simulations of the RF pulses showed a slightly broader suppression band for the optimized pulse than for the other pulses (Fig.2E). In the phantom, blood-fat CNRs were -161 (i.e. fat brighter than blood), 72, 81 and 84 for the non-suppressing, the 1-1 WE, the LIBRE and the optimized pulse, respectively (Fig.2F-I).

In vivo, the blood-pool/chest-fat CR amounted to 4.2±1.4 for the optimized pulse, 2.0±0.6 for the 1-1 WE pulse (P=0.005) and 2.8±0.9 for the LIBRE pulse (P=0.04) (Fig.3). The blood-pool/epicardial-fat CR of the optimized pulse at CR=3.8±1.3 was higher than that of the 1-1 WE pulse at CR=2.0±1.1 (P=0.004) and similar to CR=2.4±1.1 (P=0.016)obtained with the LIBRE pulse.

The RCA could be traced for a length of 5.8±1.0cm with the optimized pulse, 4.9±0.8cm (P=0.28) with the 1-1 WE pulse and 5.0±0.6cm (P=0.28) with the LIBRE pulse. Finally, the LAD could be traced for 8.7±1.3cm, 7.0±0.9cm (P=0.03) and 7.5±1.2cm (P=0.09) for the optimized pulse, the 1-1 WE and the LIBRE RF pulses respectively. In multiple volunteers, the more homogeneous fat suppression of the epicardial fat signal achieved with the optimized pulse led to better visualization of the proximal and distal parts of the LAD (Fig.4).

Discussion

The RF pulse generated with the OptiPulse framework demonstrated higher levels of fat suppression than the 1-1 WE and the LIBRE pulse in both the phantom and in vivo studies.

This was consistent with the measurement of the vessel length, which was hampered by residual strong fat signal in the 1-1 WE images. These results suggest that using a broader suppression band is beneficial for whole-heart imaging where magnetic field inhomogeneities are high.

While OptiPulse requires further validation in a large patient cohort, these preliminary results suggest both chest and epicardial fat can be simultaneously well suppressed in 3D radial whole heart imaging with OptiPulse, potentially leading to improved visualization of the coronary arteries.

Conclusion

In conclusion, we developed an optimization framework called OptiPulse to generate a B-spline-interpolated water-selective RF pulse. The optimized pulse resulted in more homogeneous lipid suppression in phantoms and human hearts at 1.5T when compared to other water-excitation pulses.

Acknowledgements

No acknowledgement found.