Tailored RF pulse optimization for magnetization inversion at ultra high field

Aaron C. Hurley, Ali Al-Radaideh, Bai Li, Uwe Aickelin, Ron Coxon, Paul Glover, Penny A. GowlandarXiv 2010 · OpenReview

Abstract

The radiofrequency (RF) transmit field is severely inhomogeneous at ultrahigh field due to both RF penetration and RF coil design issues. This particularly impairs image quality for sequences that use inversion pulses such as magnetization prepared rapid acquisition gradient echo and limits the use of quantitative arterial spin labeling sequences such as flow-attenuated inversion recovery. Here we have used a search algorithm to produce inversion pulses tailored to take into account the heterogeneity of the RF transmit field at 7 T. This created a slice selective inversion pulse that worked well (good slice profile and uniform inversion) over the range of RF amplitudes typically obtained in the head at 7 T while still maintaining an experimentally achievable pulse length and pulse amplitude in the brain at 7 T. The pulses used were based on the frequency offset correction inversion technique, as well as time dilation of functions, but the RF amplitude, frequency sweep, and gradient functions were all generated using a genetic algorithm with an evaluation function that took into account both the desired inversion profile and the transmit field inhomogeneity.

Introduction

At ultra-high magnetic fields (7 Tesla), the radiofrequency (RF) transmit field ($B_1$) becomes significantly inhomogeneous. This is primarily because the RF wavelength at 7T is comparable to the dimensions of the human head, leading to constructive and destructive interference patterns. These "bright" and "dark" spots pose a major challenge for MRI sequences that rely on precise magnetization inversion, such as MPRAGE for structural imaging and Arterial Spin Labeling (ASL) for perfusion measurement. Incomplete inversion in low-power regions leads to signal drop-outs and poor contrast, particularly in the cerebellum and temporal lobes.

Methods

To address $B_1$ inhomogeneity, the researchers developed a software-based optimization strategy using a genetic algorithm (GA). They evolved a "Time Resampled" Frequency Offset Corrected Inversion (TR-FOCI) pulse. Unlike standard FOCI pulses, the TR-FOCI pulse is defined by 11 parameters that allow the algorithm to optimize amplitude modulation, frequency sweep, and gradient functions simultaneously.

Pulse modulation functions and parameterization for the genetic algorithm.

The GA "breeds" pulses over generations, selecting for those that maximize a fitness function called Weighted Inversion Profile Accuracy (WIPA). This metric evaluates pulse performance across a range of $B_1$ amplitudes (3 to 15 $\mu$T), specifically penalizing failures at low power levels. Evaluation involved numerical Bloch simulations, phantom testing on a 7T Philips Achieva scanner, and in vivo validation in human subjects.

Results

The evolutionary approach successfully identified pulse shapes that are robust to field variations. Simulations demonstrated that the TR-FOCI pulse maintains a near-ideal inversion profile even at RF amplitudes as low as 5 $\mu$T, whereas conventional pulses like HSC and C-FOCI fail significantly below 10 $\mu$T.

Inversion profile and performance metric (IPA) across varying RF amplitudes.

In human subjects, the optimized TR-FOCI pulse significantly improved gray/white matter contrast in the cerebellum, a region typically plagued by low $B_1$ at 7T. Furthermore, these pulses remained within safety (SAR) and hardware (gradient slew rate) limits, making them suitable for clinical implementation.

Conclusion

By prioritizing consistency across a range of power levels rather than peak performance at a single point, the GA-optimized TR-FOCI pulses provide a robust solution for 7T MRI. Future work may address $B_0$ inhomogeneities and the development of a generalized pulse library to reduce the computational overhead of per-scan optimization.

Summary updated

Abstract

The radiofrequency (RF) transmit field is severely inhomogeneous at ultrahigh field due to both RF penetration and RF coil design issues. This particularly impairs image quality for sequences that use inversion pulses such as magnetization prepared rapid acquisition gradient echo and limits the use of quantitative arterial spin labeling sequences such as flow-attenuated inversion recovery. Here we have used a search algorithm to produce inversion pulses tailored to take into account the heterogeneity of the RF transmit field at 7 T. This created a slice selective inversion pulse that worked well (good slice profile and uniform inversion) over the range of RF amplitudes typically obtained in the head at 7 T while still maintaining an experimentally achievable pulse length and pulse amplitude in the brain at 7 T. The pulses used were based on the frequency offset correction inversion technique, as well as time dilation of functions, but the RF amplitude, frequency sweep, and gradient functions were all generated using a genetic algorithm with an evaluation function that took into account both the desired inversion profile and the transmit field inhomogeneity.
BibTeX
@misc{hurley2010tailored,
  author = {Aaron C. Hurley and Ali Al-Radaideh and Bai Li and Uwe Aickelin and Ron Coxon and Paul Glover and Penny A. Gowland},
  title = {{Tailored RF pulse optimization for magnetization inversion at ultra high field}},
  year = {2010},
  eprint = {1004.5051},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1004.5051},
}
Figure