Introduction

Ultra-high field (UHF) strength MRI enables quantitative mapping of the longitudinal relaxation rate (R₁ = 1/T₁) at sub-millimetre resolutions in vivo, in turn furthering our understanding of brain microstructure¹. A drawback is the poorer RF excitation uniformity which significantly biases R₁ estimates derived from dual flip angle measurements. Parallel transmission (pTx) offers the potential to improve B₁⁺ homogeneity in the imaging volume and therefore to reduce spurious spatial variance in R₁ estimates.

We investigated whether exploiting pTx by replacing the excitation pulse with optimized kT-points RF pulses would reduce B₁⁺ related biases in R₁ estimates to a level that could be addressed by post-hoc correction schemes, which are widely used at lower field strengths to remove residual biases, and do not require additional scans to map the B₁⁺ inhomogeneity.

Methods

A healthy volunteer was scanned with a 7T MRI system (MAGNETOM Terra, Siemens Healthineers, Germany) equipped with a 8Tx/32Rx head RF coil (Nova Medical, Wilmington, USA). The RF system was operated in two frameworks: (a) with the coil transmit elements in a fixed magnitude/phase relationship (approximating CP mode) using conventional non-selective sinc-shaped excitation pulses with a bandwidth-time product of 6, and (b) with dynamic pTx pulses using the kT-points trajectory. Time and SAR constraints allowed 5 kT-points: one was placed in the centre of the excitation k-space and the position of the others was optimized on the kx-ky plane offline. The RF magnitude and phase were optimized running an online optimization directly before the main scan. The pulse optimization process runs without user intervention, suiting the practical workflow of a neuroimaging center. The online optimization including acquiring B₁⁺ and B₀ maps took less than two minutes. The 5 kT-points pulse with a flip angle of 23 degrees is shown in Figure 1.

In each mode, an in-house built sequence was used to acquire 3D multi-echo FLASH images with both PD (nominal flip angle 8 degrees) and T1 (23 degrees) weightings. The other parameters were: 1 mm isotropic resolution (matrix size 224 x 256 x 176), TR 24.1 ms, eight evenly spaced gradient echoes (TE 2.5 .. 18.2 ms), GRAPPA acceleration factor 2 in both 2D and 3D phase encoding directions, gradient and RF spoiling applied, leading to an acquisition time per volume of 4.5 minutes.

Images were processed using the hMRI toolbox (http://hmri.info ²) within SPM (https://www.fil.ion.ucl.ac.uk/spm) and MATLAB (MathWorks, Natick, USA), modified to avoid the small-angle approximation, to first yield R1 maps without B1-correction. We then applied a unified-segmentation-based correction (UNICORT)³ which estimates B₁⁺ inhomogeneity and produces corrected R₁ maps without requiring measured B₁⁺ field maps.

Results

Without the application of optimized pTx pulses, signal dropouts were observed in weighted FLASH images (Figure 2) leading to severe artifacts in the uncorrected R₁ maps (Figure 3, (a)), e.g., in the posterior cerebellum and some areas of the temporal lobes. Using the optimized pTx pulses led to a marked improvement in the artifact level (Figure 3, (b)) and to a reduction in the artifactual spread of R₁ values. Within a white matter mask, the span of uncorrected R₁ values (25th-75th quartile, Figure 4) decreased from 0.65 s^-1 to 0.28 s^-1, implying a substantial improvement in B₁⁺ homogeneity. The combination of pTx with UNICORT (Figure 5) yielded rather uniform R₁ maps, although R₁ values were clearly underestimated.

Discussion and Conclusions

B₁⁺ inhomogeneities at 7T were reduced to the range where bias corrections are known to perform well at lower field strengths (<= 3T). In this first implementation, we employed a post-processing method (UNICORT) for reducing residual bias, since highly accurate and precise B₁⁺ mapping methods required for quantitative mapping are still under investigation (flip angle errors enter R₁ calculations quadratically³) and can lead to substantial error propagation when sub-optimal⁴. The UNICORT method does not account for global B₁⁺ offsets and thus led to underestimation of R₁, which may be addressed in the future by additional reference scans. Further areas of investigation are inadvertent MT effects⁵ and interaction with physiological noise.

Acknowledgements

The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement no. 616905. The Wellcome Centre for Human Neuroimaging is supported by core funding from Wellcome [203147/Z/16/Z].