Seven-tesla time-of-flight angiography using a 16-channel parallel transmit system with power-constrained 3-dimensional spoke radiofrequency pulse design

Abstract

Objectives: Ultrahigh magnetic fields of 7 T or higher have proven to significantly enhance the contrast in time-of-flight (TOF) imaging, one of the most commonly used non-contrast-enhanced magnetic resonance angiography techniques. Compared with lower field strength, however, the required radiofrequency (RF) power is increased at 7 T and the contrast obtained with a conventional head transmit RF coil is typically spatially heterogeneous.In this work, we addressed the contrast heterogeneity in multislab TOF acquisitions by optimizing the excitation flip angle homogeneity while constraining the RF power using 3-dimensional tailored RF pulses ("spokes") with a 16-channel parallel transmission system and a 16-channel transceiver head coil.

Materials and methods: We investigated in simulations and in vivo experiments flip angle homogeneity and angiogram quality with a same 3-slab TOF protocol for different excitations including 1-, 2-, and 3-spoke parallel transmit RF pulses and compared the results with a circularly polarized (CP) phase setting similar to a birdcage excitation. B1 and B0 calibration maps were obtained in multiple slices, and the RF pulse for each slab was designed on the basis of 3 calibration slices located at the bottom/middle/top of each slab, respectively. By design, all excitations were computed to generate the same total RF power for the same flip angle. In 8 subjects, we quantified the excitation homogeneity and the distribution of the RF power to individual channels. In addition, we investigated the consequences of local flip angle variations at the junction between adjacent slabs as well as the impact of ΔB0 on image quality.

Results: The flip angle heterogeneity, expressed as the coefficient of variation, averaged over all volunteers and all slabs could be reduced from 29.4% for CP mode excitation to 14.1% for a 1-spoke excitation and to 7.3% for 2-spoke excitations. A separate detailed analysis shows only a marginal improvement for 3-spoke compared with the 2-spoke excitation. The strong improvement in flip angle homogeneity particularly impacted the junction between adjacent TOF slabs, where significant residual artifacts observed with 1-spoke excitation could be efficiently mitigated using a 2-spoke excitation with same RF power and same average flip angle. Although the total RF power is maintained at the same level than that in CP mode excitation, the energy distribution is fairly heterogeneous through the 16 transmit channels for 1- and 2-spoke excitations, with the highest energy for 1 channel being a factor of 2.4 (1 spoke) and 2.2 (2 spokes) higher than that in CP mode. In vivo experiments demonstrated the necessity for including ΔB0 spatial variations during 2-spoke RF pulse design, particularly in areas with strong local susceptibility variations such as the lower frontal lobe.

Conclusions: Significant improvement in excitation fidelity leading to improved TOF contrast, particularly in the brain periphery, as well as smooth slab transitions can be achieved with 2-spoke excitation while maintaining the same excitation energy as that in CP mode. These results suggest that expanding parallel transmit methods, including the use of multidimensional spatially selective excitation, will also be very beneficial for other techniques, such as perfusion imaging.

Figure 1

Excitation k-space trajectory (top) and corresponding scheme of the sequence's excitation part for 1- and 2-spoke excitation. The energy in k-space is deposited on the blue solid lines while the dotted blue lines denote the gradient blips for the 2-spoke excitation. The RF sub-pulse duration of the 2-spokes is set to 1ms, thus the total RF pulse duration is identical to the 1-spoke RF pulse.

Figure 2

a) Locations of the 3 overlapping TOF slabs and the corresponding calibrations slices 1-7 used for B₁⁺ and B₀ mapping. Slices #3 and #5 are shared for the optimization of two adjacent TOF slabs, respectively. b) transverse view of the sum of magnitude of the 16 B₁⁺ maps including the ROIs w_i used for optimization. The ROI w₁ for calibration slice #1 was not used for optimization (see Discussion). c) B₁⁺ magnitude maps of the individual TX channels for calibration slice #6

Figure 3

a) L-curve plot demonstrating the tradeoff between the two regularization terms in equation 1, the energy term ∥b∥² and the inhomogeneity term ${\Vert \mid Ab\mid -\mid m\mid \Vert }_w^2$. The black and red curves show the results for a least-squares optimization with 1 and 32 different target phases, respectively, while the green and blue curves plot the results for the MLS optimization. b) L-curve and scatter plots for the solution of 1- and 2-spoke MLS optimization using 32 different target phases without (ε=1) and with (ε=mean(|Ab|/mean(|m|)) correction to obtain the same target flip angle for all solutions. The green and red crosses indicate the individual solutions for the 2-spoke excitation having different excitation trajectories and/or λ values and the solid lines indicate the edge for the optimal solutions. c) L-curve and scatter plots showing the tradeoff between total (normalized) energy E_n and CV for 1-, 2- and 3-spoke excitation (blue, red and green) as well as for the CP mode (black cross). The horizontal line indicates solutions having the same total E_n as the CP mode, which was the design criteria for this study.

Figure 4

a) Mean CV value and standard deviation from 7 subjects for the CP mode and for the individually optimized slabs for 1- and 2-spoke excitation. b) CV ratio, i.e. the CV values for the 1-spoke divided by the CV values for the 2-spoke excitation for each subject and each slab. The dashed lines show the average ratio value. c) Normalized voltage (top) and normalized energy E_k (bottom) for each of the 16 TX channels for 1- and 2-spoke excitation of the top slab in subject 4 normalized by the CP mode voltage (dashed line). d) Box plot for the maximum energy per channel (E_k,max) of all subjects split for the individual slab. The box indicates the median and the 25th and 75th percentile, crosses denote values (outliers) outside of 2.7*σ if a normal distribution is assumed and the whiskers extend to the extreme values which are not considered as outliers.

Figure 5

a) Maximum intensity projections (MIP) for axial, coronal and sagittal views for CP mode, 1-spoke and 2-spoke excitation. b) Corresponding flip angle maps estimated by Bloch simulations for a nominal flip angle of 20°. The black ROIs denote the regions used for optimization and voltage adjustment. As shown for the 1-spoke and 2-spoke solutions, the slices #3 and #5 are used for the optimization of 2 adjacent TOF slabs respectively. The red boxes indicate the flip angle map locations used in Figure 6.

Figure 6

Coronal native images and maximum intensity projections (MIP) for the same subject as in Figure 5 with 1-spoke (a) and 2-spoke (b) excitation. The right column shows flip angle maps estimated of calibration slice #5 which is used for the optimization of both, the top slab and the middle slab (compare red boxes in Figure 5). Note that the color scale of the flip angle, which is given in percent of the 20 degree target flip angle, is different compared to Figure 5.

Figure 7

Transversal and sagittal views of native TOF images and corresponding 15-mm-thick MIP images for CP mode, 1-spoke and 2-spoke excitation. As indicated by the arrow, the fat signal is reduced for the 2-spoke excitation.

Figure 8

Calculated flip angle map for the bottom slab in subject 8 for 1-spoke (top and middle row) and 2-spoke excitation (bottom row). In this subject a sub-optimal excitation showing a strong local flip angle void in the center of the slices (top row) could only be achieved for the 1-spoke excitation with the constraint of having the same total energy (100%) as the CP mode. A suitable solution, that successfully removed this local flip angle hole, was only found if the excitation energy was increased to 135% (middle row). In contrast, this effect was not observed for the 2-spoke excitation using 100% energy (bottom row).

Figure 9

a) Off-resonance frequency maps acquired at the 7 calibration slice positions showing strong off-resonance in the lower frontal and the lower temporal lobe. b) Native images (left) and 15-mm-thick maximum intensity projections (right) in sagittal view with CP mode, 1-spoke and 2-spoke excitation. The 2 spoke calculation was performed without (3^rd row) and with (4^th row) including B₀-induced off-resonances into the RF pulse design. This has a significant impact on the images and angiograms in the lower frontal lobe (yellow arrows). c) Corresponding transverse views for the 4 different excitation settings, showing a slice located at calibration slice position 2 (see white box in subfigure a). d) Bloch simulations for the four different settings at the location of calibration slice position 2 and 3.