While general-purpose pTx coils mostly have identical elements arranged symmetrically, coils built for a specific application may be asymmetrical. This asymmetry may lead to a high variation in local SAR distributions of individual coil-elements. For an arterial spin labelling pTx coil, we show that using 2,000,000 unconstrained random excitations to characterize the maximum local-SAR underestimates the local-SAR by 7%. We also show that for the investigated ASL coil, the neck elements, which could be exclusively used for labelling, yield 33% less SAR than the head-imaging coils. Therefore, 50% higher power limits can be used for labelling part of an ASL sequence than the imaging part.
Parallel-transmit (pTx) RF coils are used at higher field strengths to mitigate wavelength related artefacts. However, some combinations of weights applied to individual coil-elements may lead to increased local SAR. While general-purpose coils mostly have identical elements arranged symmetrically, purpose-built coils can be asymmetrical (Figure 1). This asymmetry may lead to a high variation in local SAR distributions of individual coil-elements and may make some coils more likely to be used.
Most arterial spin labelling sequences consist of two parts, i) labelling the spins in a lower slice in the neck, and ii) imaging the spins in a higher slab, in the head. To increase labelling efficiency, an ASL-coil was previously built with three elements around the neck, four around the head and one on the crown (Figure 1).
Random RF magnitude/phase combinations are commonly simulated to investigate local SAR, as these are expected to cover most real-life scenarios for a sufficiently large number of excitations (Nexc). However, using random excitations is unlikely to yield numerous cases with several coil-elements turned off, unless Nexc is impractically high.
In this study, we investigate the efficiency of using a random set of excitations to determine the operational SAR limits of an asymmetrical 8-element ASL-coil array.
Electromagnetic simulations of the coil (details given in Figure 1) were performed in Sim4Life (ZMT, Zurich, Switzerland). Virtual observation points (N=3738) were used in local SAR calculations1. Random excitations were created by randomizing the amplitudes ($$$A_{lower}≤amplitudes≤1$$$) and phases ($$$0≤phases<2π$$$) of each coil-element. The amplitudes were then scaled such that the total input power to the whole coil array was 1W.
Scenario 1: Unconstrained random shimming. Random excitations were created for Nexc=255,000 and Nexc=2,000,000 for $$$A_{lower}=0$$$.
Scenario 2: Subset of coils (Constrained random shimming). 1,000 random excitations were created for each subset of coil-elements (one-ON, two coils ON, …, one-OFF, all-ON), yielding a total of Nexc=255,000. Because only the three coil-elements around the neck could exclusively be used for labelling while the remaining five are used for imaging in an ASL sequence, we also investigated these two specific cases separately.
Scenario 3: All-ON. Calculations in scenario 2 were repeated as $$$A_{lower}$$$ was increased from 0 to 1 (phase-only shimming) at increments of 0.1. Instead of turning elements off, their amplitudes were set equal to $$$A_{lower}$$$ while the amplitudes for the selected subset were randomized between [$$$A_{lower}$$$, 1]. Phase was randomized for all elements.
The maximum local SAR was observed as 2.24W/kg (Nexc=255,000) in scenario 1 and as 2.53W/kg in scenario 2 (Figure 2). An increase of 13% was observed in calculated local SAR when elements were explicitly turned off for the same number of excitations. Using Nexc=2,000,000 (scenario 1) still yielded 7% underestimation (2.36W/kg). This difference highlights that unconstrained random excitations may not cover all possible scenarios, unless an impractical number of excitations is used, and may possibly result in an important underestimation of local SAR.
Note that explicitly diverting the power to a subset of coil-elements still does not guarantee that the global worst case is found. While increasing Nexc to 10,000 per subset did not yield any changes in our experience, increasing Nexc further or developing an interval search algorithm that searches around the local SAR maxima might increase the likelihood of finding the global maximum of local SAR.
The crown coil led to 28% higher maximum local SAR when it was used (Figure 3). Consequently, maximum local SAR when the five head coil-elements were used, was equivalent to the overall maximum of 2.53W/kg, while maximum local SAR when only the three neck elements were used with other coil-elements turned off was 1.70W/kg (Figure 4). Labelling can be achieved by using only the neck elements, which yields 33% less maximum local SAR. Therefore, we could define an input power limit that is 50% higher for labelling than for imaging, instead of penalizing the neck coils with the effect of the crown-element by using a single SAR limit. Because labelling yields comparable SAR to imaging during a pseudo-continuous ASL sequence2, this potential SAR-benefit may be used to increase labelling duration or reduce TR at 7T considerably. Also, note that the crown coil does not yield the maximum local-SAR when used alone (Figures 1, 3).
Finally, phase-only shimming (all amplitudes set to 1) is considered a safer alternative to amplitude-and-phase shimming. This is confirmed in Figure 5 where the maximum local SAR is reduced by 42% for $$$A_{lower}=1$$$ compared to unconstrained amplitude-and-phase shimming. However, setting $$$A_{lower}=0.6$$$ also yields a very similar reduction of 41% while allowing more freedom in designing the coefficients across the coil-elements for mitigating wavelength effects.
1. Eichfelder G, Gebhardt M. Local specific absorption rate control for parallel transmission by virtual observation points. Magn Reson Med 2011;66(5):1468-1476.
2. Okell TW, Chappell MA, Kelly ME, Jezzard P. Cerebral Blood Flow Quantification Using Vessel-Encoded Arterial Spin Labeling. Journal of Cerebral Blood Flow & Metabolism 2013;33(11):1716-1724.
3. Andreas C, Wolfgang K, Eckhart GH, Katharina H, Marcel Z, Esra N, Wolfgang R, Rolf J, Werner B, Ji C, Berthold K, Peter S, Hans-Peter H, Jianxiang S, Michael O, Dominik S, Anthony K, Joshua WG, Niels K. The Virtual Family—development of surface-based anatomical models of two adults and two children for dosimetric simulations. Physics in Medicine & Biology 2010;55(2):N23.