Introduction

Current commercial multi-nuclei coils suffer from stringent specific absorption rate(SAR) limits, reducing available B₁⁺ RMS below that required for many anatomical and metabolic imaging protocols. For example, a popular commercial dual-tuned proton(¹H)/sodium(²³Na) coil is three times lower with its SAR limit for proton compared to a corresponding proton-only coil. Similarly, SAR limits for the non-proton channel are much lower than for the proton-only coil. In this study, we aimed to characterise our recently proposed 16-channel ¹H/²³Na head coil design¹ by developing realistic RF simulations and validating them to enable SAR level estimations.

Methods

Experimental data collection The coil consists of eight ¹H dipoles and eight ²³Na loops as reported previously¹.(Fig1a,c) Experiments were conducted with a spherical phantom(d=15cm, 150mM NaCl).(Fig.1b) S-matrices were recorded from vector network analyser(VNA)(E5063A ENA, Keysight) on the bench(Fig.2). MR data were acquired on parallel-transmit for proton and single-transmit for sodium on 7T MR scanner(MAGNETOM Terra,Siemens,Germany). The coil was connected to a power splitter and respective TR switches(MR coiltech,UK) for each acquisition. ¹H radiofrequency field(B₁⁺) maps were acquired with TFL sequence²(TR/TE=5000ms/1.55ms,FA=5deg,2.0×2.0x5.0mm³,bandwidth=450Hz/pixel,TA=9s). Sodium B₁⁺ maps were calculated with the double-angle method³ acquired by 2D GRE sequences(TR/TE=150ms/1.92ms,FA=45/90deg,6.6×6.6x25.0mm³,32 averages,bandwidth=600Hz/pixel,TA=2min6s). For both nuclei, two modes were adopted: volume transmit mode where all eight channels are transmitting simultaneously or surface coil transmit mode where only one element of the array is transmitting at a time. In both cases signal was received through all eight channels.
Simulation data collection FDTD simulation (Sim4life 7.2, ZMT, Switzerland) was performed in three setups: phantom-centred for simulation validation, and brain-centred with Duke/Ella human model⁴ for SAR evaluation.(Fig.1d) Phantom results are tuned and matched to similar level as bench measurements with Optenni (OptenniLab5.2, Finland).(Fig.2) Same circuitry was applied to both human model simulations.
B₁⁺ magnitude mapping Experimental B₁⁺ field maps were reconstructed for individual channels from surface coil transmit mode for both nuclei while simulated B₁⁺ maps are directly exported from Sim4Life. Difference maps are calculated for the magnitudes.
B₁⁺ phase mapping The experimental individual receiver sensitivity is characterised from volume transmit mode as $$$R_i = \frac{S_i}{\sqrt{\sum(S_i \cdot S_i^*)}}$$$ where S_i is signal from individual receiver i. For given transmit channel j, the sensitivity $$$T_j = \sum \frac{S_{j,I}}{R_i}$$$ where S_j,i represents the signal acquired when transmitting with channel j and receiving with channel i in surface coil transmit mode. The relative sensitivity for each transmit channel T_rel,j is therefore calculated by taking one transmit channel T1(arbitrary choice) as the reference:$$$T_{rel,j}=\frac{T_j}{T_1}$$$. The relative phase is derived from T_rel,j (by definition channel 1 has zero relative phase). Final experimental phase maps are computed by multiplying relative phases to channel 1 simulated data to enable a straightforward comparison.
SAR calculations Q matrices⁵ were computed for 10g tissue mass-average and SAR_10g was calculated for the adult head. Virtual observations points(VOPs) were derived from the SAR matrices calculation, and maximal SAR_10g values were computed for 150k random RF shim sets. All results were normalised to 1W total input power.

Results

Linear regression of experimental and simulated S-matrices reported R² of 0.92 and 0.82 for ¹H and ²³Na respectively.(Fig.2a) The in-silico coil remained well tuned and matched when switching between different human models.(Fig.2b) The maximum normalized root-mean square error(NRMSE) for individual simulated and measured B₁⁺-field magnitude maps was 9.8% for ¹H and 17.3% for ²³Na. The averaged phase differences between individual phase maps were for 32.4 degrees for ¹H and 24.7 degrees for ²³Na within the area underneath the coil defined by signal intensities.(Fig.3) Human model simulations reported peak SAR_10g of 0.5W/kg for ¹H and 0.55W/kg for ²³Na for Duke whereas it’s 20% lower for ¹H and similar level for ²³Na in Ella.(Fig.4a) The distribution over 150K RF shims indicated 95% of the computed SAR_10g,max values are below 0.55W/kg for ¹H and below 0.8W/kg for ²³Na in the worst case.

Discussion

The 16ch dipole/loop array demonstrated robustness against load-variation when the phantom model was switched to human head models. A good match between B₁⁺ magnitude and phase maps from phantom measurements and corresponding simulations validates the model. Higher SAR_10g was reported for Duke compared to Ella due to the bulkier model. However, the worst case SAR_10g remains within relevant safety limits (i.e.head SAR limit as 3.2 W/Kg). SNR quantification and in-vivo data will be acquired for further assessment.

Conclusion

In this work, we have validated phantom results to characterise a custom-built 16-channel dipole/loop array for 7T MRI. Electromagnetic simulation results aligned well with measurements and established safe RF power limits for in vivo studies. Future development would enable simultaneous ¹H/²³Na imaging to benefit patient diagnosis.

Acknowledgements

This work was supported by King’s China Scholarship Council, by core funding from the Wellcome/EPSRC Centre for Medical Engineering [WT203148/Z/16/Z], Wellcome Trust Collaboration in Science grant [WT201526/Z/16/Z] and by the National Institute for Health Research (NIHR) Clinical Research Facility based at Guy’s and St Thomas’ NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.