Introduction

Many diseases such as cancer, diabetes or neurodegeneration involve critical changes to metabolism.¹ Magnetic resonance spectroscopy (MRS) and spectroscopic imaging (MRSI) are powerful non-invasive tools to probe metabolism in vivo. The new method of deuterium metabolic imaging (DMI) has been proposed to track glucose uptake and metabolism into glycolysis and the TCA cycle. De Feyter et al² showed that DMI with a [6,6'-²H₂]-glucose substrate could offer an alternative to ¹⁸FDG-PET. DMI can attain 1 mL nominal spatial resolution at 7T.³ Our aim is to construct a high-performance deuterium head coil with proton capability for localisation and shimming. Safe RF power limits, B₁⁺ performance and SNR of this new coil are assessed in this work.

Methods

Coil design
The coil comprises a 15-element transverse electromagnetic mode (TEM) resonator for ²H transmit, a 16-element loop array for ²H receive, and a 4-element loop array for ¹H transmit and receive. The coil is shown in Figure 1 together with a 3D rendering of the ²H and ¹H transmit structures. It was built by Virtumed LLC.

Electromagnetic field simulations
Simulations were performed in CST Studio Suite 2021 (Dassault Systemes). Specific absorption rate (SAR) and B₁⁺ fields were computed for a saline phantom (conductivity 1.0 S/m) and for the Laura and Gustav human voxel models. The ¹H and ²H transmit elements were simulated separately, since with the significant difference in the resonance frequencies, they have negligible coupling. B₁⁺ field histograms were computed based on voxels with “brain” material type in the voxel models.

Phantom studies
An agar phantom was prepared in a cylindrical phantom (Fisher) comprising water enriched to 2% abundance for HOD (singly-labelled water), agar, sufficient NaCl to yield a conductivity of 1.0 S/m⁴ and a T₁ relaxation agent (Dotarem) to target T₁=1100 ms for protons at 7T. Images were collected using the GRE pulse sequence (voltages from 20-200 V, pulse length 20 ms, T_R 900 ms, T_E 10ms, 3 averages, total acquisition time 11’33’’) with 8 slices over a 32×32 matrix. Noise was assessed from an ROI drawn outside the phantom and hence imaging signal-to-noise ratios were calculated. A 3D UTE CSI sequence with a spatial resolution of 12.5× 12.5× 25 mm³ was run. Spectra were fitted using AMARES as implemented in the OXSA toolbox⁴.
B₁⁺ maps were obtained by non-linear least-squares minimizations to the FLASH signal equations,
$$I(V)=\frac{a (\sin bV)(1-e^{-T_R/T_1})}{(1-\cos bV)e^{-T_R/T_1}},$$
where a is a complex number proportional to the receive sensitivity, and b is the flip angle per volt at that pixel (i.e. the transmit field strength). Maps of b were then converted to B₁⁺ in Hz per volt by multiplication of the number of degrees flip angle per Hz calculated for the rectangular pulse used in the GRE acquisitions.

Results

SAR
The maximum SAR-10g was 0.200 kg^-1 (Laura) and 0.352 kg^-1 (Gustav) for ²H, and 0.508 kg^-1 (Laura) and 0.432 kg^-1 (Gustav) for ¹H. The heating profiles are shown in Figure 2.

B₁ maps
Figure 3 shows the simulated B₁⁺ field distribution over the human voxel model, normalised to 1 W input power without accounting for losses. At the centre of the brain, we predict 0.77 uT/sqrt(W) B₁⁺ field strength for ¹H and 0.54 uT/sqrt(W) for ²H.
Figure 4 shows measured phantom B₁⁺ maps in comparison to an equivalent slice in the simulated model.
Our fitted ²H T₁ was 101ms. This compares to ²H T₁=409 ms report at 11.7 T by de Graaf et al.³ Our value is likely due to both the efficient quadrupolar relaxation of deuterium, and to having added Dotarem.

Signal to noise ratio
For the 120V GRE scan acquired in 11’33’’, the average SNR across all slices is 48.7 for voxels within the masked region (see Figure 5(a)). This is sufficient for clear ²H imaging and promising for highly resolved ²H-MRSI applications in the brain. Peak SNR was >200 in peripheral voxels. We see evidence of some B₁⁺ shading in the centre of the phantom.

Spectroscopy
Figure 5(c) shows an example spectrum from the phantom. There is a single clear HOD peak as expected. Taking 8 central voxels, the linewidth was 9 ± 4 ppm (mean±SD).

Discussion and conclusions

Our TEM ²H/¹H coil design delivers homogeneous B₁⁺ fields at both the ²H and ¹H 7T frequencies. Phantom scans and EM modelling results are in good agreement, allowing us to recommend appropriate k­-factors of 1.83 kg^-1 for ²H and 2.96 kg^-1 for ¹H for human in vivo scans. We expect to perform in vivo scans imminently.