9.4 T double‐tuned 13C/1H human head array using a combination of surface loops and dipole antennas

MRI at ultra‐high field (UHF, ≥7 T) provides a natural strategy for improving the quality of X‐nucleus magnetic resonance spectroscopy and imaging due to the intrinsic benefit of increased signal‐to‐noise ratio. Considering that RF coils require both local transmission and reception at UHF, the designs of double‐tuned coils, which often consist of several layers of transmit and receive resonant elements, become quite complex. A few years ago, a new type of RF coil, ie a dipole antenna, was developed and used for human body and head imaging at UHF. Due to the mechanical and electrical simplicity of dipole antennas, combining an X‐nucleus surface loop array with 1H dipoles can substantially simplify the design of a double‐tuned UHF human head array coil. Recently, we developed a novel bent folded‐end dipole transceiver array for human head imaging at 9.4 T. The new eight‐element dipole array demonstrated full brain coverage, and transmit efficiency comparable to that of the substantially more complex 16‐element surface loop array. In this work, we developed, constructed and evaluated a double‐tuned 13C/1H human head 9.4 T array consisting of eight 13C transceiver surface loops and eight 1H transceiver bent folded‐end dipole antennas all placed in a single layer. We showed that interaction between loops and dipoles can be minimized by placing four 1H traps into each 13C loop. The presented double‐tuned RF array coil substantially simplifies the design as compared with the common double‐tuned surface loop arrays. At the same time, the coil demonstrated an improved 1H longitudinal coverage and good transmit efficiency.


| INTRODUCTION
Magnetic resonance spectroscopy (MRS) and imaging (MRSI) using nuclei other than protons ( 1 H) (X-nuclei), ie 13 C, 31 P, 23 Na etc, provide valuable tools for biomedical research. However, due to a generally much lower gyromagnetic ratio and commonly smaller natural abundance (eg $1% for 13 C), these methods suffer from low signal-to-noise ratio (SNR). Thus, MRI at ultra-high field (UHF, ≥7 T), provides a natural strategy for improving the quality of X-nucleus MRS and MRSI due to the intrinsic benefit of increased SNR.
RF coils designed for X-nucleus MRS and MRSI often need to be double-tuned (DT). By double-tuning an RF coil we imply a capability of the entire RF coil structure to resonate and provide transmission and reception at two different frequencies, ie 1 H and X-nucleus frequencies. An additional ability to transmit and receive at the higher 1 H frequency is necessary for providing static magnetic field (B 0 ) shimming, spatial localization, and proton decoupling and/or application of polarization techniques in the case of 13 C nuclei. Proton decoupling as well as nuclear Overhauser enhancement can also be beneficial for other X-nuclei such as 31 P. Considering that RF coils require both local transmission and reception at UHF, the designs of DT coils, which often consist of several layers of transmit (Tx) and receive (Rx) resonant elements, 1-6 become quite complex. To simplify the DT coil development, usually the major design idea is to preserve high SNR and Tx efficiency (B 1 + /√P, where P is the RF input power) at the X-nucleus frequency, while the performance of the 1 H part of the coil is compromised intentionally. However, the quality of 1 H imaging is often equally important for many applications. In addition, the ability to use the same coil for a comprehensive examination, including both X-nucleus and 1 H scanning, without moving the subject to replace the DT coil with a single-tuned (ST) 1 H coil, is an important step toward clinical acceptance of X-nucleus MRS and MRSI.
There are two major options in designing a DT RF coil (or an array), ie double-tuning the same physical coil structure [7][8][9] (or each element of the array 1,6 ) to two different frequencies simultaneously or arranging two separate ST coils (or coil arrays) resonating at different frequencies close to each other. 1,[3][4][5]10 The first method allows optimization of the coil mainly at one frequency (commonly X-nuclei) while the Tx performance and SNR at the other frequency (commonly 1 H) is degraded. 1,7 The second method, ie combining two separate ST coils (arrays), allows for independent optimization of the DT coil at both frequencies. At the same time, the entire design becomes more complex.
Recently, we presented a DT transceiver (TxRx) 31 P/ 1 H 20-loop (10 loops at each frequency) array coil for human head imaging at 9.4 T. 11 By making the entire coil design relatively simple and limiting the number of 31 P elements to 10, we were able to place both the ST 31 P array and the ST 1 H array in the same layer and at the same distance to the head, which provided high loading and, thus, a good Tx efficiency for both arrays.
As demonstrated previously, moving one of the arrays (eg 1 H) further away from the sample into a second more remote layer degrades its Tx performance and SNR. 1 It is also important that a relatively low number of elements (10 at each frequency) was sufficient to preserve high central SNR. It is known that increasing the number of smaller surface loops in a human head Rx array improves mostly the peripheral SNR and parallel imaging performance, while the central SNR does not substantially change [12][13][14] or is even degraded due to insufficient loading. 11 For non-proton applications, intrinsically low SNR largely limits parallel imaging acceleration in any case, and highly inhomogeneous reception profiles of highdensity Rx arrays are not favorable because of a lack of normalization or quantification standards correcting for it. Hence, a smaller number of larger TxRx elements with deeper penetration and good coverage represent a better compromise for non-proton and DT coils.
In addition to high Tx efficiency and SNR, a well designed state-of-the-art human head RF coil has to provide for longitudinal (along the magnet axis) coverage of the whole brain. This, however, is difficult to realize at UHF for 1 H arrays using a single row of loops due to shortening of the RF wavelength (ie close to 100 mm at 400 MHz), and was demonstrated mainly by using substantially more complex multi-row (ie two rows) loop arrays 1,15-19 capable of 3D RF shimming. In addition, multi-row arrays improve parallel (head-to-feet direction) and multi-band imaging.
However, combining such a multi-row 1 H array with an X-nucleus array into a single layer would make the design too complex, and to the best of our knowledge has never been presented. For example, to improve the 1 H longitudinal coverage and still keep the DT coil relatively simple and robust, in the above described design of the 31  and 230 mm in height (anterior-posterior). The holder was tapered at the superior location to improve fitting to a human head and measured 155 mm in width and 185 mm in height at the very top. In addition, the geometric arrangement and length of the 13 C surface loops (ie 175 mm) was similar to the previously published 31 P array, 11 while the 1 H dipole array was very similar to our previously published ST 1 H bent folded-end dipole TxRx array. 31 All surface loops and dipoles were constructed using 1.5 mm copper wires. The choice of 1.5 mm wire was mostly determined by its mechanical properties. 1.5 mm wire is sufficiently thick to be mechanically stable, and still can be relatively easily formed manually. location. As demonstrated previously for 7 T (Reference 32) and 9.4 T33 coil designs, such an extension provides for an enhancement of the RF field in this area and thus improves the longitudinal coverage. To further improve the coverage, 31,34 we also added a flat local elliptical RF shield (175 mm Â 140 mm) placed 30 mm away from the head (Figure 1).
To minimize coupling between non-adjacent 13 C loops 35 and to efficiently decouple adjacent 1 H dipoles, 36 the array was shielded with a cylindrical RF shield placed at a 40 mm distance from the surface of the loops. The distance to the shield was chosen based on previously published data for UHF human head size arrays. 11,31,36 As we showed previously, 35 the presence of the RF shield significantly decreases coupling between non-neighboring surface loop elements of the array. This effect was demonstrated at both lower (124 MHz) and higher (300 MHz, 400 MHz) frequencies. Since decoupling of non-adjacent distantly located loops is hindered by the necessity of using long cables or wires, which have to be properly routed to avoid introduction of additional coupling, this helps to improve overall decoupling of the Tx loop array and optimize its performance. In addition, the RF shield placed close to the folded portion of the folded-end dipoles produces a mutual inductance between adjacent dipoles and compensates their intrinsic capacitive coupling. 36 The RF shield also helps to avoid interaction of the coil elements with surrounding electronic and metallic structures in the magnet bore including the gradient coils.

| EM simulations
Before constructing the array, we evaluated the new design with respect to performance and safety using numerical EM simulations. EM simulations were performed using CST Studio Suite 2019 (Dassault Systèmes, Vélizy-Villacoublay, France) and the time-domain solver based on the finiteintegration technique. In simulations of the DT 1 H/ 13 C array, we used three voxel models: a head and shoulder (HS) phantom ( Figure 1A), which was constructed to match tissue properties (ε = 58.6, σ = 0.64 S/m at 400 MHz), 37 and Virtual Family multi-tissue models, "Duke" ( Figure 1C) and "Ella", 38 cropped at chest level. For all voxel models, we used an isotropic resolution of 2 mm. In addition, we used an elliptical (140 mm width, 180 mm height, 200 mm length) phantom rounded at the top and filled with ethylene glycol. Simulations performed using the HS phantom and elliptical phantom were used for experimental comparison against the scanner-based evaluation of the 1 H and 13 C arrays, respectively.
To evaluate the interaction between 13 C and 1 H arrays and its effect on RF EM field distributions at both frequencies, the CST model of the DT array coil ( Figure 1) included all 16 elements, ie eight 13 C loops and eight 1 H dipoles. For comparison, we also simulated corresponding ST eight-element loop and dipole arrays at both frequencies. In addition to minimizing coupling between the array elements, 35 the presence of the RF shield can alter the Tx efficiency of the RF array coil. This effect is especially critical at the lower 13 C frequency due to poorer loading. Therefore, to evaluate the effect of the RF shield on the 13 C array performance, we simulated an unshielded version of the ST 13 C array coil. In all simulations we also used a large copper cylinder (640 mm in diameter and 1600 mm in length), which mimicked the RF shield of the gradient coil. As mentioned above, the presence of X-nucleus loops resonating at a substantially lower frequency can still strongly affect the performance of 1 H dipole arrays. 4 The common way to minimize the interaction of the X-nucleus and 1 H elements of a DT array at the 1 H frequency (400 MHz) is to introduce 400 MHz band stop filters (resonant 1 H traps) into the X-nucleus coil elements. In the case of stronger interaction, several traps have to be distributed along the X-loop length. 1,11 Therefore, we systematically evaluated how many 1 H traps need to be introduced into each 13 C loop to minimize alteration of the 1 H dipole B 1 + distribution and peak local specific absorption rate (SAR) value. We compared the performance of the final version of the DT 13 C/ 1 H array (four 1 H traps in each 13 C loop) with that with the same design but with only two 1 H traps. Placement of the traps is discussed below in the Section 2.3. In contrast, the performance of the X-nucleus portion of the DT coil often is not compromised by the presence of the 1 H coil, and X-nucleus traps are not required to be placed into the 1 H structure. [3][4][5] We still verified this below by numerically comparing the performance of corresponding 13 C ST and DT array coils.
B 1 + field profiles and local SAR 10g (averaged over 10 g of tissue) maps were calculated for 1 W of forward stimulated RF power at the array input and then compared with experimentally measured data. Additionally, we evaluated the Tx efficiency, both as ⟨B 1 + ⟩/√P, where P is forward RF power measured at the array input (ie feed-points of the antennas/loops), and as ⟨B 1 + ⟩/√pSAR 10g (pSAR is the peak local SAR value), ie the safety excitation efficiency (SEE). 18 The B 1 + values were averaged over a 130 mm central transversal slab, which included the majority of the human brain.

| Array construction
After EM modeling, we constructed the new 13 C/ 1 H DT array ( Figure 1B). The geometry of the array holder and all elements were the same as described in Sections 2.1 and 2.2 ( Figures 1A and 1C). Dipole and loop elements were placed on the surface of a fiberglass array holder with the wall thickness of 3 mm. As in the simulation model ( Figure 1B), the array was shielded. The RF shield was constructed from a double-sided Kapton (25 μm thickness) copper clad laminate (AKAFLEX, Krempel, Vaihingen/Enz, Germany) as previously described. 39 Figure 2A shows a schematic of a single 1 H dipole element including tuning capacitors (C tune ) (a), the matching capacitor (b), a DT cable trap 40 (c), and a home-built transmit/receive (T/R) switch circuit 33 (d). Since the dipole length was slightly larger than half of the wavelength, to tune the dipoles we used capacitors ($30 pF) instead of inductors. Low-noise preamplifiers (WanTcom, Chanhassen, Minnesota, USA) were incorporated into the respective T/R switch circuits for both 1 H and 13 C elements of the coil. Based on EM simulation data, we used variable matching capacitors of 20 pF. This was sufficient to provide matching on the HS phantom and various human heads. Following the previous work, 31,36 the length of the fold and height of the 1 H dipoles (Figure 2A) measured 30 mm and 33 mm, respectively. Figure 2B shows a schematic of a single 13 C surface loop including the tuning variable capacitor (a) (Johanson, Boonton, New Jersey, USA), matching capacitor (b), DT cable trap 40 (c), and home-built T/R switch circuit (d) integrated with a preamplifier 33 as well as four 1 H traps. 1 H resonance traps were placed on every side of the loop and distributed relatively uniformly along its length to minimize coupling between 13 C loops and 1 H dipoles. Adjacent 13 C loops were decoupled by overlapping the loops. 20,41 This also allowed us to bring the nearest non-adjacent loops (ie 1 and 3, 2 and 4 etc) closer to each other and utilize transformer decoupling to minimize crosstalk between them as suggested previously. 42,43 Optimal overlapping ($15%) was evaluated numerically and further adjusted (within 5 mm) during construction to minimize interaction between neighboring 13 C loops.  13 C experiments with proton decoupling is application of a very high (in a kW range) RF power at the 1 H frequency to the DT coil during 13 C reception. [44][45][46] To avoid saturation of the 13 C preamplifiers, a very high (À80 to À100 dB) isolation between the preamplifier inputs and the Tx input of the 1 H array is required. To provide such a high isolation, in addition to minimizing coupling between individual 1 H dipoles and 13 C loops using 1 H traps ( Figure 2B), we placed the entire 13 C eight-channel interface, which included the 13 C eight-way splitter and eight 13 C T/R switch boards (integrated with preamplifiers), in a separate box outside of the array holder. The 13 C array was connected to this interface box using an eight-channel multi-modular coaxial connector (ODU, Mueldorf, Germany). In addition, at the input of each 13 C T/R switch we introduced a second order 400 MHz band stop filter providing about À50 dB isolation. All 13 C T/R switches, preamplifiers, and filters were shielded. Finally, transmitting 400 MHz high power amplifier during 13 C reception may inject noise at the lower 13 C frequency. To avoid such noise injection, we introduced a second order 100 MHz band stop filter at the 400 MHz array Tx input. The filter provided better than À40 dB isolation at 100 MHz and less than 0.1 dB insertion loss at 400 MHz.
During transmission, the array was driven in the quadrature circularly polarized (CP) mode at both frequencies, which in our case corresponded to a 45 phase shift between adjacent elements. For this purpose, we fabricated two eight-way splitters with corresponding phase shifters constructed of coaxial cables. Eight 1 H T/R switches and the 1 H eight-way splitter were located inside of the array holder to minimize cable losses. The reconstruction of the MRSI dataset was performed in MATLAB (Version R2018a) with a self-implemented reconstruction algorithm.

| Experimental evaluation of the array performance
Image reconstruction included FFT, spatial Hanning filtering and a WSVD (whitened singular value decomposition) coil combination. 52 Signal amplitudes were calculated from a time-domain fitting using a home-built version of the AMARES algorithm (MATLAB R2018a). 53 For optimization, the fmincon solver was applied.
The 13 C SNR was calculated from the signal amplitude of the central peak of ethylene glycol (combined in the SoS manner), which was estimated from the time-domain fit, and divided by the noise. For the noise calculation, the signal from eight voxels outside of the phantom were used. Thereby, the noise was calculated as the standard deviation of the real component of the time-domain signal. As described above, a WSVD coil combination was used, which accounts for the noise correlation of the eight 13 C loops. Since the B 1 + map was relatively homogeneous, no B 1 + correction was applied.

| RESULTS
In the first step, we numerically evaluated how many 1 H traps need to be introduced into each 13 C loop to minimize the alteration of the dipole B 1 + distribution and increase of the peak local SAR value. We compared the performance of the DT 13 C/ 1 H array with that of the ST 1 H eightelement dipole array (without 13 C loops) both loaded by the Duke voxel model. The DT array was simulated in two versions, ie 13 C surface loops having two or four 1 H traps. Figure 3 presents results of this comparison. Table 1 provides more details of the simulations. As seen in Table 1, the addition of 13 C loops with four traps and two traps decreases the average ⟨B 1 + ⟩/√P of the 1 H dipole array by about 10% and 14%, respectively.
SEE evaluated for both models (ie dipoles combined with two-trap loops and four-trap loops) measure about the same value, which is about 4% lower than that of the ST 1 H dipole array (without 13 C loops). In addition, the dipole 1 H B 1 + field map in the presence of two-trap loops shows a higher B 1 + field near the nose ( Figure 3A) due to a current induced in the corresponding 13 C loops. This also increases local SAR in this location ( Figure 3B).
Evaluation of the maximum local SAR produced by a DT RF coil at both frequencies is an important step of the safety evaluation procedure. 47 Figures 3B and 3C show examples of SAR 10g maps simulated for the presented DT array loaded by the Duke voxel model at both frequencies.
Positions of maximum local SAR significantly differ at the two frequencies. While the 13  We also evaluated the effect of combining 13 C loops with 1 H dipoles on the RF magnetic field and peak local SAR distributions at the 13 C frequency. In addition, we simulated the performance of the 13 C surface loop array without the RF shield. Simulation results are presented in Table 1 and Figures Figures 4B and 4D) causes a significant reduction of the 13 C Tx efficiency (⟨B 1 + ⟩/√P) (by 16%) while SEE does not substantially change. Also, as seen in Figure 4A, the ST 13 C loop array aligned Coupling between other elements measured below À19 dB. In addition, we evaluated an interaction between 1 H dipoles and 13 C loops at both frequencies. At 400 MHz, the coupling measured À27 dB or lower. At 100.5 MHz, cross-talk was much less and measured below À40 dB between all the 1 H and 13 C elements.
Finally, we measured the isolation between the 1 H Tx port and inputs of 13 C preamplifiers. Addition of second-order filters at the input of the 13 C combined T/R switches and pre-amplifier boards and shielding the 13 C interface allows us to obtain a total isolation better than À80 dB, as required for 13 C MRS experiments with proton decoupling.
In addition to simulations, the effect of the introduction of multiple 1 H traps into 13 C loops can be evaluated simply by measuring a change in the unloaded Q-factor (Q U ) value of the 13 C loop. We made these measurements for a single 13  After constructing the DT array coil and numerically evaluating its safety, 47 we tested the coil in the scanner using phantom and in vivo experiments. Figure 6A shows an experimentally measured 13  phantom. Averaged over the 130 mm transversal slab, ⟨B 1 + ⟩/√P measured 32 μT/√kW. This value was obtained considering losses in the cable and splitter and normalized to the RF power value at the coil input, ie eight feed-points of the antennas/loops. The simulated ⟨B 1 + ⟩/√P value was higher and measured 37.2 μT/√kW (Table 1). Figure 6B presents the 13 C SNR map.
Results of experimental tests at the 1 H frequency of the presented DT array coil loaded by the HS phantom are shown in Table 1 Figure 8A) and corresponding B 1 + ( Figure 8B) and SNR maps ( Figure 8C) obtained using three head array coils, ie the 1 H part of the new DT 16-element 13   voltage of 250 V, the desired decoupling behavior is achieved. At lower voltages, only a partial decoupling is detected over the entire volume of the phantom. For a decoupling voltage of 250 V, the peak local SAR reaches 31% of the maximal allowed value compared with no significant absorption without decoupling and the applied sequence parameters. This calculation of peak local SAR was based on a conservative approach that considers an additional safety factor of 2 for the k-factors at both frequencies. 47 The increase of the noise level (standard deviation) is within the measurement error ($10%).

| DISCUSSION
As mentioned above, maintaining reasonably high Tx and R X performance of the DT coil at the 1 H frequency is very important for many applications. However, it is often compromised for the sake of design simplification. Previously, we developed the UHF DT 31 20 However, the coverage was still worse than that obtained by the 2 Â 8 ST 1 H array. 19 The presented novel design of the DT array coil combining 13 C surface loops and 1 H bent folded-end  It also important that in comparison to 31 P/ 1 H array, 11 which in addition to eight surface loops surrounding the head has two TxRx crossloops at the superior head location, the developed dipole/loop array (without having two superior cross-loops) provides comparable SNR at the superior head area and somewhat better SNR down the brain stem ( Figure 8). For comparison, an eight-loop array design, which does not include the superior cross-loops, has significantly lower SNR in the superior head area. 20 The SNR of the 1 H-dipole array can be further improved by adding a pair of Rx-only cross-loops or cross-dipoles at the top of the head.
As we demonstrated in our study, the correct choice of the number of 1 H traps and their placement in X-nucleus surface coils is critical for maintaining the 1 H dipole array Tx performance. It is not easy to provide a general recipe for choosing the trap number and their placement, which depends on the specific geometry of the array, eg the size and geometry of the X-nucleus loops. However, we would like to emphasize the following features. First, dipoles interact mainly with two 13 C surface loops located beneath adjacent dipoles and much less with the loop beneath the dipole itself. 55 In our case, this interaction is further increased because we used wider overlapped 13 C loops. Making surface loops narrower and increasing gaps between them will most likely decrease the interaction between dipoles and loops, but also compromise the performance of the X-nucleus array. However, the smaller number of 1 H traps distributed along the X-nucleus loop may suffice. Second, in our design, the length of the wire that was used to construct the 13 C loop measured about 600 mm. By interrupting the wire with two 1 H traps (Traps 2 and 4, Figure 2A), we produced two 300 mm pieces of wire. This is very close to half of the wavelength at 400 MHz and the total length of the folded-end dipoles. Thus, these wires can strongly interact with the dipoles and disturb the RF field distribution ( Figure 3A). Again, for other designs with loops of different size, the results may differ. While 1 H traps were critical to maintaining the performance of the dipole array, the performance of the 13 C loop array was not significantly altered by the presence of 1 H dipoles (Table 1) and hence no 13 C traps were placed in the 1 H elements.
Evaluation of the peak local SAR is an important part of the safety evaluation procedure. 47 As seen in Figure 3C, the 13 C surface loop array demonstrates a SAR distribution, which is characteristic for the CP mode, 17,33 ie with local SAR increased at the superior location and very low SAR near the center of the head. At the same time, the 1 H dipole array shows the highest local SAR near the center of the head. This fact is explained by coupling of the 9.4 T bent folded-end dipole array to the intrinsic TE mode of the head, which produces a tangential component of electric field near the head center. 31 As demonstrated previously, bending and folding the dipoles in the presence of the RF shield facilitates coupling of the array to the TE mode at 400 MHz. 31 Optimization of the length of the folded portion of the dipole allows minimization of the peak SAR value and SEE. 31,56 A difference in Tx efficiency was observed for both frequencies between the EM simulation results (higher) and the experimental results (lower, Table 1), which can be explained by difficulties in taking into account all losses produced in coil components and conductors in EM simulation models. Commonly, EM simulations give overestimates of the B 1 + value. 18,47 This issue is even more pronounced for dipole antennas, where relatively coarse meshing is performed over thin (1.5 mm in diameter) dipole wires. Importantly, this difference between simulated and experimental data does not compromise the coil safety due to overestimations of peak local SAR values as well.
An RF shield is commonly used in the UHF RF array coil designs to minimize radiation losses, 57 decrease coupling between non-adjacent elements, 35 and reduce interaction with the magnet bore environment including the gradient coil. At the same time, the presence of the RF shield may decrease the Tx efficiency and SNR due to out-of-phase current induced in the shield. This is more important at the lower X-nucleus frequency due to lower tissue losses, and, therefore, higher Q-factors. Therefore, we investigated the effect of the RF shield presence on the 13 (Table 1). This is explained by a significant increase of coupling between non-adjacent loops due to RF shield removal, 35 which in turn causes an increase of the reflection power and, as a result, a decrease in the coil accepted power and power deposited into tissue ( Table 1). As seen in Table 1, the tissue power deposition was reduced by 38% for the unshielded array, which matches the reduction of the 13 C-array B 1 + value. Thus, in our case, the RF shield plays an important role for both 1 H and 13 C array designs due to an improvement of the element decoupling. 35,36 Choosing an appropriate decoupling method is a critical component of any TxRx-array development. In our design, decoupling of dipole elements was provided by the presence of the RF shield. As demonstrated recently, 36 decoupling between adjacent folded-end dipoles is produced due to a capacitive coupling of the folded portion to the RF shield. 36 Double-tuning and the presence of the 13 C loops increased coupling between adjacent dipoles, which was still quite reasonable, ie average S 12 of À15.6 dB between adjacent dipoles. This is about 2-3 dB worse than that obtained for the 1 H ST bent folded-end dipole array. 36 At a lower frequency (100.5 MHz), array decoupling is more challenging. Due to substantially lower loading, there is a need to decouple adjacent as well as closest non-adjacent loops (ie 1 and 3, 2 and 4, etc). At the end, we still measured the highest coupling ($À13 dB on average) between loops separated by two elements (ie 1 and 4, 2 and 5, etc), which were not decoupled at all. Incorporating more decoupling circuits to minimize coupling between these 13 C loops would make the design quite complex, and to the best of our knowledge has never been presented before. This residual coupling caused a small inhomogeneity seen in the transversal in vivo 13 C B 1 + map ( Figure 6).
As discussed above, the new DT array substantially simplifies the coil design and provides for good longitudinal 1 H coverage without increasing the number of elements. At the same time, the Tx efficiency is still lower than that of the 1 H ST loop array. We believe that optimization of the dipole geometry and the trap placement may help to further optimize the performance of the dipole array. In addition, with only eight 1 H Rx-elements, the parallel imaging performance (mainly head-to-feet direction) and ability to perform multi-band reception are still compromised as compared with state-of-the-art 32-channel Rx head coils available on UHF systems.

| CONCLUSIONS
In this study, we developed, evaluated, and constructed a DT 13  are gratefully acknowledged.

DATA AVAILABILTY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.