Evaluation of coaxial dipole antennas as transceiver elements of human head array for ultra‐high field MRI at 9.4T

The aim of this work is to evaluate a new eight‐channel transceiver (TxRx) coaxial dipole array for imaging of the human head at 9.4T developed to improve specific absorption rate (SAR) performance, and provide for a more compact and robust alternative to the state‐of‐the art dipole arrays.

use of UHF MRI is still limited because of problems associated with strong inhomogeneity of the RF electromagnetic (EM) field and resulted safety issues.The latter is caused by shortening of the wavelength in high permittivity (∼50) conductive (∼0.5 S/m) human body tissues. 3Longitudinal (along the magnet axis) coverage of the human brain in transmission is also substantially decreased because of interaction of the RF EM field with the tissue 4 and could not be improved by increasing the length of a common quadrature birdcage-like RF coil. 5In addition, shortening of the wavelength causes a substantial increase of local tissue heating commonly assessed by evaluating the specific absorption rate (SAR). 6,7To overcome the RF field inhomogeneity and high local SAR issues, multi-element human head transmit (Tx) arrays consisting of a single-row (usually eight elements, i.e., 1×8 configuration) [6][7][8][9] of loops have been previously described.However, to deliver whole brain coverage, multi-row (e.g., double-row 2×8) arrays [9][10][11][12] in combination with dynamic parallel transmission 13 or static 3D RF shimming 14,15 are essential.To provide for relatively uniform current and voltage distributions along the conductor of the loop, each loop commonly caries a large number (e.g., 10 and more) of distributed high-voltage capacitors.This substantially increases complexity and, hence, may compromise the reliability and safe usage of human head Tx loop arrays. 8herefore, the development of novel and more reliable UHF arrays designed to overcome issues with the B 1 + inhomogeneity and high local SAR is currently in high demand. 9pproximately 10 years ago dipole antennas were introduced as elements of Tx-or transceiver (TxRx) RF array coils for imaging of the human body at UHF. 16 Because of the simplicity of the design, dipoles can provide a reliable and robust alternative to common surface loops.In the last decade, multiple studies describing the use of dipoles for designing multi-element human head 12,[17][18][19][20][21][22][23][24][25][26] and body [27][28][29][30][31][32][33][34] Tx (or TxRx) arrays have been reported.Because of the sample's size, dipole antennas for human head imaging must be designed significantly shorter than their intrinsic length equal to half of the wavelength, λ/2, for example, 50 cm at 300 MHz (7T) and 37.5 cm at 400 MHz (9.4T).Shortening of the dipole lengths (∼17-19 cm) leads to a very non-uniform (almost "triangular") current distribution with a relatively sharp peak at the dipole's center. 24,35This, in turn, reduces the longitudinal RF field coverage and causes high peak SAR (pSAR) values.In addition, common straight dipoles generate a high conservative electric field at their ends.This leads to a large (more than 10 MHz) change of the resonance frequency of tight-fitted straight dipoles with variation in the head size. 36,37This effect can be reduced by moving the dipoles further away from the head.
That, however, leads to a decrease in the Tx-efficiency and SNR.
23][24][25][26]36 In this design, the ends of the common straight dipole, which carry a very low current and high electric field values, were folded and moved away from the subject.The length of the folded-end dipole in the z-direction (along the magnet axis) was the same as that of the straight dipole with the total length of the conductor (including the folded parts) being longer and close to λ/2.As a result, the folded-end dipole produced a more uniform current distribution and lower voltage values along the straight part located near the sample.This alteration of the common dipole design resulted in the extended RF field distribution and lower pSAR value. 24In addition, the novel dipole antenna allowed minimizing the frequency shift because of head size variation, and, hence, could be used in tight-fit transceiver and receive arrays.The folded-end dipole design, however, has certain disadvantages because of bulky folded ends, which require an additional space inside the coil housing.This also makes cable routing more difficult because of the interaction between the dipole and cable going from the center of the dipole parallel to the folded ends, which carry a high electric field.In addition, this compromises the ability of using folded-end dipoles as elements of a multi-row (e.g., 2×8) array where cable routing becomes even more difficult.Another method to flatten the current distribution and stretch the RF field distribution in the longitudinal direction is using coaxial dipoles in combination with lumped inductors placed at the dipole ends between the central conductor and shield. 33This type of dipole antenna was recently used for designing a TxRx body array and demonstrated lower pSAR compared to an array consisting of straight fractionated dipoles. 33n this work, we developed a new 9.4T eight-element human head TxRx coaxial dipole array with a compact and robust design.Because of significant differences in design constrains between UHF RF array for human body and head, which are mainly determined by the large difference in object size in comparison to the RF wave length inside the human tissue (10-12 cm at 400 MHz), previous data obtained from body array simulations 33 cannot be directly applied to the head array design.Therefore, in this work, we performed a complete numerical optimization, bench and scanner evaluation of the coaxial dipole array developed for human head imaging at 9.4T.First, to minimize pSAR value and the frequency shift because of the head size variation, we optimized a single coaxial dipole element.The eight-channel array was designed and evaluated numerically based on single element optimization results.Finally, results of numerical simulations were verified on a bench and in the scanner including in vivo measurements on a healthy volunteer.To the best of our knowledge, this is the first example of developing a tight-fit coaxial dipole array for human head MRI at UHF.Another very important and novel aspect of our work, which has never been demonstrated previously, is finding, based on the numerical optimization, that the coaxial dipoles allow minimizing frequency shift because of the head size variation and, therefore, are suitable for tight-fit human head array designs.

Coaxial dipole design
Similarly to the previous work, 38 the dipole antenna is constructed using a coaxial cable with two gaps in the shield introduced at a distance d from both ends of the antenna as shown in Figure 1A.In addition, following the design described in work, 38 lumped inductors, L end , are introduced between the center conductor (core) and shield at both ends of the dipole.The excitation port is placed at the dipole center.The main difference with the original design 38 is that the dipole in the new design is not driven on the core of the cable, but directly on the shield, which is shorted with the inner conductor at the center of the dipole (Figure 1A). Figure 1B demonstrates the equivalent schematic of the coaxial dipole shown in Figure 1A.This antenna corresponds to the fractionated dipole 33 with two inductors inserted at the distance d from both ends of the antenna.Each inductor is mainly produced by two inductors connected in series, that is, the small inductor formed by the short-ended coaxial cable between the gap and the center, L 1 , and inductor L d created by the short piece of the cable at the dipole end loaded by L end (Figure 1C).With a 50-Ω cable and a distance between the gap and the center of ∼70 to 80 mm, inductance L 1 equals to 20 to 25 nH.L d is substantially larger and, for example, for L end of 40 nH and a short cable of 20 to 30 mm, ranges from 85 to 190 nH.Therefore, a large inductance in the fractionated dipole design can be replaced by a substantially smaller inductor at the end of the coaxial dipole.As demonstrated in work, 38 by adjusting the position of the gaps and value of L end , the current distribution along the dipole length can be flattened and, therefore, pSAR minimized.Figure 1D,E show unfolded straight and fractionated dipole also simulated for comparison.

EM simulations of a single dipole element
For single-element simulation and optimization, we used the finite-element method in the frequency domain implemented in CST Studio Suite 2021 (Dassault Systèmes).In the first step, the numerical simulation setup consisted of a 176-mm diameter cylindrical phantom mimicking human head tissue properties 39 (ε = 58.6,σ = 0.64 S/m at 400 MHz), a 215-mm diameter cylindrical FR-4 holder, and a single dipole element (Figure 2A).The 190-mm in length coaxial dipole antenna was designed using a 50-Ω coaxial cable with a dielectric diameter of 3.45 mm and center conductor diameter of 1 mm.PTFE/Teflon (ε = 2.1, tanδ = 2.2⋅10 −5 ) was used as a dielectric.A relatively large cable diameter was chosen for the purpose of a reasonable (<3-4 h) simulation time of array designs loaded by human voxel models.Otherwise, a very fine local mesh is required that substantially increases the simulation time.According to our numerical simulations of a single dipole element, there is a very minor influence of the coaxial cable diameter on the dipole performance.Reducing the Coaxial (A) and its equivalent fractionated (B) dipoles.(C) General representation of equivalence between a lumped inductor, L d , and 50-Ω coaxial line terminated with a lumped inductor, L end .Additional dipole designs, that is, straight (D), and folded-end (E), simulated for comparison.coaxial cable diameter to 1.48 mm, corresponding to the diameter of RG-405 semi-flexible cable, led only to 1.04 times drop in SAR-efficiency in the phantom center, but required a significantly larger simulation time.L end was simulated as a lumped inductor with a Q-factor of 150.The Q value was evaluated on a bench for a self-made inductor constructed using 1.2-mm diameter copper wire.Five gap positions, d, were considered in the numerical simulations: 10, 20, 30, 40, and 50 mm from the cable end.In addition to the coaxial dipole, we simulated the fractionated dipole (Figure 1B), straight (Figure 1D), and folded-end (Figure 1E) dipoles for comparison.All these dipoles were made of 1.5-mm annealed copper wire and measured 190 mm in the longitudinal z-direction (along the cylindrical axis of the phantom and magnet).The locations of inductors in the fractionated dipole were similar to the positions of gaps in the coaxial dipoles.The height of the folded-end dipole was 35 mm and the fold measured 30 mm (Figure 1E).Because the outer diameter of the coaxial dipole was ∼2 times larger than the size of the wire, which may affect the dipole performance, we also simulated the fractionated dipole using thicker 3.5-mm wire.Finally, we simulated a fractionated dipole with four distributed lumped elements, that is, the original fractionated dipole design. 33Each dipole element was tuned and matched using an L-network consisting of two series inductors and a parallel capacitor using the CST Schematic module.Matching was adjusted until S 11 measured −30 dB or below.For all single dipole elements, we evaluated SAR-efficiency, B 1 + / √ pSAR 10g , and Tx-efficiency, B 1 + / √ P stim , where P stim is the power stimulated at the coil's input and pSAR 10g is the peak SAR averaged over 10-g of the tissue mimicking phantom.The current distribution along the dipole length was evaluated as the corresponding component of the RF magnetic field (H x ) measured at a 5-mm distance from the wire center.
In addition, for each single dipole shown in Figure 1 we evaluated the shift of the resonant frequency with phantom size variation mimicking variation in the human head size.For this purpose, after tuning and matching each dipole on the larger 176-mm diameter phantom, we replaced the phantom with a smaller 142-mm diameter phantom and simulated the S 11 -parameter without adjusting capacitors and inductors in the matching circuit.
Finally, for quantitative comparison, we simulated the original coaxial dipole design. 36

Eight-channel dipole array numerical optimization
The final numerical model of the array consisted of eight 190-mm long coaxial dipoles (d = 20 mm, L end = 40 nH) placed on an elliptical FR-4 holder with 3 mm wall thickness and uniformly surrounding the head (Figure 3A).The shape of the holder was the same as previously used in our 9.4T dipole head arrays designs 21,23,24,26 and measured 200 mm from left to right and 230 mm from top to bottom.Similar to previous works, for improving the longitudinal coverage, a local RF shield (Figure 3A) was added at the superior location. 11In addition, a 640 mm in diameter copper cylinder with 1600-mm length was added to the model to mimic the RF shield (bore) of the MRI scanner.Simulations were performed in CST Studio Suite 2021 using the finite-integration method in time-domain.For comparison, we also simulated eight-element folded-end and fractionated dipole arrays of the same length and shape of the housing.All arrays were loaded with multi-tissue "Duke" and "Ella" voxel models with 2-mm resolution (Zurich MedTech).A local mesh with a maximum voxel size of 1 mm was applied to the 3D objects for all simulated arrays.The arrays were simulated using a server equipped with three Nvidia Tesla V100 GPUs for acceleration.Each array had ∼70 million mesh cells, with simulation times of ∼3 to 4 h each.After finishing EM simulations, inputs of all array elements were matched to the 50-Ω source in the schematic module of CST.All array configurations used a lumped element L-matching network with parallel capacitance and two series inductances.The level of input matching for all coil elements in numerical simulations was better than −30 dB.
In the next step, circularly polarized (CP-mode) (45 • phase increment between the channels) B 1 + field and SAR distributions were calculated for 1 W power stimulated at the array inputs.SAR was averaged over a 10-g tissue mass using the CST Legacy method.Additionally, we calculated the average and deviation of the B 1 + value over the 130-mm transversal slab, which includes the majority of the human brain.Based on these results, we evaluated both SAR-efficiency and Tx-efficiency for all three simulated arrays and two voxel models.

Array construction
After finishing EM simulations, we constructed the eight-channel transceiver coaxial dipole array (Figure 3B).The geometry of the array holder and all elements were similar to the geometry in the simulation (Figure 3A).The array consisted of eight 190-mm coaxial dipole antenna elements (d = 20 mm) with L end = 40 nH.A 50-Ω semi-flexible non-magnetic RG405 cable with a 1.48 mm outer diameter (Carlisle Interconnect) was used for dipole construction.Eight panel-mount BNC connectors were placed on the polycarbonate plate (Figure 3B) and connected to each dipole using a RG405 cable.Figure 3D shows a schematic of a single coaxial dipole element including inductors at the ends (L end ) and matching inductors L t (∼15 nH), a matching capacitor C m (1-19 pF, Johanson), and two floating ground cable-traps 40 (baluns), which were placed to prevent wave (the common mode) propagation along the feeding cable shield.All inductances were self-made using 1.2-mm tinned copper wire.
A home-built transmit/receive interface, including a set of high-power PIN-diode TxRx switches 6 and low-noise amplifiers (WanTcom), was used to connect the coil to the MRI scanner.

Bench and in vivo measurements
Before in vivo measurements, the dipole array was evaluated on a bench and in the scanner using a homogenous head and shoulder phantom (ε = 58.6,σ = 0.64 S/m at 400 MHz) according to the safety procedure developed in our lab. 41The local ethics committee approved the in vivo study, and informed consent was obtained from each subject.In vivo T 1 -weighted and T 2 *-weighted images of a healthy volunteer were acquired using MPRAGE (TI = 1340 ms, TR = 3360 ms, GRAPPA 2, matrix size 264 × 264 224 mm, resolution 0.8 mm isotropic, inversion preparation by adiabatic hyperbolic secant pulse (HS4)) 42 and gradient echo (TR = 11 ms, TE = 7 ms, GRAPPA 2, FA = 5 • , matrix size 264×264×224 mm, resolution 0.8 mm isotropic).A B 1 + map was measured using a presaturated TurboFLASH sequence 43 (TR = 2.5 ms, T rec = 7.5 s, TE = 0.73 ms, GRAPPA 2×2, FA = 2 • /70 • , matrix: 64×64×64, resolution 3.5-mm isotropic).From that, following, 43 the B 1 + field normalized by input power was calculated as from the obtained flip angle α, the impedance R = 50Ω and the duration  and amplitude U of the rectangular pre-saturation pulse.RF power was evaluated at the coil plug taking into account cable losses.In all experiments, the array was driven in the CP mode.

Numerical optimization of a single coaxial dipole element
First, we evaluated the effect of the gap position d and L end values on the shape of the current distribution along the dipole length.Numerical simulations of coaxial dipoles with d of 40 mm and larger showed the "triangular" current distribution similar to the straight dipole.Additionally, simulations of the dipole with the gap placed at 10 mm from the end resulted in a large value of L end , (85 nH).Therefore, in this step, we only considered d values of 20 and 30-mm.Figure 4A,B display current distributions obtained for 20 and 30-mm gap positions and several L end values.In the case of the coaxial dipole with the 20-mm gap position, the maximum L end value was 40-nH, which produced the flattest current distribution.Further increasing the L end value worsened the distribution.This effect was more pronounced for the 30-mm gap position as shown in Figure 4B.Although the L end value of 25-nH resulted in relatively flat current distribution, the L end value of 30-nH led to a substantial decrease of the current at the center of the dipole.Figure 4C compares current distributions for all four types of dipole elements shown in Figure 1 including the common straight dipole, folded-end dipole, 20-mm/40-nH coaxial dipole, and fractionated dipole.In the case of the fractionated dipole with the gap at 20 and 30 mm, L end measured 290 and 200 nH, respectively.Both the coaxial and fractionated dipole designs produced the flattest current distribution.However, the fractionated dipole required a substantially greater L end value.Increasing the size of the wire to 3.5-mm for the fractionated dipole has an almost negligible effect on the B 1 + (1.025 times drop in the center for 1.5-mm dipole compared to the 3.5-mm dipole) and SAR (1.012 times drop of pSAR for 3.5-mm dipole compared to the 1.5-mm dipole) distributions but leads to a decrease of L end to 110 and 100-nH for the gap positioned at 20 and 30 mm, respectively.
In the next step, to evaluate the influence of the current distribution on the RF magnetic field, we calculated the SAR-efficiency (B 1 + / √ pSAR 10g ) distributions for all four types of dipole elements loaded by the cylindrical phantom.Figure 5 shows simulated central coronal SAR-efficiency maps.In agreement with data shown in Figure 4A-C, coaxial dipoles with smaller L end values and stronger central current demonstrate an RF field distribution that is very similar to that of the straight dipole.Also, flattening the current distribution led to extending SAR-efficiency maps in the longitudinal direction with best results obtained for 20-mm/40-nH and 30-mm/25-nH coaxial dipole designs configurations and corresponding fractionated dipoles.Further increase of the L end value to 30 nH for the 30-mm gap coaxial dipoles resulted in a drop of the B 1 + field near the center of the phantom.Even further increase to 40 nH led to a very inhomogeneous RF field distribution and very high dipole input impedance that could not be matched to a 50-Ω source using the L-circuit with realistic inductance and capacitance values.In addition, the coaxial 20-mm/40-nH dipole demonstrated the lowest pSAR value among all dipole configurations except for the 30-mm/40-nH coaxial dipole.The latter is rather impractical because of the low penetration depth and Tx-efficiency (Figure 5).For quantitative comparison, Figure 4D shows a plot of the SAR-efficiency obtained along the central axis of the cylindrical phantom for the straight, fractionated, folded and coaxial dipole designs.Both coaxial and fractionated dipoles demonstrated 1.09 times higher SAR-efficiency than the folded-end dipole.However, the coaxial dipole design allowed using substantially smaller L end .Therefore, a coaxial dipole with d = 20-mm and L end = 40-nH was chosen for the final array design.Figure S1 shows comparison of the original coaxial dipole design and our version including the current, voltage, and RF field distributions.In both cases, the gap positions and L end values

F I G U R E 5
Simulated SAR-efficiency maps for the folded-end, straight, fractionated, and coaxial dipole antenna elements in the central sagittal plane of the cylindrical phantom.
were the same.Apparently, both designs are very similar with our design having slightly better (∼1.01 times) SAR-efficiency.
Finally, to evaluate sensitivity of the considered dipole designs to the change in the sample size, we numerically calculated the shift of the dipole resonant frequency resulting from variation in the size of the phantom (Figure 2B).As seen in Figure 2B, the frequency shift for the 20-mm/290-nH fractionated, folded-end, and optimized 20-mm/40-nH coaxial dipoles were very similar and measured ∼5 MHz, whereas the shifts measured for the straight dipole and suboptimal 20-mm/20-nH coaxial dipole were substantially larger, (∼10 MHz).Such a significant frequency shift implies that the tight-fit straight and suboptimal coaxial dipole array designs are not very practical because such coils will require manual tuning on each subject.To understand the effect, we simulated the conservative electrical field distribution along the length of the dipole for both the optimized coaxial and straight dipole as a reference.Figure S2 shows this comparison.As seen in the figure, the area of strong electrical field is substantially reduced in the coaxial dipole design.This reduction is responsible for the smaller sensitivity toward load variation.
Additionally, to evaluate the effect of adding more lumped inductors along the dipole length, we simulated the original fractionated dipole design with four inductors.Two additional inductors were placed closer to the dipole's center.Figure S3 shows a comparison of current distributions obtained for dipoles with two and four inductors.As seen in the figure, adding more inductors near the center increases the current at the center and leads to an increase in the pSAR value.

Eight-channel coaxial dipole array simulation
After optimizing the single coaxial dipole element, we simulated the final array design consisting of eight 20-mm/40-nH coaxial dipoles (Figure 3A). Figure 6 shows numerically calculated central sagittal Tx-efficiency maps obtained using the 20-mm/40-nH coaxial dipole array as well as the 20-mm/290-nH fractionated and folded-end dipole arrays for comparison.All arrays were loaded by the Duke voxel model.Figure 7 shows SAR 10g distributions for the same three arrays calculated in the central sagittal plane as well as in transversal planes cut through the locations of the maximum local SAR.Figure 8 shows central sagittal SAR-efficiency maps for the same arrays.Numerically calculated SAR and SAR-efficiency distributions obtained for the developed coaxial array loaded by the Ella voxel model are presented in Figure S3.Table 1 summarizes all simulated data obtained for the developed array using both voxel models.The presented data show that the use of coaxial dipoles allows for substantially decreasing the pSAR value in comparison to the folded-end dipole (up to 1.46 times) and fractionated dipole (up to 1.13 times) arrays.At the same time, the folded-end dipole array demonstrates the highest Tx-efficiency, that is, ∼1.1 times higher than both the coaxial and fractionated dipole designs.However, most importantly, the developed coaxial dipole array shows the highest SAR-efficiency, which is up to 1.1 times higher than that of the two other arrays (Table 1).

Coil evaluation and in vivo measurements
After numerical optimization, we constructed the eight-element coaxial dipole array (Figure 3B). Figure 3C shows the full eight-port S-parameter matrix obtained for the developed array loaded with the homogenous head and shoulder phantom.Coupling between adjacent elements was not worse than −10.9 dB with the average value of −11.1 dB.Non-adjacent array elements had coupling better than −21 dB with the average value of −25.1 dB.
Figure 9 shows an in vivo MPRAGE image (Figure 9A) and B 1 + field map (Figure 9B) acquired using the developed array and a healthy male volunteer (head size-59 cm in circumference).The measured mean B 1 + over the Electromagnetic simulated central sagittal transmit (Tx)-efficiency maps for the optimal coaxial, fractionated, and folded-end dipole arrays all loaded by the Duke voxel model.The averaging volume, that is, the 130-mm transversal slab, is marked by a white dashed line.

F I G U R E 7
Electromagnetic SAR 10g distributions in the central sagittal plane and transversal plane cut through the maximum of SAR 10g simulated for optimal coaxial, fractionated, and folded-end dipole arrays loaded by the Duke voxel model.

F I G U R E 8
Electromagnetic central sagittal SAR-efficiency maps simulated for the optimal coaxial, fractionated, and folded-end dipole all loaded by the Duke voxel model.

T A B L E 1
Comparison of the dipole array transmit performance for the optimal coaxial, fractionated, and folded-end dipoles for the Duke and Ella voxel models.

Array
Voxel model pSAR pSAR ratio <B 130-mm region was 0.283 μT/ √ W normalized to the coil input.Losses in the cables and Tx-Rx interface were considered in this evaluation.

DISCUSSION
We developed, constructed, as well as numerically and experimentally evaluated a prototype of a new array coil consisting of eight coaxial dipoles for MRI of the human head at 9.4T.Similarly to the recently developed folded-end dipole antennas, coaxial dipole elements provide for more uniform current distribution along the dipole length compared to the common straight dipoles.Flattening the current distribution allows extending the RF magnetic field longitudinally and decreasing the pSAR value.In addition, coaxial dipoles are substantially less sensitive to the variation in the human head size than common straight dipoles (Figure 2B), which allows using this type of antennas in tight-fit transceiver array designs for improved Tx-efficiency.Although the folded-end dipoles provide similar benefits, such antenna design has several disadvantages in comparison to the coaxial dipoles.First, to reduce interaction between folded and non-folded dipole parts, which carry opposite currents and may lead to field cancelation and inhomogeneities (i.e., a "vertical loop" effect 36 ), the dipole height should be relatively large 12,[21][22][23][24][25][26]36 (e.g., 30 mm). Ths substantially increases the array's size in the axial direction.Another disadvantage of having folded parts, which carry a high conservative electrical field, is their interaction with cable traps and with the feeding cable positioned parallel to the dipole wire.In contrast, the coaxial dipole provides a very compact design and an option of moving cables away from the dipoles.Alternatively, a balanced driving of the coaxial dipole can be realized using transformer coupling to one of the inductors at the end of the dipole.This option allows further reduction of coupling between the dipole and cable.Finally, the coaxial dipole array provides for ∼1.1 times improvement (Table 1) in the SAR-efficiency in comparison to the folded-end dipole array of similar geometry.
In addition, the coaxial dipole design provides great flexibility in adjusting the resonance frequency of the dipole by varying the L end value.The same array design can be easily transformed into lower (e.g., 300 MHz for 7T) or higher frequency array simply by modifying the inductance value and/or the position of the gaps.Also, as seen in Figure 5, the RF field distribution can be relatively easily adjusted by altering the inductance value and/or the gap position depending on the requirement of the specific application.For example, by altering the inductance (or gap position) at one side more than at the other side, the dipole element with asymmetrical current distribution can be produced. 44Combining such asymmetrical dipoles into an array can be beneficial for parallel transmit 14 or receive 45 imaging at UHF by virtually increasing the number of channels.
Technically, the coaxial dipole design is very similar to the fractionated dipole antenna with two lumped inductors placed at the gap positions.Therefore, fractionated dipoles tuned to the resonant frequency can also substantially reduce pSAR and flatten the current distribution (Figures 4 and 5).However, such dipoles require a significantly larger inductances for tuning than the coaxial dipoles, that is, 290-nH for a 1.5-mm dipole wire with inductors placed at 20-mm from the dipole ends.Size of such an inductor exceeds 10-mm, which may produce interference with the RF field in the sample when the inductor not properly positioned.In addition, tuning the resonance frequency of the coaxial dipole can be easily performed by a small adjustment of the end inductors.It is much harder to produce the same relative change of a much larger (300-nH) inductor.Increasing the wire size to 3.5-mm reduces the inductor value to 110-nH, which is still a large and relatively bulky inductor.Using such a thick and inflexible wire for constructing dipoles is also not very practical.In addition, the coaxial dipole array provides up to 1.1 times higher SAR-efficiency (Table 1) than the fractionated dipole array.
In this work, we tested a prototype of the new array design consisting only of eight TxRx elements.This is much smaller than the number of Tx and Rx elements in the current state-of-the-art UHF array design.
The number of TxRx dipole elements can be further increased to 16 as demonstrated previously. 46Furthermore, the array can be further optimized by increasing the number of Rx-elements.For example, this can be done by combining coaxial TxRx-dipoles with Rx-only loops as previously suggested. 22According to theoretical 47,48 and experimental data, 36,49 such a combination is required at UHF to provide for optimal SNR near the center of the human head.

CONCLUSION
In this work, we developed, constructed, and numerically and experimentally evaluated an eight-element TxRx coaxial dipole array for human whole-brain imaging at 9.4T.
The array has a more compact and robust design with superior SAR performance compared to the previously described folded-end dipole array.The proposed coaxial dipoles could also be easily modified and used as elements of a multi-row TxRx-array at lower (e.g., 7T) or higher (e.g., 11.7T) field to improve brain coverage and lower pSAR compared to other dipole designs.

ACKNOWLEDGMENTS
Electromagnetic simulation model of a single dipole element loaded by a cylindrical phantom; (B) simulated S 11 plots for different dipoles loaded by the small and large cylindrical phantoms.

F
I G U R E 3 (A) Electromagnetic simulation model of the eight-channel coaxial array; (B) photo of the constructed dipole array without the cover for better visualization; (C) measured S-parameter matrix for the eight-channel 20 mm/40 nH coaxial dipole array loaded with homogenous head and shoulder phantom; (D) schematic of a single coaxial dipole array element.

F I G U R E 4
Numerically calculated current distributions along the coaxial dipole with 20-mm (A) and 30-mm (B) distances from the dipole's end; (C) comparison of current distributions of straight, folded-end, fractionated, and optimal (20 mm/40 nH) coaxial dipoles; (D) specific absorption rate (SAR)-efficiency numerically evaluated along the central axis of the cylindrical phantom for the straight, folded-end, fractionated, and optimal (20 mm/40 nH) coaxial dipoles.

F I G U R E 9
In vivo MPRAGE brain images (A) and corresponding B 1+ maps (B) of a healthy volunteer in axial, sagittal, and coronal planes.The array was driven in the CP mode.