Scalable and modular 8‐channel transmit and 8‐channel flexible receive coil array for 19F MRI of large animals

To introduce an RF coil system consisting of an 8‐channel transmit (Tx) and 8‐channel receive (Rx) coil arrays for 19F MRI of large animals.


INTRODUCTION
F MRI has been implemented successfully in rodents to study myeloid cell uptake and migration after myocardial infarction (MI) and in atherosclerosis, [1][2][3][4][5][6] but only few reports are published in which this imaging method has been used in human-size models such as pigs 4,6 ; however, the transfer of 19 F MRI into large animals is crucial for both method validation and development of clinical applications. A major challenge of upscaling the 19 F MRI studies to larger animals is the design and construction of a dedicated 19 F RF coil system. At B 0 = 3 T, the Larmor frequency of the 19 F nucleus of 120.2 MHz is only 7.5 MHz lower than that of 1 H with 127.7 MHz, which requires special decoupling circuits to avoid unwanted coupling. Furthermore, electromagnetic and mechanical compatibility with the existing 1 H RF coils is also required for studies in which interleaved measurements of multiple nuclei are performed. Another limitation is the low peak RF power of the broadband transmit system for X-nuclear MRI: In clinical 3T whole-body MRI systems, often RF power amplifiers with only a few kW are available at the 19 F frequency, whereas the narrow-band transmitter for conventional 1 H MRI can deliver peak powers up to 15 kW and more. In general, target design parameters for a 19 F coil system are homogeneous excitation and high SNR within the target anatomy (in this case, the heart). Unlike 1 H MRI, in 19 F MRI, the signal solely originates from the administered 19 F tracers, and no background signal from the body is present. Thus, the effective imaging volume can be restricted to a specific tissue or organ. Various multinuclear coils have been designed for preclinical studies on small animals. 7 A double-tuned birdcage coil can be realized using either a trap circuit 8,9 or a PIN diode, 10 and mechanical actuators were proposed to avoid losses from the PIN diode. 11 To construct multinuclear birdcage coils, alternating rungs were individually tuned to two different frequencies, 12,13 or the end rings were designed as four-ring birdcage coils. 14 However, double-tuned birdcage coils are difficult to realize when the two resonance frequencies are very close. In Joseph et al, two degenerate modes were tuned to the 19 F and 1 H resonances by nonuniform distribution of capacitances in the rungs of birdcage coil. 15 Insertion of an additional shield with a smaller diameter was also proposed to switch the resonance frequency to 19 F Larmor frequency. 16 Muftuler et al suggested an automatically tuned birdcage coil using variable capacitance (varactor) diodes. 17 A double-tuned birdcage coil was developed using the coupled resonator method. 18 In Choi et al, 19 PIN diodes with inductors were used to dual-tune a birdcage to the 19 F and 1 H frequencies.
A double-tuned coil can also be constructed using separate coils, such as two birdcage coils in co-axial configuration [20][21][22][23] or concentric loops. 24,25 Although each coil resonates at a different frequency, mutual coupling can cause interference between the coils due to induced current between the loops. Counter-rotating currents can cause significant loss in magnetic field intensity of the coils. 21 Geometric decoupling can also be used in double-tuned coils. [26][27][28][29][30] Although there are many examples of 19 F coils for small animal measurements, only a few designs were demonstrated for human and large animal applications: 19 F knee MRI at 7 T, 31 human liver MRS, 32 and lung MRI 33 at 1.5 T, pig at 3 T using a dual-tuned transmit (Tx), and a 19 F-surface loop coil. 34 In general, dual-tuned 19 F/ 1 H coils show an inferior performance compared with mononuclear coil designs. 7,18 In clinical studies, use of the conventional 1 H coils for 1 H MRI increases reproducibility and intersubject consistency. Therefore, in this study, we use mononuclear RF coil systems for 19 F MRI. The advantages of using mononuclear coils that are easily exchangeable to allow consecutive measurements on the same animal have also been demonstrated for preclinical setups at different field strengths. 35,36 As a design target, we also ensured that 1 H and 19 F coils can easily be exchanged without moving the animal.
The purpose of this study is to introduce an RF coil system for 19 F cardiac MRI at 3 T in a pig model. We developed an 8-channel Tx and 8-channel receive (Rx) coil array tuned to 115.9 MHz that can be used in combination with existing commercial 1 H RF coils. The coils were designed to have high transmit efficiency with homogeneous field distribution, and high sensitivity within the target volume of the heart. For Tx coil design optimization, finite-difference time-domain (FDTD) simulations were performed to compare 8-channel independent loop elements to a high-pass birdcage coil for different animal sizes. To improve Tx efficiency, a coil housing was constructed that can be adapted to animals of various sizes. An Rx coil array was designed to fit tightly on the curved abdomen of a large pig. The performance the 19 F RF coil system was evaluated in silico, in phantom, and ex vivo.

METHODS
In the following, the design of the 19 F RF coil system is described, which consists of an 8-element Tx volume coil array, and an 8-element Rx surface coil set distributed to anterior and posterior parts. All coil elements were designed as single-resonant systems operating at the 19 F frequency, and for the 1 H MRI measurements, a surface coil array with 18 loop elements arranged in The animal size may differ significantly. Simulation settings to test the effect of the size of the volume covered by the transmit (Tx) array are driven by real animal models. A transverse slice from the 3D-FLASH image of a large (A) and small pig (B). The Tx array element models covers a larger volume to enclose the large pig model, where, the heart is indicated with a darker color and around the heart, lungs are visible (C). Large pig large coil (LPLC) model (C). Small pig and small coil (SPSC) model (D). The distance between the neighboring coils in LPLC and SPSC configurations is measured as 104.6 mm and 89.3 mm, respectively 3 × 6 matrix form (Body 18; Siemens Healthcare) was used. To determine the optimal geometry of the RF coil system, T 2 -weighted 3D-FLASH 1 H images (voxel size: 2 × 2 × 3 mm 3 ) of previous animal experiments were segmented using 3D Slicer, 37 in which the pig is in a supine position on the patient table. In transverse images, the heart was very close to the anterior chest surface, and it extended from 4 to 13 cm into the thorax for the large animal and 3 to 11 cm for the small animal ( Figure 1A,B). For two different animal sizes (small and large pig), the heart, the lung, and the outer contour of the animals were manually segmented ( Figure 1C,D). Based on these geometric data, the following Tx coil array and tight-fitting Rx coil array designs were realized.

Transmit array
Based on the geometries of the thorax, an outer Tx coil array was designed for RF excitation that consists of eight independent loop coils (size: 6 × 15 cm 2 ; Figure 2D) symmetrically arranged in a circle (Ø = 42 cm) around the animal and the Rx coils, five of which are fixed on a detachable upper section made of a polycarbonate ( Figure 2C). The Tx loop coil elements were etched on a 1.6-mm-thick FR4 printed circuit board with a 105-μm-thick Cu layer ( Figure 2D). Each loop was segmented by four symmetrically distributed tuning capacitors, and it was matched to 50 Ω by series capacitors (25-8R2-G-2500-S; Voltronics). Variable capacitors (60-0752-15 005-901; Alfred Tronser) were added in parallel to the tuning capacitor at the feed port and the matching capacitors. Active detuning was implemented using a high-power PIN diode (MA4PK2000; MACOM). The coil elements were attached to the cylindrical coil former by 3D-printed (i3 Mk3S; Prusa Research) spacers ( Figure 2C). The spacers allowed modularity and scalability of the Tx array. For tight fitting of the coils also to the smaller animals, spacers allowed to adapt the inner diameter of the Tx array so that the Tx elements are always close to the thorax. The spacers were designed as detachable mounts that hold the Tx coil elements and can be slid and fixed to the rails on the coil holder. If the coil elements are mounted on 12-mm or 72-mm spacers, the inner diameter of the Tx array is scaled from 420 mm to 300 mm, respectively. Spacers also allowed modularity by simply removing the coil elements and configuring arrays with fewer channels using the same coil holder. The Tx coil was connected to a clinical 3T MRI system (Prisma Fit; Siemens Healthcare), which was equipped with a broadband RF power amplifier (8 kW peak power, 30-130 MHz, LPPA 13080 W; Stolberg HF-Technik) that can provide 3.2 kW peak RF power at 115.9 MHz with a maximum pulse duration of 3 ms. To split the RF power equally to each channel, an eight-way and a five-way high-power Wilkinson power dividers were designed so that the Tx coil could either be used with all eight channels, or with only the anterior five channels for cardiac applications ( Figure 2E). To drive the eight-channel/five-channel Tx coil in circularly polarized mode, 45 • /72 • phase increments between the adjacent channels were introduced by adjusting the length of the coaxial cables between the coil elements and the output ports of the power divider. A schematic diagram of the interface between the Tx coils and the MRI system is shown in Supporting Information Figure S1. S-parameters S ij , B 1 + homogeneity, and transmit efficiency of the Tx coil were measured. Printed circuit board layouts for the coil elements and the power divider are available at https://github.com/ozenEEE/ UKF_19FCOIL.

F I G U R E 2
Schematic and layout of the dedicated 19

Receive array
An inner Rx coil array with eight channels was constructed that consists of a six-channel semiflexible anterior and a two-channel rigid planar posterior array ( Figure 2B). Four of the anterior elements and the two posterior elements were constructed from enameled copper wire (∅: 1 mm) that were fixed in 3D-printed polyethylene terephthalate glycol-modified wire tracks. The elements were segmented by two tuning capacitors and series-matching capacitors (S111DUE; Johanson Technology), and variable capacitors (JR300 [Knowles Voltronics] or BFC280811339 [Vishay Electronic]) were added in parallel to the tuning capacitor at the feed port and the matching capacitors. The other two anterior Rx coil elements were shielded loop resonators (SLRs) 38-42 driven in self-resonance, 42 which were constructed using a 22-cm-long semirigid coaxial cable (SUCOFORM_47_CU; Huber+Suhner) with a single gap on the shield symmetric to the feed port. Active detuning was implemented using a PIN diode (MA4P7102F-1072T; MACOM) in parallel to the tuning capacitor at the feed port. Preamplifier decoupling was achieved by adjusting the length of the coaxial cable (G_02232_D; Huber+Suhner) between the Rx coil and the preamplifier (Stark Contrast). Detailed circuit schematics and component values can be found in https://github. com/ozenEEE/UKF_19FCOIL. A schematic diagram of the interface between the Rx coils and the MRI system is shown in Supporting Information Figure S2.
The anterior loop coil elements were arranged as a 2 × 2 matrix, in which the two elements in the first row were realized as loops with a diameter of 12 cm, and the loop elements in the second row were shaped to fit between the legs of a pig (see Figure 2B). Two SLR elements were placed between the loop coil elements ( Figure 2B). The neighboring loop coil elements in a column were decoupled by geometrical overlapping, whereas the elements in each row were decoupled capacitively. This decoupling strategy has two advantages: First, a better isolation is achieved as the next nearest neighbors are further away from each other (Supporting Information Figure S3). Second, capacitive decoupling is less sensitive to small changes in coil geometry, which allows us to bend the anterior coils around the animal's dome-shaped thorax to maximize the filling factor. The rigid posterior part consists of two geometrically decoupled rectangular coils of 8 × 14 cm size, which are fixed below a platform that is used for animal positioning ( Figure 2B).
Performance of the Rx coil array was characterized in the test bench using a network analyzer (ZVB4; Rohde & Schwarz) to measure the S-parameters, S ii , and in the MRI system to determine the ratio of loaded and unloaded quality factors, Q i L /Q i UL , both in a phantom and in vivo without changing the tuning and matching. For this, a tissue-mimicking phantom was prepared using 13.6 L distilled water with 3 g/L NaCl and 1 g/L CuSO 4 filled in a half-cylinder container (∅: 20 cm, height = 40 cm). Phantom liquid had ε r = 78.2 and σ = 0.46 S/m at 115.9 MHz measured with DAKS12 (Zurich Med Tech). The S-parameters were also measured in vivo inside the MRI system.

Simulations
To assess the effect of animal size, coil arrangement (tight vs distant placement with respect to the body) as well as amplitude, phase mismatches, and electromagnetic field simulations were performed on the Tx array elements using the FDTD solver of Sim4Life (v6.2; Zurich Med Tech). Simulation results were used to calculate the B 1 + homogeneity, Tx efficiency, and the maximum 10 g-averaged peak local specific absorption rate (SAR 10g ). Again, the two different animal models, a large pig (LP) and a small pig (SP), were used as shown in Figure 1C,D.
The lungs and the heart were modeled as separate uniform organs with isotropic dielectric properties of lung parenchyma and heart muscle, respectively, 43 and the rest of the body was assigned ε r = 78 and σ = 0.5 S/m. Coil conductors were modeled as perfect electric conductors. For the simulation, the coils were first tuned to 115.9 MHz with a series of multiport broadband simulations with a 100-MHz bandwidth. Once a reflection coefficient ≤ 15 dB was achieved for all of the coil elements, a harmonic simulation was performed. The Tx arrays were driven in CP mode under ideal matching conditions with the real parts of the input impedances equal to the source impedance. For comparison, an eight-leg high-pass birdcage coil of the same diameter was modeled and simulated. To demonstrate the scalability of the Tx array, a tight-fit Tx array for the smaller pig model was simulated in which spacers were introduced to bring the Tx array elements closer to the animal. In the simulations, B 1 + homogeneity, Tx efficiency, and SAR 10g were evaluated in the following nine scenarios: 1. Large pig model and large-diameter Tx array (LPLC); 2. Birdcage coil with the same diameter as the LPLC case, yet longer in z direction (rung length = 40 cm), evaluated with small and large pig models (LPBC and SPBC); 3. Small pig model large-diameter Tx array (SPLC); 4. Small pig model and tight-fit Tx array with spacers (SPSC); and 5. Both pig models, upper five and seven elements of the Tx array configurations (LPLC7x, LPLC5x, SPSC7x, and SPSC5x). B 1 + homogeneity was evaluated across the heart. As a figure of merit, peak deviation from the mean value within the heart was calculated. To test the effect of imperfections in the power splitting, differences in phase shifters, as well as reflections, B 1 + homogeneity was evaluated for various combinations of the Tx array elements with random amplitude and phase deviations within 20% of the CP-mode values (ie, equal amplitude and 360/N phase increment, where N = 5 or 8 is the number of Tx channels). Transmit efficiency was calculated as |B 1 + |/ √ P in in μT/ √ kW. All of the results were normalized to total input power of 3 kW (details of the simulation settings are provided in the Supporting Information. Pig models and the Python scripts used in analysis of the simulation results can be downloaded from https://github.com/ ozenEEE/UKF_19FCOIL. To test the scalability of the Tx coil system to higher field strengths, the individual coil elements were tuned to 19F Larmor frequencies at 7 T, 9.4 T, and 10.5 T at 280, 376, 420 MHz, respectively. Field simulations were performed for the SPSC case, and CP-mode results were compared with the 3T (116 MHz) results. Excessive B 1 + inhomogeneity is expected at ultrahigh fields already starting from 4 T. 44,45 To achieve a homogenous excitation at beyond 4 T, B 1 shimming is required. [46][47][48][49][50] To show that B 1 shimming can improve the homogeneity, an optimization algorithm was run with the phase values of the individual coils being the free parameters, and log the objective function calculated within a volume that covers the whole heart. The integrated optimization package of Sim4Life v7.0 was set up for a global search method based on Jones et al, 51 with maximum number of iterations limited to 300. The optimization routine started with the initial eight optimization variables with a 45 • phase shift. The RF shimming results for the optimal solution were incorporated in the postprocessing step, and the objective function was compared with the CP-mode results.

Magnetic resonance imaging measurements
A 19 F grid phantom filled with perfluoropolyether lubricant (Fomblin Y LVAC 16/6; Solvay Fluor) was prepared to assess the coil performance. The phantom consists of acrylic tubes (ø = 10 mm) that are distributed over a homogeneous 1 g/L CuSO 4(aq) body (300 × 200 × 400 mm 3 ) as shown in Figure 3I. The SNR analysis was performed using a 3D gradient-echo sequence with TR/TE = 8.4/4 ms, α = 30 • , 2-mm isotropic resolution, FOV = 240 × 320 mm 2 , 120 × 160 acquisition matrix, 120 slices, 12 averages, and 09:52 min:s total acquisition time. A 3-ms-long Gaussian RF pulse was programmed to excite only the central line of the perfluoropolyether spectrum so that chemical shift artifacts caused by the multiple 19 F-resonances are avoided. B 1 + maps were calculated using the double-angle method 52

F I G U R E 3
Reflection, coupling, and noise correlation matrices. The S matrix of the Rx array measured with phantom (A) and in vivo (B) loading. In the phantom, an Sii < −20 dB and an Sij < −12 dB were achieved, and in the animal an Sii < −14 dB and an Sij < −19 dB were achieved. Noise correlation matrix of the Rx array (C). Reflection and transmission coefficients of the Tx array elements for large phantom and large coil (D), small phantom and small coil (E), and small phantom and large coil (F). Photos of the large (G) and the small (H) coil, respectively. Photo of the large-grid phantom (I). The coils were tuned initially for the large-phantom and large-coil setting. No additional tuning and matching were applied using the same acquisition matrix for α 1 /α 2 = 20 • /40 • . To demonstrate modularity, the same measurements were also repeated using the upper five elements of the 19 F Tx array. A five-channel Wilkinson power splitter was also interfaced to the Tx output of the MRI system. Receive coil noise correlation matrix was computed from a noise-only scan. The 1 H MR images were acquired using the same imaging parameters as in the 19 F protocol, but with a single average and 1 × 1 × 2 mm 3 voxel size using the system's body matrix coil. For 1 H MRI, the 19 F coils were unplugged, the anterior 19 F Rx array and the upper part of the Tx array (ie, five elements) were removed, and the remaining coil elements were manually detuned by a battery-supplied detuning signal (Supporting Information Figure S4). To test the coil system under realistic conditions, a cadaver measurement was performed on a pig using a 3D-printed heart-shaped phantom with six perfluoroctylbromide (PFOB)-filled acrylic tubes that was introduced into the thorax after the heart was removed. During the measurements, because the phantoms and the cadaver were not moved for the 19 F and 1 H coil exchange, the data sets were intrinsically co-registered. For imaging, the same imaging protocol was used as the grid-phantom measurements. To demonstrate scalability, MRI of the heart phantom with PFOB tubes was acquired with the 12-mm and 72-mm spacers corresponding to LC5x and SC5x arrangements, respectively.
Finally, to test the detection sensitivity, we mixed two different PFCs, PFOB and FC-84 (perfluoroheptane, Fluorinert; 3 M), because PFOB is not soluble in water and only minimally soluble in oil (37 mM/L). The CF 2 Br peak of the PFOB is separated by 1.97 kHz from the closest peak of the FC-84. 53 The 5-ml Eppendorf tubes with PFOB/FC-84 volume ratio of 0.1, 0.25, 0.5, 0.75, and 1 (pure PFOB) were prepared. Exciting only the CF 2 Br frequency with a narrow-band pulse using the LC setup and the same receive arrays, the SNR of each tube was calculated 54 for the same imaging protocol as the phantom and cadaver tests.

RESULTS
In vitro and in vivo S ij matrices are shown in Figure 3. The S-matrix of the Tx array shows that the isolation between the elements is higher than 13 dB, and the input impedance of each coil is close to 50 Ω ( Figure 3A-C). The Wilkinson power divider introduced an additional insertion loss up to 1.5 dB at the output ports. The amplitude deviation from the ideal value was measured to be < 12% for both five-way and eight-way power dividers. Incremental phase for the CP mode deviated by < ±3.5 • , and ± 6.1 • for eight-way and five-way power divider, respectively. When the Tx coils are used with spacers (small pig arrangement; Figure 3B), reflection coefficients improve by up to 5.7 dB compared with the larger coil diameter ( Figure 3C) due to higher loading. For the Rx array, coupling between the elements was always below −16 dB without preamplifier decoupling, which improves the isolation by up to 18 dB. An isolation of 38 ± 8 dB was measured between the 19 F Tx and Rx elements. Between the phantom and in vivo, S parameters deviate < 10%, which demonstrates that the RF coil is not sensitive to load variations. The ratio, Q i L /Q i UL , was below 0.45 for all RF coil elements, which indicates that the sample noise is dominant for the Rx array, whereas for the Tx array most of the RF energy is absorbed by the load.
In Figure 4, simulated B 1 + maps for various pig and coil pairs are shown for a transverse slice across the center of the heart. Using seven-channel and five-channel coils increases Tx efficiency in the heart compared with eight channels for the same total input power. Scalability was demonstrated in simulations in which the coil elements were placed to form an array of a smaller diameter of 300 mm, and 40.2% higher Tx efficiency was computed for SPSC compared with the SPLC case.
Quadrature birdcage coils have a higher Tx efficiency for both small and large pigs. Compared with the SPBC and LPBC settings, SPSC5x and LPLC5X has up to 35% and 62% lower Tx efficiency, respectively. The B 1 + line profile analysis showed that the Tx homogeneity is better than 30% for all cases except for LPLC, in which up to 45% deviation from the mean is observed along the line profile, L1 (ie, the head-foot direction) (Supporting Information Figure S5). Field combinations with 20% random deviations from CP resulted in < 6% deviation in B 1 + homogeneity (Supporting Information Figure S6). For the smaller pig, birdcage and the smaller coil array have superior homogeneity in the longitudinal direction.
An axial slice from the B 1 + simulations for the eight-channel Tx array tuned for 7 T, 9.4 T, and 10.5 T are shown in Figure 4. At the ultrahigh frequency band (ie, 0.3-1.0 GHz), CP mode yields extremely inhomogeneous B 1 + with deviations up to 80 dB between the minimum and maximum B 1 + magnitude, even for a limited volume of interest ( Figure 5E). After phase-only RF shimming, the objective function (ie, log ) is improved by 44, 30, and 27 dB at 7 T, 9.4 T, and 10.5 T, respectively. The resulting phases assigned to each channel were ∅ 7T = {118,120,181,282,294,60,220,152} • , ∅ 9.4T = {127,4, 313,359,224,20,121,1} • , and ∅ 10.5T = {340,2, 81,224,359,1, 176,38} • . For the 9.4T and 10.5T simulations, the optimization algorithm did not converge; however, the phase values for the best objective among the tested cases was assigned.
In Table 1, peak local SAR values at 3-kW input power are given for all pig sizes and coil pairs. As expected, when the coil elements are further away from the body in the SPLC case, the peak SAR 10g decreases. In the SPSC arrangements, the distance between the coil elements and the body is slightly larger than the LPLC case (∼5 mm). Thus, consistently lower peak SAR10g was calculated for SPSC than for LPLC. For lower number of Tx array elements, the peak SAR 10g is also higher as the peak RF power per coil increases, with the exception of SPSC5x, which has the lowest peak SAR 10g . The

F I G U R E 7
Ex vivo 19 F MRI using the eight-channel Tx and eight-channel Rx coil system. Photo of the cadaver measurement setup (A). The 3D-printed phantom rendered from high-resolution heart images of a large pig with six perfluoroctylbromide-filled tubes (B). Transverse slices of the cadaver, in which the 19 F image (green) is overlaid on the 1 H gradient-echo (gray) image (C). The B 1 + maps for SPSC5x (D) and SPLC5x (E) configurations are shown to verify simulation results that demonstrate scalability measured and simulated B 1 + maps, respectively. When only the upper five elements were used (ie, LPLC5x), Tx efficiency was 2.29 ± 0.39 μT/ √ kW versus 2.41 ± 0.37 μT/ √ kW in the measured and simulated B 1 + maps, respectively. The higher Tx efficiency for the lower number of channels is also visible in the axial slices from the measured B 1 + maps given in Figure 6B,C. For SPSC, SPSC7x, and SPSC5x cases, Tx efficiency increased from 2.91 ± 0.34 μT/ √ kW to 3.55 ± 0.57 μT/ √ kW and 4.48 ± 0.81 μT/ √ kW, respectively. Only eight-channel and five-channel large coil arrangements were tested experimentally due to the grid-phantom size. Note the reduced sensitivity in the central tube in the top row, which can be attributed to the lower Rx coil sensitivity along the center of the anterior coil array (Figure 6). Scalability was also demonstrated using the heart-shaped PFOB phantom ( Figure 7D,E), in which the phantom size corresponds to a small pig and the SPSC5x and SPLC5x configurations were compared. A similar result as the simulations, comparing the SPSC5x and SPLC5x cases, was also obtained by phantom MRI measurements. Here, the SC5x case yielded 31.3% higher transmit efficiency than LC5x for the PFOB heart phantom ( Figure 7D,E). No coupling between the 19F Tx coils and the integrated 1 H body coil was observed.
For the 1 H imaging, only the upper five elements of the Tx array and the flexible Rx array placed on the top could be removed. A decoupling over 30 dB between the 1 H and the remaining 19 F elements was measured. To determine the reproducibility of the 19 F signal, 10 acquisitions were then performed using the transmitter voltage of 64 V. For this reference transmitter voltage, a mean SNR of 79 ± 3 was observed for the 19 F signal from the perfluoropolyether liquid at the central transverse slice. This translates into a signal variation of 3.8%, assuming the noise is constant.
The PFOB-filled vials could be clearly identified inside the cadaver with an SNR of 275 ± 51 for a 3D gradient-echo sequence with 2-mm isotropic resolution and 12 averages, acquired in 9:52 min:s. The overlay of 19 F images (green) and 1 H anatomical images (gray) provides visualization of the PFOB at the area of myocardium (Figure 7).
In the sensitivity tests, the image SNR was < 10 for 10% and 25% PFOB solutions (Supporting Information Figure S7). Detection sensitivity is calculated as 2193/mol in image SNR. Note that this measurement accounts only for the CF 2 Br peak.

DISCUSSION
In this study, an eight-channel Tx/eight-channel Rx RF coil array was developed for 19 F MRI of large animals at 3 T. The coil setup can be used in combination with clinical 1 H coils by allowing removal of the 19 F coils during 1 H imaging without moving the animal. The RF coil array was tested in a grid phantom and an ex vivo pig model under various loading conditions. A modular and scalable design scheme was introduced using spacers to adapt the Tx coils to various animal sizes. The proposed 19F coil system differs from other coil designs in that it combines the use of spacers for the Tx coil elements to be able to adapt the circumference of the Tx array with a modular design, where the number of Tx elements can be changed, and a receive coil system with different decoupling techniques (capacitive and geometric) having a hybrid design made of loops and SLRs. The proposed coil system can be used, for example, to investigate MI. Myocardial infarction is one of the main causes of mortality and long-term disability worldwide. 55 After MI, the imbalance of the monocytes' inflammatory and reparative response leads to aggravated myocardial damage and suboptimal tissue repair. Currently, novel therapeutic strategies are being investigated that modulate monocyte migration and response to MI, to limit ischemic damage and to promote healing. Thus, there is an urgent clinical need to develop and validate new in vivo readouts of monocyte migration and response after MI as a potential prognostic marker of cardiac outcomes, or to characterize the efficacy of new immune-modulating compounds. The proposed coil can also be used to track rapid myeloid cell egress from the spleen and bone marrow. 56 The results show that the Tx array driven in CP mode provides a homogeneous excitation field over the heart region with < 15% deviation from the mean. The modular design enabled us to use five channels as well to increase the RF power per channel, thereby improving the transmit efficiency while keeping the B 1 + homogeneity within 19% of the mean. Decreasing the size of the coil by replacing the thin by thicker spacers allowed us to scale the coil to the animal size. The results showed that 31.3% higher B 1 + can be obtained when Tx coil array is arranged to form a cylinder with a smaller diameter.
As seen in the FDTD simulations, a homogeneous B 1 + distribution can be created over the heart of an adult pig with an array of local Tx loop coils. In combination with an Rx coil array, also a high SNR can be achieved if the Rx coil shape conforms to the anatomy of a pig in supine position. More detailed modeling of pig anatomy and coils could also increase the accuracy. In Carluccio et al, 57 it was reported that anatomical modeling with fewer tissue segments does not result in significant changes in local SAR; therefore, it might be preferred to reduce computational overload. 1 H imaging is not significantly affected (< 5% change in regional SNR; Supporting Information Figure S8) by the presence of the 19 F coils. The anterior part of the Rx and Tx coils can be removed without moving the animal, whereas the posterior elements are unplugged and detuned during 1 H MRI. Thus, 19 F MRI can be performed as an add-on experiment, for example, to an MR-guided interventional procedure, and commercially available 1 H coil arrays can be used for 1 H imaging. A double-resonant coil was not preferred, as the combination of different resonance frequencies leads to a suboptimal performance.
The 19 F MRI system can be upscaled or downscaled for human measurements. The size of the Rx elements could be reduced for humans to increase the receive sensitivity. Similarly, the Tx coil efficiency can be increased by placing the coil elements closer to the patient. This concept was tested for decreasing the diameter of the Tx coil holder for the small pig model from 420 mm to 300 mm, and the Tx efficiency was increased by 40.2% around the center of the heart. The size of the coil was adapted by using adjustable coil holders such as in Lopez Rios et al. 58 This approach is especially suitable for the proposed Tx coil system, as each coil element is electrically and mechanically separated.
FDTD simulations with random deviations from the ideal CP mode showed that the Tx array is robust against phase and power deviations at the input ports. Although the Tx efficiency of quadrature birdcage coils is higher than that of the Tx array, the modular, scalable Tx array has several advantages: It is simple to build by 3D printing; it does not need electrical or mechanical connection between the elements; coil elements can be tuned independently and individually; and it can potentially be used with parallel Tx for RF shimming and local SAR management. 47,50,59 A birdcage coil can also be designed in multiple segments that can be attached together with robust galvanic connections between the pieces (eg, two half cylinders), yet this would require other means of production equipment. However, we think a modular birdcage structure can also be constructed using rapid prototyping facilities such as printed circuit board etching and 3D printing, and being able to be replaced without moving the subject inside would be very useful; however, this is beyond scope of this study. Unlike birdcage coils, this coil can also be adapted to higher fields after retuning, without further modifications, as shown in Figure 5.
In both simulations and MRI experiments, the five-channel Tx array could provide sufficient Tx homogeneity within the heart and with the advantage of increased maximum B 1 + . This is especially useful for nuclei such as 19 F, for which only a limited peak RF power was available on the clinical MRI system we used (ie, 3.2 kW). If other organs such as the spleen are studied, the arrangement with reduced Tx channel needs to be further investigated.
The proposed coil array opens design options for densely packed multichannel transmit 19 F RF coil arrays, even at higher magnetic field strengths. Future solutions will involve designs that use different building blocks for signal transmission and reception, to balance the constraints dictated by sensitivity, depth penetration, transmission field uniformity, SNR, and SAR. 60 In this context, a recent report suggested the combination of dipole antenna and loop elements. 61 At 3 T, CP mode with fixed amplitude an phase distributions resulted in satisfactory β 1 + homogeneity values with a maximum deviation of 17%, even for the five-channel Tx array arrangement. Our 19 F phantom studies with PFOB showed a steady 19 F signal over a series of independent measurements with a signal variability of 3%. A closer examination revealed a detection limit of 3.9 mmol (or 23 × 10 20 PFOB molecules) for the proposed setup, accounting only for the CF 2 Br spins.
In all measurements, a narrow-band RF pulse was used, which excited only the central resonance and ignored the other peaks in the spectrum. For complex 19 F molecular structures, optimized sequences that can excite the whole spectrum using larger RF bandwidths and prevent chemical shift artifacts at the same time are required to increase SNR. One possible approach could be to perform sine-bell acquisition-weighted CSI with selective refocusing pulses. 62 Similar to the scalable Tx array idea, the proposed Rx array can also be designed in a scalable form to increase SNR from the target volume such as demonstrated in Lopez Rios et al. 58 Although the anterior part of the current Rx array has a semiflexible design and is placed directly on the abdomen, maximizing the filling factor, the size and geometric extent of the Rx array configuration can be modified to bring all of the elements closer to the heart region, even for smaller animals. This can be accomplished by designing all of the array elements as flexible modular coils, 42,63 or by keeping the four loop coils as the core structure and using the remaining two SLR elements as separate portable elements that can be placed closer to the heart in smaller animals. We designed the size of the proposed Rx array elements by considering the geometric extent of the heart from the anterior surface. This distance is smaller in smaller animals. A more straightforward approach would be to construct a downscaled version of the existing anterior part of the Rx array for smaller animals.

CONCLUSIONS
The proposed modular eight-channel 19 F Tx and the eight-channel semiflexible Rx RF coil array is an important step to realize 19 F-MRI experiments in the heart, to assess the biodistribution and bioavailability of fluorinated drugs. The modular design allows us to adapt the coil array to pigs of various sizes, improving the Tx coil performance. The 19 F coil can be combined with existing 1 H coil equipment, which is advantageous in complex studies involving MR-guided interventions and functional imaging.

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher's website.
FIGURE S1: Schematic diagram for interfacing the transmit (Tx) array to the MRI system. A coil plug (Tx plug) was used to deliver the RF power from a broadband RF power amplifier (RFPA), as well as the DC bias voltage for active detuning of the individual Tx coil elements, as shown in the detuning circuit for the first element. The detuning logic is predefined in the coil files, which is controlled through the RF excitation and signal acquisition entries in the pulse sequence. To avoid excessive cabling, the DC signals were coupled with the RF paths, leading to the individual Tx coil elements at the power splitter FIGURE S2: Schematic diagram for interfacing the receive (Rx) coils to the MRI system. A coil plug (Rx plug) was used to deliver the amplified MR signal to the receiver unit of the MRI system, as well as the DC bias voltage for active detuning of the individual Rx coil elements. Detuning circuits for loop and shielded loop resonator (SLR) elements are shown for the first elements. An additional cross-diode is used at the preamplifier inputs (not shown) after the DC-blocking capacitor, C dc , for passive protection. Preamplifiers were biased through the "MR signal" path, which is a coaxial cable with a DC signal coupled, which is activated during the receive cycle. The detuning logic is predefined in the coil files, which is controlled via the RF excitation and signal acquisition entries in the pulse sequence. To avoid excessive cabling, the DC signals for active detuning were also coupled with the RF paths on the coil side FIGURE S3: Comparison of the next-nearest-neighbor distances in the four-loop array 2 × 2 configurations. A, B, The proposed approach in (A) results in a distance of 5.5 cm between L11 and L22, whereas the arrangement in (B) a distance of 1.4 cm. Additional decoupling of 4.3 dB between L11 and L22 is gained by choosing (A) over (B) FIGURE S4: A schematic diagram of the battery-supplied detuning used for the unplugged 19F coils that are left inside the magnet bore during 1 H measurements. The battery output is simply connected to the coaxial cable of the 19F coil elements, thereby activating the detuning PIN diode to minimize coupling with the 1 H coils FIGURE S5: Line profile results of the simulations for various pig model and coil pairs in the ideal input matching case normalized to 1-kW input power. Magnitude of the B 1 + field was plot along the lines in L1 (head-foot), L2 (anterior-posterior), and L3 (right-left) directions for each combination FIGURE S6: Effect of the phase and amplitude mismatches on the B 1 + homogeneity. Thirty-two random amplitude and phase values within 20% of the ideal values were used to calculate field distributions and maximum deviation from the mean along the lines in L1 (head-foot), L2 (anterior-posterior), and L3 (right-left) directions FIGURE S7: PFOB was mixed with FC-84 (perfluoroheptane) to test sensitivity of the coils by exciting only the CF2Br peak of perfluoroctylbromide (PFOB) using a narrow-band (3 kHz) Gaussian RF pulse. The image SNR for a 2-mm isotropic FLASH sequence (32 averages) was < 10 for the 5-ml tubes with 10% and 25% solutions FIGURE S8: Comparison of 1 H gradient-echo (GRE) images with and without the remaining 19F elements, which were actively detuned. No significant artifacts or sensitivity change was observed with structural similarity index (SSIM) of 0.9913 and normalized mean square error (NMSE) of 0.55%