Simulation guided design of the MRcollar: a MR compatible applicator for deep heating in the head and neck region

Abstract Purpose To develop a head and neck hyperthermia phased array system compatible with a 1.5 T magnetic resonance (MR) scanner for noninvasive thermometry. Methods We designed a dielectric-parabolic-reflector antenna (DiPRA) based on a printed reflector backed dipole antenna and studied its predicted and measured performance in a flat configuration (30 mm thick water bolus and muscle equivalent layer). Thereafter, we designed a phased array applicator model (‘MRcollar’) consisting of 12 DiPRA modules placed on a radius of 180 mm. Theoretical heating performance of the MRcollar model was benchmarked against the current clinical applicator (HYPERcollar3D) using specific (3D) head and neck models of 28 treated patients. Lastly, we assessed the influence of the DiPRA modules on MR scanning quality. Results The predicted and measured reflection coefficients (S11) of the DiPRA module are below −20 dB. The maximum specific absorption rate (SAR) in the area under the antenna was 47% higher than for the antenna without encasing. Compared to the HYPERcollar3D, the MRcollar design incorporates 31% less demineralized water (−2.5 L), improves the predicted TC25 (target volume enclosed by 25% iso-SAR contour) by 4.1% and TC50 by 8.5%, while the target-to-hotspot quotient (THQ) is minimally affected (−1.6%). MR experiments showed that the DiPRA modules do not affect MR transmit/receive performance. Conclusion Our results suggest that head and neck hyperthermia delivery quality with the MRcollar can be maintained, while facilitating simultaneous noninvasive MR thermometry for treatment monitoring and control.


Introduction
Hyperthermia has proven beneficial when added to standard radiotherapy and chemotherapy for treatment of many tumor locations, including the head and neck (H&N) [1][2][3][4]. In order to elevate the temperature in the H&N region to the desired range of 40-44 C, we have previously developed the unique HYPERcollar3D applicator consisting of 20 patch antennas operating at 434 MHz [5,6]. At Erasmus MC, hyperthermia using the HYPERcollar3D is being applied once a week for 75 min after the radiotherapy treatment [7]. The temperature during the hyperthermia is measured superficially and, whenever possible, interstitially by sets of fiber optic probe sensors placed in closed tip catheters. Magnetic resonance (MR) imaging has shown to have a clinical potential for noninvasive temperature measurement in the pelvic region, however no device exists for MR-hyperthermia in the H&N [8,9]. In 2012, we therefore started to study the feasibility of integrating the hyperthermia H&N device into an MR scanner for noninvasive MR thermometry by developing a novel applicator: the MRcollar. Integrating a hyperthermia system into the MR requires the use of nonmagnetic components and design principles that minimize system-induced image artifacts. One example for this is the minimization of large electrically conductive surfaces that can generate undesired gradient-induced eddycurrents or shield certain regions of interest from the MR radiofrequency (RF) transmit fields. Previously, we redesigned the HYPERcollar3D and experimentally showed the feasibility of applying hyperthermia in a 1.5 T GE 450w MR scanner using the 'MRlabcollar', a laboratory prototype consisting of 12 patch antennas operating at 434 MHz [10][11][12]. In those measurements, we observed that the connectors of the antennas caused image distortion in the MR thermometry images, even if the effects were located on the periphery of the water bolus structure and not within the region of interest. Motivated by the negligible MR distortions found for the self-grounded bow-tie antenna, we developed the Yagi-Uda antenna, which has a smaller cross-section of metal obstructing the B 1 þ field [13][14][15].
Noninvasive MR thermometry using the MR scanner is clinically available for deep hyperthermia in the pelvic region using the Sigma Eye system (Pyrexar Medical, Salt Lake City, USA) [16,17]. Sigma Eye and MRlabcollar use independent systems for MR thermometry and hyperthermia heating. This differs from other approaches where the hyperthermia and MR imaging functionalities are embedded into a common antenna structure. It was demonstrated that a 3 T MR surface loop coil array can be modified in real time into a C-shaped dipole antenna phased array for hyperthermia heating using fast RF switches [18,19]. As an alternative, Winter et al. proposed to use a bowie-tie antenna system for imaging and heating at 7 T (298.6 MHz) [20,21]. Both these approaches use the MR scanner RF amplifiers for heating and thus don't allow truly simultaneous heating and imaging. Further they also restrict the heating frequency to the Larmor frequency, which is often not the optimum choice considering focusing and operational costs. Hence, the optimum choice might be an appropriate co-design of a heating antenna array and imaging coil arrays. We selected 434 MHz as operating frequency as it is within the band of optimum frequencies [22] and an ISM (industrial, scientific and medical) frequency for which low-cost amplifiers are available. This frequency is also sufficiently far from the Larmor frequency (64 MHz) and higher order modes of a 1.5 T MR scanner, which enables decoupled operation for truly simultaneous noninvasive MR thermometry [9].
The purpose of this study was to use 3D electromagnetic (EM) field modeling to design a clinical applicator prototype for MR guided deep hyperthermia treatment in the H&N region. In the first step, we designed the dielectric parabolic reflector antenna (DiPRA) module. In the second step, we used a dedicated measurement setup to verify the predicted reflection coefficient and radiation characteristic of the DiPRA module. Based on this module, we then created a phased array applicator (MRcollar) model and tailored its dimensions to the average H&N patient. Forth, the predicted heating capabilities of the MRcollar design were evaluated against the clinically used HYPERcollar3D applicator [23]. Finally, MR experiments were carried out to assess the influence of the DiPRA modules on MR scanning quality.

Methods
Antenna encasing -DiPRA module , for which heating performance and MR compatibility were verified in a previous study [13]. The antenna was manufactured at an FR-4 printed circuit board and operates at 434 MHz. It is partly submerged in the de-mineralized water bolus, which allows reducing the antenna size, efficient EM energy transfer and cooling of the patient surface. The director of the printed circuit board (PCB) Yagi-Uda antenna (Figure 1(a)) was removed to reduce axial size and because it induced significant sensitivity in impedance characteristics to the cavity enclosing. This removal resulted in a reflector backed dipole antenna shown in Figure 1(b). To reduce the amount of water needed, minimize the cross-coupling and improve focusing, we designed a specific encasing with a parabolic back shape shown in Figure 1(c), a dielectric parabolic reflector antenna (DiPRA). The width of the DiPRA encasing (w e ), shown in Figure 1(c), is determined by the 70 mm printed reflector backed dipole antenna width. The height (h e ) was selected such that the specific absorption rate (SAR) in the muscle equivalent phantom was not influenced by this dimension. The distance between antenna and cavity top (d a-e ) and the thickness of the top encasing part (d et ), shown in Figure 1(d), were selected to have a minimum influence on the antenna reflection characteristics. We placed the DiPRA module on a 30 Â 300 Â 300 mm 3 water bolus and a 100 Â 300 Â 300 mm 3 muscle equivalent phantom model, and studied reflection coefficient and SAR characteristics. For the FR-4 substrate we applied e r ¼ 4.66, r ¼ 1.9 mS/m, q ¼ 1850 kg/m 3 dielectric properties, for the antenna encasings e r ¼2.6, r ¼ 4 mS/m, q ¼ 1180 kg/m 3 , for the water bolus e r ¼80, r ¼ 0.04 S/m, q ¼ 1000 kg/m 3 and for the phantom e r ¼56.9, r ¼ 0.81 S/m, q ¼ 1090 kg/m 3 .

DiPRA module verification measurements
To validate this embedded antenna concept, we created a dedicated measurement setup in Sim4Life, as shown in Figure 2(a). Figure 2(b) shows the manufactured setup using polyethylene terephthalate glycol (PET-G) filament printed with 0.4 mm nozzle diameter, 0.25 mm layer height and 100% filling factor at Prusa i3 MK3 (Prusa Research, Prague, Czech Republic) 3D printer. A two-component epoxide glue (Chemex POX Z 21, Prague, Czech Republic) was applied for antenna fixation and coating of the 3D printed part ensuring water resistance of the entire measurement setup. We also 3D printed a 30 mm height frame for the water bolus filled with demineralized water. For the reflection coefficient measurements, we created a phantom with muscle equivalent properties at 434 MHz consisting of demineralized water, salt and isopropyl alcohol. The concentrations of the individual components were adjusted in a few iterative steps using dielectric assessment kit DAK-12 (SPEAG, Z€ urich, Switzerland) connected to N9923A FieldFox vector network analyzer (Keysight, Santa Rosa, USA). This procedure resulted in a liquid phantom with a relative permittivity of e r ¼56.4 and an electric conductivity of r ¼ 0.79 S/m. These values were very close to the literature values for muscle of e r ¼56.9 and r ¼ 0.81 S/m used for the hyperthermia treatment planning purposes within this study. The reflection coefficient characteristics were measured for different encasing top thicknesses, ranging from 0.5 mm to 1.5 mm, using a liquid phantom and a FSH8 (Rohde&Schwarz, Munich, Germany) vector network analyzer.
For the SAR measurements, we prepared a solid phantom following the procedure recommended for superficial hyperthermia quality assurance [24]. The phantom consisted of demineralized water, salt, agar powder, TX-151 and polyethylene powder and its dielectric properties (e r ¼66, r ¼ 0.92 S/m) were measured using DAK-12. The phantom was filled in eight 10 mm 3D printed frames which were placed on the top of another inside the 3D printed container. After heating period, the whole phantom was taken out from the container and two 2D profiles in depth of 12.8 mm and 24.5 mm depths measured by removing top phantom layers. These thicknesses correspond to the manually measured thicknesses of the first and second phantom layers after phantom creation. The SAR characteristics were measured using E60 FLIR (FLIR Systems, Wilsonville, USA) for the scenario without antenna encasing cover and for the scenario with a 1 mm thick cover. The 75 W harmonic signal at 434 MHz was excited using a PG 70.150.2 (SSB Electronics GmbH, Lippstadt, Germany) power generator and guided via a ZGBDC30-372HPþ (MiniCircuits, Brooklyn, USA) bi-directional coupler toward the DiPRA panel SMA type connector. Forward and reflected power were acquired using two PWR-6GHZ (MiniCircuits, Brooklyn, USA) power meters connected with 20 dB additional attenuation to the outputs of the bidirectional coupler. Its attenuation and coupling characteristics were measured at the beginning of the experiments with a  FSH8 vector network analyzer. We calculated SAR using the specific heat capacity (c ¼ 3640 J/kg/K) multiplied by the empirical rate of temperature rise. The latter is computed from the ratio of measured temperature increase (subtracted 2D infrared images after and before the heating) and the measurement time (four minutes). SAR was normalized to 1 W input power to enable comparison of predicted and measured SAR distributions. The predicted 2D SAR profiles were extracted at corresponding depths from homogenous phantom model. To provide more evenly distributed water bolus pressure to the patient and a more homogenous water circulation throughout the water bolus, it is desirable to keep the water bolus thickness constant for the whole MRcollar model. We approximated the surface of the representative patient with a radius of curvature of 150 mm, to which we added 30 mm of water bolus thickness. This resulted in the final 180 mm radius of the both 2 Â 3 arrays parts of the MRcollar model shown in Figure 3(b). A rounding with a radius of 180 mm was applied to the footprint of the antenna modules in order to provide better contact between water bolus and the modules.

MRcollar model heating capabilities
Heating capabilities of the clinically used HYPERcolar3D and the MRcollar model were evaluated on data from 28 H&N patients for which hyperthermia treatment planning (HTP) was previously done. The patient group consisted of 14 oropharynx, five neck node, three nasopharynx, three larynx, two oral cavity and one hypopharynx tumors, i.e., 19 males and nine females with a mean age of 63.1 ± 11.2 years. Patient specific 3D H&N models were created from computed tomography (CT) images using an atlas based automatic segmentation routine followed, if necessary, by manual adjustments in iSeg (v. 3.8, Z€ urich MedTech AG, Z€ urich, Switzerland) [25]. Then the patient models were imported into Sim4Life, together with the HYPERcollar3D or the MRcollar model, for EM field simulations. All dielectric properties were assigned from the IT'IS database available in Sim4Life at 434 MHz [26]. For the HYPERcollar3D models, we Afterwards, all simulations were exported to VEDO, our inhouse developed visualization tool for electromagnetic dosimetry and optimization [27]. VEDO allows pretreatment and online optimization of the amplitude and phase feeding signals of individual antennas in order to maximize the EM field energy focus in the target region. It uses the particle swarm optimization toolbox to optimize target-to-hotspot quotient (THQ) representing a ratio between the average SAR in the target and the average SAR in the hotspot (total volume with the highest 1% SAR outside the target) [28,29]. In addition to THQ, we also compared the target volume coverage of the 25% and 50% SAR iso-contour, TC25 and TC50.

MR assessment of antenna encasing -DiPRA module
We investigated the influence of the DiPRA module on MR quality following the tests recommended in Hoffmann et al. [30]. Three comparative tests were used to assess the absence of field distortions on B 0 and B 1 þ as well as SNR

Results
Antenna encasing -DiPRA module Figure 4(a) shows the influence of the distance between the antenna and the encasing cover d a-e (mm) on the antenna reflection coefficient, with the minimum S 11 ¼ À44.6 dB obtained for a distance of 5 mm. We selected d a-e ¼4.5 mm, to account for a larger distance in the curved antenna encasing, ensuring good matching (S 11 ¼ À32 dB). Figure 4( Figure 5(a) shows the normalized SAR depth profiles for the reflector backed dipole antenna placed in: (1) a flat (brick-shaped) volume of deionized water, (2) encasing without the cover and (3) in full encasing, i.e., the DiPRA. In Figure 5(a), penetration depths equal to 20 mm are observed for the flat water model, as are penetration depths of 21 mm for both scenarios with encasing. These values were calculated following the single antenna guidelines of ESHO [24]. When introducing the encasing around the reflector backed dipole antenna, the maximum SAR increased by 47% in a muscle equivalent phantom, from 3.4 W/kg to 5 W/kg for 1 W input power. The normalized 2D SAR distribution at 10 mm depth for the DiPRA module in the phantom is shown in Figure 5(b). Figure 5(c,d) show the normalized SAR profiles at 10 mm depth for all three studied scenarios along Y axis and Z axis, respectively. The encasing without the aperture cover had the best SAR symmetry from all three studied scenarios at 10 mm depth of the phantom. A 6 mm maximum SAR shift in -Y direction was observed for the DiPRA module. Figure 6(a) shows the reflection coefficient comparison for the prediction and two measurements of the DiPRA. For all cases, S 11 is below À20 dB, which represents maximum reflection of 1% from the input power. Figure 6     Normalized 1 g SAR predictions and verification measurements for the DiPRA module without and with the 1 mm top cover are shown in Figure 7(a,b) (with) and Figure 7(c,d) (with). The distributions were normalized to the maximum SAR in the 2D plane at 12.8 mm phantom depth, i.e., the height of the first 3D printed frame filled with the agar phantom. For the DiPRA scenario without the top cover, we obtained comparable predicted (2.04 W/kg) and measured (2.09 W/kg) maximum SAR. For the scenario including a 1 mm top cover, the measured maximum SAR (2.41 W/kg) was higher than predicted (2.01 W/kg). In this case, measurement (Figure 7(d)) confirmed the predicted SAR shift when adding 1 mm top cover. Figure 8 shows the corresponding normalized SAR figures at 24.5 mm depth in the agar phantom. At this depth, predicted SAR is higher than measured SAR for both studied DiPRA scenarios.

MRcollar design
For 28 simulation setups we obtained a mean water bolus volume for the HYPERcollar3D of 8.0 ± 0.3 L and 5.5 ± 0.3 L for the MRcollar design. For 22 adjacent DiPRA modules inside the array and all 28 HTP setups, we obtained a cross-coupling of À23.6 ± 5.1 dB. The maximum cross-coupling of À14.9 dB was observed for the two adjacent modules in the top left corner of the 2 Â 3 array. Figure 9 shows the THQ for the MRcollar against the HYPERcollar3D model for 28 H&N patients. For the HYPERcollar3D a mean THQ ¼ 1.25 ± 0.39 was obtained, and 1.23 ± 0.3 was the mean THQ for the MRcollar model, demonstrating comparable heating capabilities of both systems. Figure 10(a,b) show the comparison of target coverage in terms of TC25 and TC50. Higher mean values for the MRcollar model of TC25 ¼ 87.8 ± 13% and TC50 ¼ 55.5 ± 23.8% in comparison to TC25 ¼ 83.7 ± 15.6% and TC50 ¼ 47 ± 22.1% obtained for the HYPERcollar3D demonstrate improved target coverage for the MRcollar design.  of DiPRA modules had a standard deviation of 8.3 Hz within the ROI. The presence of two DiPRA modules improved B 0 field homogeneity (standard deviation ¼ 7.9 Hz). The mean flip angle within the ROI was always 32.5 ± 1.3 and hence independent of the presence of DiPRA modules. Finally, the image SNR in the ROI was 156 dB, when no DiPRA modules were present, while the image SNR was slightly higher when one or two DiPRA's were present (163 dB). Overall, a negligible impact on B 0 homogeneity, average flip angle and SNR were observed.

Discussion
The MRcollar design improves the predicted SAR target coverage (TC25¼ þ4.1% and TC50¼ þ8.5%) and achieves a comparable target-to-hotspot quotient (THQ ¼ À1.6%) when comparing it to our clinically used HYPERcollar3D system. These improvements are present even though the MRcollar design has two antenna rings in comparison to three rings of HYPERcollar3D. The reflector backed dipole antenna shows an excellent predicted and measured matching, i.e., S 11 <À20 dB, when placed in a specifically designed dielectric encasing. Cross-coupling between two adjacent antennas is predicted to be always better than À14.9 dB, ensuring  independence of the individual channels. Encasing the DiPRA also improves symmetry in the z-direction but adding the cover of the antenna encasing causes a 6 mm shift of the maximum SAR in 10 mm depth of muscle equivalent phantom. This predicted SAR shift when adding 1 mm top cover was verified using a dedicated measurement setup. It is  caused by the y-components of the electric field that are generated since the two dipole 'arms' are on different sides of the PCB. This phenomenon is small but may impose a need to ensure a matching orientation of the antennas inside the Sim4Life model and the manufactured device. Unfortunately, the required technology to perform the ultimate experiment, i.e., microwave heating with the DiPRA modules while simultaneously measuring the temperature distribution using MR-thermometry measurements, are presently not available. We are currently in the process of implementing additional 434 MHz filtering in the MR-signal readout pathway as well as obtaining sufficient 434 MHz amplifiers to enable the final proof of functionality and future clinical hyperthermia application using the MRcollar for patients with H&N cancer. Nevertheless, the MR experiments presented here combined with those earlier reported by Sumser et al. [35] do illustrate that the DiPRA modules do not alter MR transmit/receive performance. For the MRcollar design, we decided to replace the manually manufactured HYPERcollar3D patch antennas by reflector backed dipole antennas designed on a PCB. A downside of this antenna is that very small FDTD grid steps, i.e., 0.75 mm and 0.5 mm, are necessary to correctly resolve the thin metallic antenna parts and the antenna feeding port. This increases the computation time for each simulation approximately two times in comparison to those for the HYPERcollar3D. However, we expect that ongoing improvements in graphical processor units will allow us to complete the whole treatment planning process in around two hours, which is sufficiently fast for clinical HTP modeling. Most importantly, the antenna orientation parallel to the radiated RF pulse of the MR birdcage coil indeed ensured a limited flip angle (B 1 þ ) distortion within the antenna encasing.
The maximum SAR of the single reflector backed dipole antenna module inside the muscle equivalent phantom was increased by 47% when introducing the encasing around the antenna, leading to higher power delivery at the target region. This will decrease the maximum power requirements for the microwave amplifiers, which is the most expensive sub-system in RF hyperthermia systems. The antenna encasings also reduce the amount of water needed in the system by 2.5 L (31%) in comparison to the HYPERcollar3D, making it lighter and thus easier to install. The DiPRA modules also enable an independent water circulation system for the antennas from the water bolus circulation system. In this way, the temperature inside the antenna cavities can be kept constant while the temperature in the water bolus can be tailored to the superficial cooling required. When changing the water temperature by 1 C, the resonant frequencies for the patch and waveguide antennas shift by 1.1 MHz and by 0.89 MHz respectively [33,34]. A lower water volume will also reduce water circulation-based MR distortions. In addition, the space between the DiPRA modules allows integration of an MR surface coil array close to the patient surface [35]. Such a surface coil array will increase imaging SNR and potentially enable multi-coil acceleration techniques to obtain MR thermometry maps with higher temporal resolution. With these combined approaches, the accuracy and motion-robustness of MR thermometry in RF hyperthermia treatments is expected to improve.

Conclusion
Our results show that the novel MRcollar design enables achieving comparable heating capabilities as compared to the current clinical HYPERcollar3D. The DiPRA antenna module provides excellent predictive and measured matching characteristics and also predicted cross-coupling between adjacent antennas is predicted to be always better than À14.9 dB. The MRcollar design also reduces the amount of water needed by 31% (2.5 L), which will reduce applicator weight and water circulation-based MR distortion. The flip angle map distortions caused by the DiPRA module are found to be restricted to within the module itself. The H&N hyperthermia delivery quality with the MRcollar can be maintained, while facilitating simultaneous noninvasive MR thermometry for treatment monitoring and control.