A phased array applicator based on open ridged-waveguide antenna for microwave hyperthermia

Radiative hyperthermia is a clinically applied cancer treatment modality where antenna design is crucial to achieving therapeutic goals. Serving as the building block of a phased-array configuration, antennas are typically arranged in a cylindrical or elliptical array called applicator. This short communication proposes an elliptical phased array applicator based on a compact, UWB design from the category of double-ridged horn antennas customized for hyperthermia systems. The performance of the antenna, named open ridged-waveguide, has been experimentally assessed based on the quality metrics of the hyperthermic community. The proposed design achieves an ultra-wideband range of operation from 400 to 800 MHz with an aperture size of 3 by 4 cm. Moreover, thanks to the shielding provided by the metal-lic housing, the design proves good isolation better than (cid:1) 30 dB throughout the band. The power deposition capability of the proposed applicator followed by the thermal analysis is also investigated for a realistic head and neck patient model. The results indicate very good quality metrics achieved in the treatment planning of the patient.


| INTRODUCTION
During the past couple of decades, antennas have found various applications in biomedical engineering, leading to a growing branch of research called bio-electromagnetics (BioEM). BioEM includes, but is not limited to, microwave imaging 1 and hyperthermia (HT). 2 The aim of HT is to increase the efficacy of routine cancer treatment modalities, such as radiotherapy and chemotherapy, by targeting focused EM energy at the tumors under treatment. This selective targeting gives a mild rise to tumor's temperature (of about 5-7 C above the normal body temperature), and as a result, enhances the drug absorption in the tumor. 3 Devices for heating up the target tissue are called applicators in the HT context. Dealing with lossy human tissues, an appropriate antenna for HT applicators needs to operate in the lower part of the UHF band to penetrate deeply enough into the body. This is because the penetration depth falls rapidly at higher frequencies. This also means that a matching medium, commonly referred to as water bolus, between the antenna and the body is needed to not only facilitate the wave transition into the body but also to prevent the heating of the body surface. 4 In general, an ideal antenna design for HT is directive and compact and preferably has a symmetrical radiation pattern with a reflection coefficient better than À10 dB throughout a wide frequency band of operation.
For deep HT treatments, several antennas are typically arranged in a cylindrical or elliptical configuration. 4 A variety of antenna types for this purpose have been already proposed in the literature. [5][6][7] However, they either suffer from a very narrow frequency bandwidth or need an RF matching network/balun to achieve a symmetrical radiation field and stable electrical performance. Besides making the antenna larger and more complex, matching networks absorb a substantial amount of power and thus decrease the efficiency of these designs. Therefore, in terms of reliability and reproducibility, antennas in which an RF matching network can be avoided have a clear advantage when used for a clinical treatment. 5 Furthermore, the optimum phase and amplitude of each element in the array need to be determined so that the focal point is steered on the target and that hot-spot formations in the healthy tissue are suppressed. Several beamforming methods have been proposed in the literature among which we count the generalized eigenvalue problem, 8 time reversal (TR), 9,10 particle swarm optimization (PSO), 11 and focusing via constrained optimization. 12 This study proposes an elliptical phased array applicator for deep HT that uses a new antenna model from the doubleridged horn family. 13 The antenna, henceforth called open ridged-waveguide (ORWG), has a UWB characteristic and requires no matching network or balun. 14 To assess the efficacy of the proposed applicator, the treatment planning of a realistic head and neck (H&N) patient model is investigated. As far as the SAR-based beamforming of the plan is concerned, a hybrid TR-PSO beamforming method is proposed to minimize the cost function by taking advantage of TR fast computation and using it as the starting point. For the thermal analysis, an open-source computational platform, called FEniCS, 15 is used.

| ORWG design
The ORWG antenna, whose design and optimization procedures are given elsewhere, 14 is a water-filled and compressed version of a conventional double-ridged horn antenna. The different parts of the antenna are visualized in Figure 1, and their dimensions are as follows (all in centimeter): a ¼ 3:

| Antenna measurement setup
Two ORWG antennas were fabricated, and a thin rubber diaphragm was attached to their apertures to act as a lid. Separating the antenna from the surface bolus, this lid made it possible for the water to circulate inside the structure through a small inlet and outlet. Experimental verification was performed in a standardized setup with a muscle tissue load as described in HT quality assurance guidelines 16,17 and depicted in Figure 2A.
A hydrogel bolus 18 with the dimensions of 36 Â 16 Â 3 cm was placed on top of a muscle phantom with dimensions 36 Â 25 Â 6 cm. Consisting of 99% water and 1% polymer, a hydrogel bolus is a convenient solution with the potential to replace water-filled plastic bags in HT procedures. The relative permittivities and conductivities of the hydrogel and muscle phantom were assessed by Agilent 85070E dielectric probe kit and shown in Figure 2B,C, respectively.

| Patient model and the phased array applicator
A realistic patient model of a 34-year-old man called Duke 19 with a tumor in the tongue region was investigated. The model, shown in Figure 3A, was segmented by a trained radiation oncologist and the tumor was then inserted into it. The voxel model consists of 45 tissue types whose thermal and dielectric properties are taken from the IT'IS database 20 at 500 MHz. To form a phased array applicator for the treatment planning of this model, 16 ORWG antennas were arranged in an elliptical configuration which has two rings with an inter-ring distance of 5 cm. Shown in Figure 3B is the antenna placement around the patient model simulated in CST Microwave Studio. 21 The water bolus was modeled using the first-order Debye model in CST ( Figure 3C) and its major axis, minor axis, and height are 14, 11, and 14 cm, respectively.

| SAR beamforming
Implementation of PSO in EM problems is straightforward. A thorough discussion about it and about how to choose its parameters is given by Robinson et al. 22 The main steps taken in this process are listed below which also include the integration of the TR technique in the initialization phase of the PSO algorithm.
• Definition of the problem space: 15 particles were set to search the solution space of a unit circle in the complex plane. So each particle was a 16 by one vector of complex numbers in that space. • Initialization of the 1st swarm: TR setting was assigned to the first particle in the set because we wanted to take advantage of having a good initial guess. • Cost function: Hotspot-tumor quotient (HTQ) was selected as the cost function in this study as follows where SAR V 1% ð Þ and SAR T ð Þ are the highest percentile of SAR distribution in the healthy tissue and the mean SAR absorbed by the tumor, respectively. • Termination condition: the optimization was terminated when the maximum number of iterations (M) or the maximum number of stall iterations (S) was reached, whichever came first. In this problem, M and S were set to 30 and 5, respectively.

| Bioheat equation and thermal simulation
The deposited EM energy in terms of power loss density (PLD) eventually turns into heat due to the Joule heating effect. Because of blood circulation in body, biological models involve perfused tissues. Taking into account the blood perfusion rate, the simplest way to model this phenomenon is by using Pennes Bioheat equation 23 as follows: where T C ð Þ is temperature; ρ kg=m 3 ð Þ is mass density; c p J= C=kg ð Þis heat capacity; k W= C=m ð Þis thermal conductivity; and T b , c b , and w b kg=m 3 =s ð Þ are temperature, heat capacity, and perfusion rate of blood, respectively.
While one can go for commercial solvers like CST and Comsol to tackle such a multi-physics problem to some extent, we believe that a modular, open-source computational engine provides more flexibility to model, simulate, and optimize the phenomenon. It can also save a lot of computational time upon efficient implementation. Hence after exporting the EM field from CST, we chose to perform the thermal simulation with a powerful FEM-based computational engine called FEniCS. 15 Tetrahedron mesh generation for the solver was done by Iso2mesh, 24 an open-source MATLAB/Octave-based package.
3 | RESULTS Figure 4A shows the measured S 11 and S 21 of the fabricated ORWG antennas. The setup for this measurement was according to Figure 2A. As can be seen from the black solid line, the return loss of the design remains below À10 dB from 375 to 815 MHz; a bandwidth of 440 MHz. Moreover, as shown with the solid blue line in this plot, the two antennas placed in the close vicinity of each other have a low level of mutual coupling, less than À35 dB throughout the whole band.
The return loss and cross-talk among the adjacent elements of the phased array applicator of Figure 3B are shown in Figure 4B. As can be seen, the design's characteristic (S 11 ) for this realistic scenario is slightly different from the S 11 measured when the antenna was in front of a flat muscle F I G U R E 4 (A) Measured return loss and the mutual coupling between the two elements shown in Figure 2A. (B) Simulated return loss and the cross-talk between the 16 elements in the array configuration of Figure 3B [Color figure can be viewed at wileyonlinelibrary.com] phantom. This was expected due to the variation of water bolus thickness resulting from the natural body curvature. However, the design still enjoys a very good return loss below À10 dB within the desired band. Another interesting observation is that the plot shows a higher level of cross-talk among elements arranged in an elliptical configuration than the level already recorded between those placed on top of a flat muscle phantom. According to Figure 4B, when the first antenna acts as the transmitter, the highest level of mutual coupling happens to its immediate neighboring elements, that is, antennas number 2 and 8 in the top ring. Figure 5A shows the cost function minimization at 500 MHz using the TR-PSO method. The HTQ indicator of TR is 1:3 which was used to initialize the PSO routine for further optimization. It is almost axiomatic that using an educated initial guess can not only help to reduce the optimization time in an otherwise lengthy process (typically more than 100 iterations), but it can also lower the odds of getting trapped in local minima. This hybrid routine resulted in the final value of 1:09 for the HTQ in less than 25 iterations (the five last iterations were the stall iterations as discussed in Section 2.4). Regarding the computational cost, the whole optimization process on an ordinary laptop with an Intel Corei7 ® CPU of 2.8 GHz took around 260 s to converge.
The resultant point SAR distribution of the optimized plan along with the temperature distribution are shown in Figure 5B, D in transverse, coronal, and sagittal planes, respectively. The normalized excitation power per channel was 0:5 W by default in CST. After beamforming, the amplification factor for the combined PLD used in the thermal analysis was set to be 10 to raise the maximum temperature up to 44 C. As can be seen, the tumor's coverage is very good with no significant hotspot in the healthy tissue. Plus, some quantified metrics for the thermal analysis are depicted in Figure 5C. Temperature-volume histogram in this sub-figure indicates that almost the whole volume of tumor has a temperature above 40 C at the steady-state phase while only less than 6% of the healthy tissue volume experiences a temperature higher than the normal body temperature during the treatment. Observe that the T < 37 C is caused by the fixed temperature of the bolus (20 C) applied as Dirichlet boundary condition to the patient skin. To quantify the quality of the treatment outcome based on commonly used metrics in the HT community, SAR treatment quantifier of target coverage 25% (TC 25 ) and thermal indicators of T 90 , T 50 , T 10 were measured as follows: TC 25 ¼ 97%, T 90 ¼ 41:8 C, T 50 ¼ 43 C, and T 10 ¼ 43:9 C. TC 25 indicates the percentage of the tumor volume that is enclosed by the iso-contour of 25% of the maximum deposited SAR in the tumor. Similarly, the thermal indicator of T x determines the temperature in the target tumor exceeding by x percent of all the temperature values. Obviously, the larger the TC 25 and T x , the better the treatment quality.

| CONCLUSIONS
A new antenna model, recently proposed for HT applications, was further investigated in this study. An elliptical phased array applicator hiring 16 ORWG antennas was analyzed in the treatment planning of an H&N patient. Both the characteristics of the antenna and the treatment outcome of the plan are very good and satisfactory. This motivates further development and implementation of the proposed applicator for clinical applications.