Design of Multiresonance Flexible Antenna Array Applicator for Breast Cancer Hyperthermia Treatment

Hyperthermia therapy has recently become one of the primary treatment principles due to its prominence and success in curing deep-rooted malignancies, especially for breast tumors. Therefore, a proposed multi-resonance applicator (MRA) was designed and investigated in this study and compared with the performance of a single resonance applicator (SRA). The MRA applicator includes a set of 35 antenna elements operating simultaneously in different frequency bands. This is accomplished by applying a modified version of the gravitational search algorithm hybridized with particle swarm optimization to design the antenna element and focusing the heat on the tumor. Machine learning was used to build antenna elements operating in different frequency bands. Several scenarios considering realistic thermal and dielectric breast properties with a single tumor of different volumes and positions are addressed to evaluate the applicator’s performance and capability. The results showed the capability of the proposed MRA applicator to use a microwave power of 60 watts to elevate the temperature of the tumor to over 42.5° C with better performance indicators than SRA. The suggested methodology of using MRA can be considered to treat successfully other malignancies in different organs, such as brain, liver and larynx tumors.

methods, the need for a microwave transmission to cause 28 The associate editor coordinating the review of this manuscript and approving it for publication was Seifedine Kadry . centralized hyperthermia in tumors has the great attraction 29 [5], [6], [7]. 30 Clinical microwave hyperthermia (MWH), which involves 31 mild heating in the range of 39-44 o C for 30-90 minutes, 32 has been shown to be a viable adjuvant to radiation and/or 33 chemotherapy for a variety of malignancies [8], [9], [10], 34 [11], [12]. On the other hand, the thermal model has some 35 limits which can be divided into passive and active systems. 36 The process through which the human body and the physical 37 world around it exchange heat and moisture is known as a 38 passive system. The body generates heat through metabolism, 39 which is continually lost to every region of the body via 40 the circulatory system. The body's characteristics play a key 41 role in the estimation of this portion of heat. The factors 42 taken into account include blood flow, bone, fat, and mus-43 cle mass. As a result of internal heat conduction, it is lost 44 noninvasive monitoring of deep tissue temperature and more 101 antennas. In [14], hyperthermia applicator with 24 directed 102 microstrip spiral antennas constituting a hemispheric array 103 at 0.434 GHz is designed for the treatment of breast cancer, 104 whereas, the antenna phases are adjusted to concentrate all of 105 the fields in the desired location. Asili et al. [15], have intro-106 duced a flexible mild MWH antenna applicator for chemo-107 thermotherapy of the breast that operates at 1.6 GHz using 108 3 × 3 network antennas warp by 45 o . Using a 24-element 109 antenna array at 4200 MHz and differential beam-steering 110 sub-arrays, an epitome of MWH of patient-private breast 111 paradigms is demonstrated in [16]. The focusing procedure 112 is carried out by applying the particle swarm optimization 113 (PSO) to compute the best excitation phases of the planned 114 three-dimensional antenna array, considering advanced phase 115 shifts. In [19], the authors have designed and evaluated a new 116 applicator with 48 antenna elements resonating at a single 117 frequency of 750 MHz. The applicator had the capability to 118 heat multiple brain tumors simultaneously based on adaptive 119 beamforming technique. 120 Recently, Machine Learning (ML) models have been incor-121 porating into an optimization method to determine the best 122 antenna settings and performance [20], [21]. The artificial 123 neural network (ANN), which is the foundation of the ML, 124 allowing to learn and interpret data. The output may be 125 correctly calculated once the ANN model has been correctly 126 trained using the input data. 127 In this research, a numerical investigation of a non-invasive 128 microwave beamforming technique for producing localized 129 hyperthermia in the breast is presented. To begin, the ML 130 technology is used to propose a really well-designed antenna 131 element with high gain, efficiency, and low reflection coef-132 ficients at the resonance frequency of 2.5 GHz. Then a flex-133 ible antenna array consisting of 35 elements is designed to 134 construct the single resonance applicator. The applicator is 135 implemented considering the shape of the breast. On the 136 other hand, the multi-resonance applicator (MRA) has been 137 designed considering different resonance antennas at 1.5, 2.5, 138 3.5 and 4.5 GHz using ML techniques. The performance 139 of the two applicators was compared in terms of accuracy 140 and speed of tumor heating considering a real setting of 141 female's breast phantom; this environment contains tumors 142 with different volumes and positions. As detailed described 143 before in [19], the hybrid modified version of gravitational 144 search algorithm and particle swarm optimization (MGSA-145 PSO) algorithm [22], [23] is used to optimize the antenna 146 elements feeding weights to utilize the power at tumor posi-147 tions while preserving normal temperatures in normal tissue. 148 To that aim, the compiler of computer simulation technology-149 microwave studio (CST-MWS) is coupled to Matlab program 150 to run thermal and electromagnetic simulations for testing 151 applicator performance.

152
The following is how the paper is structured: Section II, the 153 antenna construction and array applicator setup are discussed. 154 The research results are presented and analyzed in Section III. 155 Lastly, Section IV wraps off with the conclusions. profile in the breast can be determined using the CST-thermal 169 analysis and a well-known Pennes bio-heat equation [24].
Each antenna array element's phase ϕ n , on the other hand, 211 is chosen from an area of choice of [180 • , 180 • ]. The opti-212 mization procedure is stopped when the tumor centre and 213 the present optimum location are separated by just under 214 1 mm or when the number of iterations reaches its limit, 215 whichever comes first. Thermal simulations in CST are then 216 used to determine the temperature distribution T . The optimal 217 phases ( ϕ * ) are employed in this thermal study, whereas every 218 antenna element has an amplitude α n that is optimized to 219 increase the tumor temp to 42.5 o C. As a result, the objective 220 function in this scenario is: where, V 37 o C tumor is the total volume of tumor i at 37 o C of normal 223 body temperature and V 42.5 o C tumor ( α) is the tumor i volume at 224 which the temp of 42.5 o C is achieved using the amplitudes 225 that have been optimized α * . The exciting amplitudes α, 226 on the other hand, are placed anywhere between 0.5 and 227 3 in the decision space. Finally, normal tissue hot spots are 228 evaluated, and if any are found, the antenna array feedings 229 are re-optimized using the previously acquired global best 230 solution of phases and amplitudes. The CST-MWS package's 231 compilers and Matlab software are coupled to address this 232 difficult task. Matlab is used to control the beamforming 233 process, while the thermal simulations and electromagnetic 234 computation are performed in CST. As a consequence, the 235 link is put together using several command lines to allow 236 Matlab to manage CST-MWS from the outside. The essential 237 steps of the proposed methodology followed in this paper are 238 depicted in Fig. 1, which is a simplified flowchart.

240
In this section, the design of antenna element is introduced. 241 In addition, a brief description of using ML technique for 242 modeling and training will be presented. Furthermore, the 243 constitutional design of the applicator with flexible array 244 antenna would be described. Moreover, the EM modeling 245 considering breast phantom and tumors will be illustrated. 246 The problem formulation is also included in the description 247 and the scenarios those will be considered. To begin, the proposed antenna's design configuration using 250 ML technique is discussed. Fig. 2 shows the dimensions 251 of the antenna to be trained using ML technique for best 252 performance of reflection coefficient, radiation efficiency, 253 and high gain. The antenna element is 2 cm × 2 cm in size 254 at 2.5 GHz on a dielectric constant ( r ) of 3.5 and a loss 255 tangent (δ) of 0.002, a Kapton polyimide-based substrate with 256 a height of 0.11 mm is considered for its flexibility, resilience, 257 and thermal durability. As depicted, the antenna consists of 258    For MRA implementation, antennas resonating at different 277 frequencies are required; therefore, the ANN architecture 278 with five layers was applied including the input layer, three 279 hidden layers, and the output layer [21]. In order to generate a 280 database for modeling the ML, simulations of 160 times with 281 different dimensions are performed using CST-MWS solver. 282 Fig. 3 shows the topology representation of the variables of 283 the modeled antennas. The antenna parameters are divided 284 into four groups based on their size of 25 × 25 mm 2 , 20 × 285 20 mm 2 and 15 × 15 mm 2 , 10 × 10 mm 2 . Each group has 286 40 one element that comprise parameter combination of W 1 , 287 W 2 , W 3 , W 4 , W 5 , L 1 . e.g. for the first group of 25 × 25 mm 2 , 288 there is 40 one element including the parameter combination. 289 The 160 one-element data is separated into two datasets: 290 dataset #140 and dataset #20. The CST will determine the 291 simulated resonant frequency f rsim of each element with a 292 specific antenna setting.

293
In this study, the five-layer a machine learning architecture 294 with three hidden layers is modeled, as illustrated in Fig. 4. 295 VOLUME 10, 2022  for the antennas protrude from the applicator, where they are 331 attached to cables that are connected to the switching matrix, 332 phase shifters to regulate excitation phases, and attenuators 333 to control excitation amplitudes. A free space separation 334 distance of 1 mm is considered in this study between the 335 applicator and the female patient breast.

337
The breast phantom used in this study is derived from 0.5 mm 338 spatial resolution magnetic resonance imaging (MRI) data 339 [26]. A well-known Virtual Population (ViP3.1) model called 340 Ella is considered as one from the family available through 341 the IT'IS Foundation [27]. Ella is a 26-years old female, 342 1.63 m tall and weighs 57.3 kg. Ella's breast phantom model 343 mimics the physical shape and anatomy of the fatty human 344 breast. It has 0.5 mm × 0.5 mm × 0.5 mm resolution with 345 320 × 360 × 258 voxels. The width of the breast base 346 is 14.7 cm. The areolar diameter is 3.75 cm in both hori-347 zontal and vertical planes. The nipple of the Ella's breast 348 is situated 8.79 cm from the mid-sternal line and the dis-349 tance from the nipple to the inframammary fold is set to 350 6.34 cm. The nipple diameter is 1.3 cm with a projection 351 of 0.85 cm. The breast phantom model includes the main 352 tissue types: skin, muscle, blood and fat. The tissue dielectric 353  and thermal properties are adopted from [27] as reported 354 in Table 1

400
Firstly, the results of antenna design and ML will be pre-401 sented in this section. Then, the power and thermal results of 402 the designed applicators in addition to the quality indicators 403 will be discussed to show the capability of the proposed 404 MRA applicator for heating breast tumor with different vol-405 umes/positions with high efficiency. 406 Fig. 7 shows the effect of antenna parameters on the radi-407 ation characteristics performance. Therefore, a comparison 408 between an optimized antenna element resonating at 2.5 GHz 409 and random antenna dimensions regarding the antenna char-410 acteristics (reflection coefficient, radiation efficiency, gain, 411 and axial ratio) is introduced in the figure. Table 2 presents 412 the optimized and initial antenna dimensions resonating at 413 2.5 GHz. It is clear that the length of the four curved arms 414 (L 1 ) is responsible for improving reflection coefficient S 11 as 415 shown in Fig. 7a. Also, the length of antenna core on the feed 416 point (W 4 ) has a significant impact on the antenna radiation 417 efficiency as illustrated in Fig. 7b. The effect of the cross-418 oval rings (W 2 ) on improving the realized gain is depicted 419 on Fig. 7c. Finally, Fig. 7d presents the impact of the semi-420 oval heads at the end of each arm (L 2 ) on enhancing the axial 421 ratio. So, the improvement in the antenna gain will reflect the 422 enhancement in the antenna field distribution. As shown in 423 Fig. 8, the field intensity and distribution of the optimized 424 antenna in Fig. 4a is enhanced compared to the random 425 antenna design antenna in Fig. 4b     Regarding the antenna element design, Fig. 9 depicts the 427 simulated resonant frequency variation versus antenna num-428 ber. It is clear that as the antenna exterior dimension grows 429 larger; the resonant frequency lowers, resulting in a signifi-430 cant degree of nonlinearity between the resonant frequency 431 and the antenna parameters. As a result, calculating the 432 resonance frequency of a single element is a difficult and 433 nonlinear task. So, the ML model is trained, validated and 434 tested on dataset #140 as depicting in Fig. 10a and dataset 435 #20 as depicting in Fig. 10b for the purpose of verifying and 436 comparing the ML with the earlier established.

437
When the ML is applied to the antenna parameters, it is 438 evident that the obtained antennas can achieve good perfor-439 mance in terms of reflection coefficient, realized gain, and 440 radiation efficiency, at different resonance frequencies of 1.5, 441 2.5, 3.5 and 4.5 GHz as shown in Fig. 11. Whereas, the 442      The parameters of the one element antenna and some 458 testing data contain the computed and measured resonant 459 frequency values are given at the Table 3. It should be noted 460 that, the antenna parameters (W 1 , W 2 , W 3 , W 4 , W 5 , L 1 ) 461 are the main parameters those greatly effecting the antenna 462 performance, while other parameters are kept constant as 463 depicted in Table 2.

464
Now, the antenna curvature effect on the reflection coeffi-465 cient either in free space or in the presence of breast phan-466 tom is presented in Fig. 13. It is clear that, the reflection 467 coefficient is increased from -25.57 dB to -23.6 dB due to 468   number of antenna elements at higher frequencies in the pre-510 sented MRA applicator. Thus, the optimum choice depends 511 on the required focusing resolution and penetration depth, 512 which are patient-specific. If the frequency is too low, the 513 heated area is larger than the desired area associated with poor 514 resolution. If the chosen frequency is too high, then higher 515 EM power is absorbed near the skin due to the reduced pen-516 etration of higher frequencies. Therefore, the optimal design 517 of the applicator should include combination of antenna array 518 operating in different frequency bands.

519
As depicted in Figs. 18d and 19d, the capability of the 520 designed MRA to heat the tumor effectively better than SRA. 521 The obtained results of Q-power distribution and thermal 522 profiles is compared to the previously introduced results. It is 523 clear that, the amount of power dissipated in the tumor using 524 MRA is significantly higher than that obtained using SRA. 525 Furthermore, low power is almost dissipated in the rest of the 526 healthy tissues surrounding the tumor in the case of MRA.

527
Consequently, the temperature distribution is more con-528 centrated into the tumor tissues in case of using MRA, and 529 it is almost non-existent in the healthy tissues around the 530  tumor This confirms the validity of the multi-frequency effect 554 on the accuracy and performance.

555
Regarding the quality indicators of aPA, RMTi and TC 50% , 556 they will be computed to measure the proposed MRA appli-557 cator heating performance. Fig 23 illustrates the quality 558 indicators of the proposed MRA applicator for different sce-559 narios and compared to SRA at different frequencies. When 560 calculating the aPA coefficient, it was noticed that its value 561 decreases with increasing frequency when using the SRA 562 applicator, and also decreases with the increase in the volume 563 of the tumor. But in the case of using the MRA applicator, 564 it was noticed that the aPA value was large compared to 565 the SRA with an average value of 2.5, 2.72, and 3.91 in 566 each the 1cm 3 , 2cm 3 , and 3cm 3 tumors, respectively. While 567 calculating the RMTi coefficient, it was noticed that its value 568 increases with increasing frequency when using the SRA 569 applicator, and also decrease with the increase in the volume 570 of the tumor. But in the case of using the MRA applicator, 571 it was noticed that the RMTi value was large compared to 572 the SRA with an average value of 3.62, 1.85, and 1.93 in 573 each the 1cm 3 , 2cm 3 , and 3cm 3 tumors, respectively. Finally, 574 when calculating the TC 50% coefficient, it was noticed that 575 its value increases with increasing frequency when using 576 to 42.5 o after 7.7 min, increased to 13.7 min for larger tumor 622 of 3 cm 3 .

623
Now, the effect of the blood perfusion rate on the tumor 624 temperature is presented. Fig. 25 illustrates the relation 625 between the blood perfusion rate and the temperature, 626 whereas, the blood flow rate in the model was changed at 627 a rate from 0.5 to 0.9 with step 0.1. It was found that by 628 increasing the rate of blood flow in the tumor and blood ves-629 sels, this led to a prolonged heating period, meaning that the 630 tumor needed more time to reach the required temperature. 631 As depicted in the figure, it required only 10 min to heat the 632 tumor to 42.5 o at a perfusion rate of 0.5 which increased to 633 25 min at a perfusion rate of 0.9.

634
In addition, the rate of change in the tumor volume due to 635 the rise in temperature has been studied. As an example, the 636 tumor volume of 2 cm 3 has been considered. Therefore, the 637 thermal transient solver is executed for localized heating with 638 the exposure time to monitor the tumor volume extension 639 along x, y, and z axes across the center of the tumor during 640 heating process. As depicted in Fig. 26, the tumor volume 641 extension is clearly increasing with time as the temperature 642 rises. After 55 minutes whereas the temperature is increased 643 to 45 o , the tumor volume is extended up to 15%.

644
Finally, the effect of the proposed MRA applicator on the 645 temperature of the breast skin is introduced. As presented in 646 [29], the thermal boundary conditions at the skin surface are 647 found to be 40.6 o at 2.5 GHz. As shown in Fig. 27a, the use 648 malignancies in different organs, such as brain, liver, and 661 larynx tumors can be treated successfully with the suggested 662 methodology of using multi-resonance applicator (MRA).

663
Regarding the feeding complexity of the applicator, it is 664 worth noting the great progress in the field of cellular mobile 665 whereas, with the advent of 5G communications, the increas-666 ing demand for the effective use of beamforming technique 667 has been more accelerated than ever. Recently, various types 668 of antenna arrays have been designed and implemented at 669 different frequency bands. For example, IBM and Ericsson 670 announced the world's first reported Si mmWave phased 671 array antenna module operating at 28 GHz [30]. The module 672 consists of four monolithic microwave integrated circuits 673    678 results were compared with those obtained by a single reso-682 nance applicator. A realistic human breast model and irregu-683 lar tumor shape was considered to create a practical environ-684 ment. Also, this paper discussed the applicator's capability 685 to treat tumors of different volumes and placements. The 686 obtained simulation results revealed that the MRA outper-687 form the performance than the SRA in addition to decreasing 688 the required time to reach up to 42. Communications and Electronics Engineering. He is also the Vice Dean 817 of Community Service and Environmental Development. He has published 818 over 80 refereed journals and conference papers in addition to one book on 819 reconfigurable microwave filter. His current research interests include the 820 areas of 5G mm-wave and optical nano-antennas design for modern wire-821 less applications using metaheuristic optimization techniques, microwave 822 hyperthermia, microwave filters design, and radar cross section reduction 823 techniques. with the Electrical Engineering Department, Fac-831 ulty of Technology and Education, Sohag Univer-832 sity, Sohag, Egypt. He has authored more than 833 35 papers on microwave-based smart antenna, 834 conformal array devices, and mmWave antennas. He has served as an edi-835 tor/a reviewer of many international journals. His current research interests 836 include the areas of microwave applications in biomedical, especially in 837 breast and brain cancer, hyperthermia, and using millimeter wave for cancer 838 detection.