Acta Polytechnica https://doi.org/10.14311/AP.2023.63.0383 Acta Polytechnica 63(6):383–389, 2023 © 2023 The Author(s). Licensed under a CC-BY 4.0 licence Published by the Czech Technical University in Prague APPLICATION OF AN ARTIFICIAL INTELLIGENCE SEGMENTATION FOR DEEP HYPERTHERMIA TREATMENT PLANNING IN THE PELVIC REGION Tomas Drizdal∗, Marek Novak Czech Technical University in Prague, Faculty of Biomedical Engineering, Department of Biomedical Technology, nám. Sítná 3105, 272 01 Kladno, Czech Republic ∗ corresponding author: tomas.drizdal@fbmi.cvut.cz Abstract. During a microwave hyperthermia oncology treatment, the target region temperature is elevated to the temperatures of 40–44 °C, which improves the therapeutic effect of a standard radiotherapy and/or chemotherapy treatments. Amplitudes and phases of antenna input signals in the phased array setup surrounding the 3D patient model are optimised with respect to maximise the energy deposition in the target region. In this study, we successfully integrated an automatic artificial intelligence segmentation routine, used for patient-specific 3D model generation, into the hyperthermia treatment planning process. This allows us to apply more realistic patient 3D model for the online hyperthermia guidance including detailed retrospective analyses of the overall treatment quality, possibly leading to a widespread clinical use of the hyperthermia treatment planning. Keywords: Hyperthermia, specific absorption rate, phased array applicator, hyperthermia treatment planning. 1. Introduction Microwave regional hyperthermia is an oncology treat- ment during which a target region is heated to the tem- perature range of 40–44 °C for the duration of 60–90 minutes [1]. It is typically applied in combination with radiotherapy or chemotherapy, utilising constructive interference from multiple external electromagnetic field (EM) sources, converted into a heat within the human body [2, 3]. Individual EM waves are produced by the antennas surrounding the body and forming the phased array regional hyperthermia microwave system [4, 5]. Amplitudes and phases of individual antenna input signals are optimised in order to ob- tain maximum energy deposition within the target region [6, 7]. This is achieved by applying the hyper- thermia treatment planning (HTP) process in which a patient-specific 3D models created from computed tomography (CT) scan in combination with a model of the microwave phased array system for EM field simulations and optimisation is used [8, 9]. HTP is applied for online clinical guidance, the am- plitudes and phases of individual antennas input sig- nals to be adjusted during the treatment, if high tem- perature is measured outside the target region or if the patient complains [10, 11]. It is also used as an in- clusion criterion for the hyperthermia treatment (HT) to test whether it is possible to heat selected target or if it is safe for the patient to undergo the treat- ment [12]. HTP also allows a detailed retrospective analysis of the entire treatment process and design of novel applicator systems [13, 14]. For all HTP appli- cations, the patient 3D model is pivotal for a correct implementation of the selected HTP application. Cur- rently, in clinical practice the 3D patient model usually consists of tissues discriminated by a CT scan, such as muscle, fat, bone, lung and internal air [15]. With the development of automatic segmentation routines based on atlas registration or artificial intelligence (AI) algorithms, generating complex 3D models for the HTP purposes becomes feasible [16–18]. The purpose of this study was to test the applica- tion of an automatic AI segmentation routine for deep HTP purposes in the pelvic region. From publicly available CT scans, we created a 3D patient-specific 3D model using the automatic artificial intelligence based segmentation available in TotalSegmentator ex- tension of 3D Slicer [18–21]. This 3D model was then imported in Sim4Life (version 7.0, Zürich MedTech AG, Switzerland) for electromagnetic field simulations using a phased array system consisting of 12 dipole antennas operating at 100 MHz placed in two rings. We optimised amplitudes and phases of each antenna input signal in order to maximise the specific ab- sorption rate (SAR) within the pancreas assigned as a HT target region. We compared clinically applied SAR quantities for the comparison of detailed patient- specific 3D model consisting of internal organs and for a 3D model in which all internal organs were assigned as muscle, representing the current standard clinical segmentation practice. 2. Materials and methods 2.1. Segmentation of the CT images First, we created a patient-specific 3D model from the series of the CT images available from the Cancer Imaging Archive [19, 20]. The selected CT dataset 383 https://doi.org/10.14311/AP.2023.63.0383 https://creativecommons.org/licenses/by/4.0/ https://www.cvut.cz/en Tomas Drizdal, Marek Novak Acta Polytechnica was for a pancreatic cancer patient. This patient data group has a bad prognosis as the symptoms of- ten appear in the later stage. The tumour is also present in challenging locations for the radiotherapy treatment [22]. The CT images were imported into the 3D Slicer (version 5.2.2) for the AI segmentation for internal organs using the Total Segmentator ex- tension [18]. Two segmentations, using the “Body” and the “Total” options were created and saved in the *.nrrd format. We used the “Total” segmentation option, since it contains individual organs, however without fat tissue. For fat assignment, we used the “Body” segmentation, which ensures that each CT pixel is assigned to a specific tissue. Afterwards, we combined these two segmentations and merged corre- sponding tissues, such as all muscle and bone parts, into a single file in the *.vtk format inside MATLAB (version 9.13, Natick, Massachusetts, USA) using the list shown in Table 3 in the Appendices. Then, we im- ported the CT images to the iSeg (version 3.10.57.98, Zürich MedTech, Switzerland) along with the MAT- LAB file and assigned individual organs. For fat tissue delineation, we used the “Body” AI segmen- tation inside the 3D Slicer and for muscle tissue, we used a threshold operation inside iSeg. Since Virtual Population models used for various in-silico studies, including those in hyperthermia filed, contain carti- lage, we manually segmented the cartilage as well [23]. Figure 1a) shows an example of the CT image and Fig- ure 1b) the corresponding segmentation in iSeg. The whole process took 5 minutes of GPU computational time and 40 minutes of operational power from which the manual cartilage segmentation took 20 minutes. 2.2. Hyperthermia treatment planning setup The selected HT applicator consisted of twelve 100 MHz dipole antennas placed in two rings within the 500 mm long elliptically-shaped shell with a width of 600 mm and a height of 500 mm. An elliptical shape was selected to obtain more uniform distance between the patient surface and the shell compared to a cylindrical setup. This 3D phased array system was positioned around the 3D model in order to place the target region (pancreas) into the centre of the ap- plicator system and provide at least a 40 mm distance from the patient surface to the wire dipole antenna elements (Figure 2). We have selected pancreas as a target region in this study, which represents with respect to its shape and location, a challenging organ for the HT. 2.3. Electromagnetic field simulation A harmonic signal with 15 periods at 100 MHz was used for the excitation of all 12 antenna elements in the phased array setup. A 5 mm global discreti- sation FDTD step with a refinement to 2 mm for the region within the shell and to 1 mm for the wire Figure 1. a) axial CT image, b) segmented CT image in iSeg including internal organs. Figure 2. HTP setup consisting of elliptical phased array (12 wire 100 MHz dipole antennas placed in two rings) and a patient-specific 3D model. Antenna names correspond to the ring and antenna number, e.g. antenna 2_4 is fourth antenna in second antenna ring. dipole antenna was used. Absorbing UPML bound- ary conditions were placed 20 cm from the applicator model. The dielectric properties shown in Table 1 were assigned from the ITIS tissue database avail- able in the Sim4Life package. All metalic parts were assigned as a perfect electric conductor (PEC) mate- rial. It took around 10 minutes to compute one EM field simulation accelerated by RTX 3080 Ti graphical processor unit. 384 vol. 63 no. 6/2023 Application of AI segmentation in hyperthermia treatment planing Name ϵr ρ σ [−] [kg m−3] [S m−1] adrenal gland 64.2 1028 0.64 air 1 1.2 0 blood vessel 76.8 1050 1.23 bone 15.28 1908 0.06 cartilage 55.8 1100 0.47 esophagus 77.9 1040 0.9 fat 12.7 911 0.07 gallbladder 79 1071 1.01 heart 76.8 1050 1.23 kidney 98.1 1067 0.81 large intestine 81.8 1088 0.68 liver 69 1079 0.49 lung 31.6 394 0.31 muscle 66 1090 0.71 pancreas 68.8 1087 0.79 shell 2 1180 0.004 small intestine 96.5 1030 1.66 spleen 90.7 1089 0.8 stomach 77.9 1088 0.9 water 80 1000 0.002 Table 1. Dielectric properties at 100 MHz used in HTP [24, 25]. 2.4. Optimization and evaluation Individual EM field distributions were then imported to MATLAB for amplitude and phase optimisation of the antenna input signals using the particle swarm op- timisation technique. We optimised the target to hot spot quotient (THQ) representing the ratio of aver- aged SAR inside the target region to the average SAR within the hotspot, which were assumed as a volume with 1 % highest SAR outside the target region [26]. Aside from the THQ (–), we evaluated the SAR target coverages (TC) parameters TC25 (%) and TC50 (%), determining the volume of tumour coverage with 25 % or 50 % SAR iso-contour. We have compared two models, • organs – for this model, we assumed a complete segmentation including internal organs, • muscle – for this model, we assigned all internal organs (except lung) as a muscle and cartilage was split into bone structure and muscle using histogram threshold segmentation, which is currently the clin- ically applied segmentation strategy for patient- specific 3D models used in HTP. We also computed the absolute THQ difference |dTHQ| (%). |dTHQ| = ∣∣∣∣ (THQmuscle − THQorgans THQorgans ) ∣∣∣∣ × 100 (1) where |THQmuscle| is the THQ value obtained using muscle patient model and |THQorgans| value obtained 0 25 50 75 100 norm. SAR (%) 0 25 50 75 100 v o lu m e ( % ) TC25 @muscle =91 % TC25 @organs =78 % TC50 @muscle =46 % TC50 @organs =6 % Patient (muscle) Pancreas (muscle) Patient (organs) Pancreas (organs) Figure 3. SAR volume cumulative histograms of pancreas and whole patient model for complete seg- mentation model (organs) and for model with organs assigned as muscle (muscle). from complete segmentation model including internal organs. 3. Results The optimised power and phase coefficients of individ- ual antenna input signals are shown in Table 2. Power coefficients are shown for 1 W of the total power of the phased array system. Antenna Porgans Φorgans Pmuscle Φmuscle [W] [◦] [W] [◦] 1_1 0.077 0 0.072 0 1_2 0.085 -5 0.103 24 1_3 0.087 140 0.069 78 1_4 0.029 -51 0.068 85 1_5 0.015 69 0.098 36 1_6 0.096 69 0.063 96 2_1 0.129 -8 0.075 6 2_2 0.122 49 0.108 93 2_3 0.083 3 0.079 54 2_4 0.100 -63 0.067 -3 2_5 0.098 -18 0.090 7 2_6 0.079 65 0.108 86 Table 2. Optimised powers P [W] (normalised to 1 W of the total power) and phases Φ [◦] of antenna input signals for organs and muscle segmentations. For the complete segmentation model including in- ternal organs, we obtained TC25 = 78 %, TC50 = 6 % and THQ = 0.63, while for the model with internal or- gans assigned as muscle, TC25 = 91 %, TC50 = 46 % and THQ = 0.71, the |dTHQ| = 12.7 %. Figure 3 shows SAR cumulative histograms of the patient and pancreas for the detailed segmentation model includ- ing internal organs and for the patient-specific 3D model in which all internal organs were assigned as a muscle. 385 Tomas Drizdal, Marek Novak Acta Polytechnica Figure 4. Normalized SAR CT overlay for complete segmentation model a) axial c) sagital slices and for model in which internal organs were assigned as muscle b) axial and d) sagital slices with highlighted pancreas contour in green. Figure 4a) and Figure 4c) show axial and sagi- tal slices of the optimised SAR for the complete pa- tient model while Figure 4b) and Figure 4d) show the patient-specific 3D model with muscle dielectric properties assigned to organs. 4. Discussion We have successfully integrated AI automatic seg- mentation into our overall HTP process in the pelvic region. It takes around 45 minutes to generate de- tailed patient-specific 3D models used in automatically generated HTP setups. We received difference of 13 % for TC25 and 40 % for TC50 between the detailed segmentation including internal organs and the model segmented into muscle, fat, lung, air and bone, which is currently standard practice in clinical HTP. These differences, observed in the challenging pancreatic re- gion, highlight the importance of implementing the de- tailed segmentation in this region. However, even with the detailed segmentation, we obtained TC25 = 78 %, which is considered to be sufficient (TC25 ≥ 75 %) for the HT. We estimated that manually creating a com- plete segmentation model would take an experienced individual around ten hours, which is significantly more than the 45 minutes required for the AI-assisted segmentation. We expect that this time will be fur- ther reduced in near future, to 25 minutes, by adding an automatic segmentation of the cartilage. The observed difference in THQ of 12.7 % when comparing two studied models falls into a range of 7.0–22.8 % difference found by VilasBoas-Ribeiro et al. [15] when comparing a clinical model containing muscle, fat, bone, lung and internal air and a very de- tailed model of the Virtual Population models. These models are commonly used for in-silico studies of EM field exposure, nevertheless they represent healthy models of healthy volunteers which led to the develop- ment of Erasmus Virtual Patient Repository (EVPR) created from CT scans of real HT patients [27]. How- ever, the EVPR models consist of a limited amount of tissues, such as fat, muscle, bone, internal air and tumour. The implemented AI segmentation allows to conduct simulation studies including temperature and thermoradiotherapy modelling with realistic pa- tient 3D models [28–30]. These 3D models can be obtained from widely available public repositories, en- abling non-clinical groups to also work in this area of research. The optimised coefficients of antenna input signals differ between HTP results based on the muscle and organs segmentations. This was caused by different dielectric properties of individual organs with respect to muscle tissue. Tissue permittivity and conductiv- ity influence the wavelength and attenuation of the propagating EM wave and thus also the optimised amplitudes and phases of individual antenna input signals. Predicting local areas with high temperatures (hot-spots), which usually appear at tissue interfaces with high dielectric contrast, might be improved with HTP based on a detailed segmentation. However, a study investigating whether a detailed segmentation leads to a higher EM dose within the target region and thus temperatures during the HT would need to be verified in clinical practice, which is outside the scope of this paper. A total of 104 available tissues were segmented us- ing the TotalSegmentator, which can also be used as constrained masks during the online optimisation of amplitudes and phases input signals of individual antennas in the phased array system. Detailed models also allow more accurate simulation guided design of HT devices using, for example, meta-material struc- tures or compact antenna structures [31–33]. Further- more, this technology might also enable the creation of more complex anthropomorphic phantoms for hy- perthermia quality control measurements including dielectric properties of specific phantom materials [34– 36]. In clinical practice, dielectric properties can be obtained individually for every patient using electric property tomography, which uses magnetic resonance imaging (MRI) scanner [37, 38]. MRI, along with Elec- tric Impedance Tomography and Microwave Imaging, can be employed for non-invasive temperature moni- toring during the HT, improving the overall treatment quality [39–42]. 5. Conclusion The implementation of an AI segmentation routine to create patient-specific 3D models enables a more re- 386 vol. 63 no. 6/2023 Application of AI segmentation in hyperthermia treatment planing alistic and tailored HTP approach that can be applied in clinical practice. This could extend the availability of HTP treatment guidance to more clinics, improving the overall quality of HT. Detailed models are also suitable for testing the accuracy of temperature tissue model predictions. Acknowledgements This study was supported by the Student Grant Compe- tition of CTU, grant number SGS23/199/OHK4/3T/17. Further authors would like to thank Zürich MedTech AG (https://www.zmt.swiss), for providing the Sim4Life for Science license enabling this research. References [1] N. R. Datta, D. Marder, S. Datta, et al. Quantification of thermal dose in moderate clinical hyperthermia with radiotherapy: a relook using temperature-time area under the curve (AUC). International Journal of Hyperthermia 38(1):296–307, 2021. https://doi.org/10.1080/02656736.2021.1875060 [2] H. Peeters, E. M. van Zwol, L. Brancato, et al. Systematic review of the registered clinical trials for oncological hyperthermia treatment. International Journal of Hyperthermia 39(1):806–812, 2022. https://doi.org/10.1080/02656736.2022.2076292 [3] J. L. Guevelou, M. E. Chirila, V. Achard, et al. Combined hyperthermia and radiotherapy for prostate cancer: a systematic review. International Journal of Hyperthermia 39(1):547–556, 2022. https://doi.org/10.1080/02656736.2022.2053212 [4] O. Fiser, I. Merunka, J. Vrba. Waveguide applicator system for head and neck hyperthermia treatment. Journal of Electrical Engineering and Technology 11(6):1744–1753, 2016. https://doi.org/10.5370/JEET.2016.11.6.1744 [5] H. P. Kok, E. N. K. Cressman, W. Ceelen, et al. Heating technology for malignant tumors: a review. International Journal of Hyperthermia 37(1):711–741, 2020. https://doi.org/10.1080/02656736.2020.1779357 [6] Z. Rijnen, J. F. Bakker, R. A. M. Canters, et al. Clinical integration of software tool VEDO for adaptive and quantitative application of phased array hyperthermia in the head and neck. International Journal of Hyperthermia 29(3):181–193, 2013. https://doi.org/10.3109/02656736.2013.783934 [7] D. Baskaran, K. Arunachalam. Implementation of thinned array synthesis in hyperthermia treatment planning of 434 MHz phased array breast applicator using genetic algorithm. IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology 7(1):32–38, 2023. https://doi.org/10.1109/JERM.2022.3224294 [8] H. P. Kok, J. van der Zee, F. N. Guirado, et al. Treatment planning facilitates clinical decision making for hyperthermia treatments. International Journal of Hyperthermia 38(1):532–551, 2021. https://doi.org/10.1080/02656736.2021.1903583 [9] H. P. Kok, J. Crezee. Hyperthermia treatment planning: Clinical application and ongoing developments. IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology 5(3):214–222, 2021. https://doi.org/10.1109/JERM.2020.3032838 [10] M. Franckena, R. Canters, F. Termorshuizen, et al. Clinical implementation of hyperthermia treatment planning guided steering: A cross over trial to assess its current contribution to treatment quality. International Journal of Hyperthermia 26(2):145–157, 2010. https://doi.org/10.3109/02656730903453538 [11] H. P. Kok, J. Crezee. Adapt2Heat: treatment planning-assisted locoregional hyperthermia by on-line visualization, optimization and re-optimization of SAR and temperature distributions. International Journal of Hyperthermia 39(1):265–277, 2022. https://doi.org/10.1080/02656736.2022.2032845 [12] M. M. Paulides, G. M. Verduijn, N. Van Holthe. Status quo and directions in deep head and neck hyperthermia. Radiation Oncology 11(1):21, 2016. https://doi.org/10.1186/s13014-016-0588-8 [13] R. Gaffoglio, M. Righero, G. Giordanengo, et al. Fast optimization of temperature focusing in hyperthermia treatment of sub-superficial tumors. IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology 5(3):286–293, 2021. https://doi.org/10.1109/JERM.2020.3043383 [14] T. Drizdal, K. Sumser, G. G. Bellizzi, et al. Simulation guided design of the MRcollar: a MR compatible applicator for deep heating in the head and neck region. International Journal of Hyperthermia 38(1):382–392, 2021. https://doi.org/10.1080/02656736.2021.1892836 [15] I. VilasBoas-Ribeiro, G. C. van Rhoon, T. Drizdal, et al. Impact of number of segmented tissues on SAR prediction accuracy in deep pelvic hyperthermia treatment planning. Cancers 12(9):2646, 2020. https://doi.org/10.3390/cancers12092646 [16] R. F. Verhaart, V. Fortunati, G. M. Verduijn, et al. CT-based patient modeling for head and neck hyperthermia treatment planning: manual versus automatic normal-tissue-segmentation. Radiotherapy and Oncology 111(1):158–163, 2014. https://doi.org/10.1016/j.radonc.2014.01.027 [17] V. Fortunati, R. F. Verhaart, W. J. Niessen, et al. Automatic tissue segmentation of head and neck MR images for hyperthermia treatment planning. Physics in medicine and biology 60(16):6547, 2015. https://doi.org/10.1088/0031-9155/60/16/6547 [18] J. Wasserthal, H.-C. Breit, M. T. Meyer, et al. TotalSegmentator: Robust segmentation of 104 anatomic structures in CT images. Radiology: Artificial Intelligence 5(5):e230024, 2023. https://doi.org/10.1148/ryai.230024 [19] K. Clark, B. Vendt, K. Smith, et al. The cancer imaging archive (TCIA): Maintaining and operating a public information repository. Journal of Digital Imaging 26(6):1045–1057, 2013. https://doi.org/10.1007/s10278-013-9622-7 [20] N. Mayr, W. T. Yuh, S. Bowen, et al. Cervical cancer – tumor heterogeneity: Serial functional and molecular imaging across the radiation therapy course in advanced cervical cancer (CC-tumor-heterogeneity), 2023. [2023-02-14]. https://doi.org/10.7937/ERZ5-QZ59 387 https://www.zmt.swiss https://doi.org/10.1080/02656736.2021.1875060 https://doi.org/10.1080/02656736.2022.2076292 https://doi.org/10.1080/02656736.2022.2053212 https://doi.org/10.5370/JEET.2016.11.6.1744 https://doi.org/10.1080/02656736.2020.1779357 https://doi.org/10.3109/02656736.2013.783934 https://doi.org/10.1109/JERM.2022.3224294 https://doi.org/10.1080/02656736.2021.1903583 https://doi.org/10.1109/JERM.2020.3032838 https://doi.org/10.3109/02656730903453538 https://doi.org/10.1080/02656736.2022.2032845 https://doi.org/10.1186/s13014-016-0588-8 https://doi.org/10.1109/JERM.2020.3043383 https://doi.org/10.1080/02656736.2021.1892836 https://doi.org/10.3390/cancers12092646 https://doi.org/10.1016/j.radonc.2014.01.027 https://doi.org/10.1088/0031-9155/60/16/6547 https://doi.org/10.1148/ryai.230024 https://doi.org/10.1007/s10278-013-9622-7 https://doi.org/10.7937/ERZ5-QZ59 Tomas Drizdal, Marek Novak Acta Polytechnica [21] A. Fedorov, R. Beichel, J. Kalpathy-Cramer, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magnetic Resonance Imaging 30(9):1323–1341, 2012. Quantitative Imaging in Cancer. https://doi.org/https: //doi.org/10.1016/j.mri.2012.05.001 [22] A. van der Horst, H. P. Kok, J. Crezee. Effect of gastrointestinal gas on the temperature distribution in pancreatic cancer hyperthermia treatment planning. International Journal of Hyperthermia 38(1):229–240, 2021. https://doi.org/10.1080/02656736.2021.1882709 [23] A. Christ, W. Kainz, E. G. Hahn, et al. The virtual family-development of surface-based anatomical models of two adults and two children for dosimetric simulations. Physics in Medicine and Biology 55(2):N23, 2010. https://doi.org/10.1088/0031-9155/55/2/N01 [24] S. Gabriel, R. W. Lau, C. Gabriel. The dielectric properties of biological tissues: III. Parametric models for the dielectric spectrum of tissues. Physics in Medicine and Biology 41(11):2271–2293, 1996. https://doi.org/10.1088/0031-9155/41/11/003 [25] P. Hasgall, F. D. Gennaro, C. Baumgartner, et al. ITIS database for thermal and electromagnetic parameters of biological tissues, 2018. [2022-02-23]. https://doi.org/10.13099/VIP21000-04-0 [26] R. A. M. Canters, P. Wust, J. F. Bakker, G. C. V. Rhoon. A literature survey on indicators for characterisation and optimisation of SAR distributions in deep hyperthermia, a plea for standardisation. International Journal of Hyperthermia 25(7):593–608, 2009. https://doi.org/10.3109/02656730903110539 [27] G. G. Bellizzi, K. Sumser, I. VilasBoas-Ribeiro, et al. Standardization of patient modeling in hyperthermia simulation studies: introducing the Erasmus Virtual Patient Repository. International Journal of Hyperthermia 37(1):608–616, 2020. https://doi.org/10.1080/02656736.2020.1772996 [28] I. VilasBoas-Ribeiro, S. A. N. Nouwens, S. Curto, et al. POD-Kalman filtering for improving noninvasive 3D temperature monitoring in MR-guided hyperthermia. Medical Physics 49(8):4955–4970, 2022. https://doi.org/10.1002/mp.15811 [29] H. P. Kok, G. C. van Rhoon, T. D. Herrera, et al. Biological modeling in thermoradiotherapy: present status and ongoing developments toward routine clinical use. International Journal of Hyperthermia 39(1):1126–1140, 2022. https://doi.org/10.1080/02656736.2022.2113826 [30] H. P. Kok, J. Crezee. Validation and practical use of Plan2Heat hyperthermia treatment planning for capacitive heating. International Journal of Hyperthermia 39(1):952–966, 2022. https://doi.org/10.1080/02656736.2022.2093996 [31] D. Vrba, J. Vrba. Novel applicators for local microwave hyperthermia based on zeroth-order mode resonator metamaterial. International Journal of Antennas and Propagation 2014:631398, 2014. https://doi.org/10.1155/2014/631398 [32] D. Vrba, J. Vrba, D. B. Rodrigues, P. Stauffer. Numerical investigation of novel microwave applicators based on zero-order mode resonance for hyperthermia treatment of cancer. Journal of the Franklin Institute 354(18):8734–8746, 2017. https://doi.org/10.1016/j.jfranklin.2016.10.044 [33] D. Baskaran, K. Arunachalam. Design and experimental verification of 434 MHz phased array applicator for hyperthermia treatment of locally advanced breast cancer. IEEE Transactions on Antennas and Propagation 69(3):1706–1715, 2021. https://doi.org/10.1109/TAP.2020.3016462 [34] T. Drizdal, M. M. Paulides, K. Sumser, et al. Application of photogrammetry reconstruction for hyperthermia quality control measurements. Physica Medica 101:87–94, 2022. https://doi.org/10.1016/j.ejmp.2022.08.008 [35] T. Pokorny, D. Vrba, J. Tesarik, et al. Anatomically and dielectrically realistic 2.5D 5-layer reconfigurable head phantom for testing microwave stroke detection and classification. International Journal of Antennas and Propagation 2019:5459391, 2019. https://doi.org/10.1155/2019/5459391 [36] O. Fiser, S. Ley, M. Helbig, et al. Temperature dependent dielectric spectroscopy of muscle tissue phantom. International Journal of Microwave and Wireless Technologies 12(9):885–891, 2020. https://doi.org/10.1017/S1759078720000203 [37] E. Balidemaj, H. P. Kok, G. Schooneveldt, et al. Hyperthermia treatment planning for cervical cancer patients based on electrical conductivity tissue properties acquired in vivo with EPT at 3 T MRI. International Journal of Hyperthermia 32(5):558–568, 2016. https://doi.org/10.3109/02656736.2015.1129440 [38] S. Gavazzi, C. A. T. van den Berg, A. Sbrizzi, et al. Accuracy and precision of electrical permittivity mapping at 3T: the impact of three , mapping techniques. Magnetic Resonance in Medicine 81(6):3628– 3642, 2019. https://doi.org/10.1002/mrm.27675 [39] I. VilasBoas-Ribeiro, S. Curto, G. C. van Rhoon, et al. MR thermometry accuracy and prospective imaging-based patient selection in MR-guided hyperthermia treatment for locally advanced cervical cancer. Cancers 13(14):3503, 2021. https://doi.org/10.3390/cancers13143503 [40] T. V. Feddersen, D. H. J. Poot, M. M. Paulides, et al. Multi-echo gradient echo pulse sequences: which is best for PRFS MR thermometry guided hyperthermia? International Journal of Hyperthermia 40(1):2184399, 2023. https://doi.org/10.1080/02656736.2023.2184399 [41] R. Poni, E. Neufeld, M. Capstick, et al. Feasibility of temperature control by electrical impedance tomography in hyperthermia. Cancers 13(13):3297, 2021. https://doi.org/10.3390/cancers13133297 [42] O. Fiser, M. Helbig, J. Sachs, et al. Microwave non-invasive temperature monitoring using UWB radar for cancer treatment by hyperthermia. PIER 162:1–14, 2018. https://doi.org/10.2528/pier17111609 388 https://doi.org/https://doi.org/10.1016/j.mri.2012.05.001 https://doi.org/https://doi.org/10.1016/j.mri.2012.05.001 https://doi.org/10.1080/02656736.2021.1882709 https://doi.org/10.1088/0031-9155/55/2/N01 https://doi.org/10.1088/0031-9155/41/11/003 https://doi.org/10.13099/VIP21000-04-0 https://doi.org/10.3109/02656730903110539 https://doi.org/10.1080/02656736.2020.1772996 https://doi.org/10.1002/mp.15811 https://doi.org/10.1080/02656736.2022.2113826 https://doi.org/10.1080/02656736.2022.2093996 https://doi.org/10.1155/2014/631398 https://doi.org/10.1016/j.jfranklin.2016.10.044 https://doi.org/10.1109/TAP.2020.3016462 https://doi.org/10.1016/j.ejmp.2022.08.008 https://doi.org/10.1155/2019/5459391 https://doi.org/10.1017/S1759078720000203 https://doi.org/10.3109/02656736.2015.1129440 https://doi.org/10.1002/mrm.27675 https://doi.org/10.3390/cancers13143503 https://doi.org/10.1080/02656736.2023.2184399 https://doi.org/10.3390/cancers13133297 https://doi.org/10.2528/pier17111609 vol. 63 no. 6/2023 Application of AI segmentation in hyperthermia treatment planing A. Appendices Tissue name in HTP Tissue name from TotalSegmentator adrenal gland right_adrenal_gland, left_adrenal_gland air trachea blood vessel aorta, inferior_vena_cava, portal_vein, left_common_iliac_artery, right_common_iliac_artery, left_common_iliac_vein, right_common_iliac_vein bone L5_vertebra, L4_vertebra, L3_vertebra, L2_vertebra, L1_vertebra, T12_vertebra, T11_vertebra, T10_vertebra, T9_vertebra, T8_vertebra, T7_vertebra, T6_vertebra, T5_vertebra, T4_vertebra, T3_vertebra, T2_vertebra, T1_vertebra, C7_vertebra, C6_vertebra, C5_vertebra, C4_vertebra, C3_vertebra, C2_vertebra, C1_vertebra, left_rib_1, left_rib_2, left_rib_3, left_rib_4, left_rib_5, left_rib_6, left_rib_7, left_rib_8, left_rib_9, left_rib_10, left_rib_11, left_rib_12, right_rib_1, right_rib_2, right_rib_3, right_rib_4, right_rib_5, right_rib_6, right_rib_7, right_rib_8, right_rib_9, right_rib_10, right_rib_11, right_rib_12, left_humerus, right_humerus, left_scapula, right_scapula, left_clavicle, right_clavicle, left_femur, right_femur, left_hip, right_hip, Sacrum esophagus esophagus fat body gallbladder gallbladder, urinary_bladder heart heart, left_atrium, left_ventricle_of_heart, right_atrium, right_ventricle_of_heart, pulmonary_artery kidney right_kidney, left_kidney large intestine colon liver liver lung superior_lobe_of_left_lung, inferior_lobe_of_left_lung, su- perior_lobe_of_right_lung, middle_lobe_of_right_lung, infe- rior_lobe_of_right_lung muscle left_gluteus_maximus, right_gluteus_maximus, left_gluteus_medius, right_gluteus_medius, left_gluteus_medius, right_gluteus_medius, left_erector_spinae_muscle, right_erector_spinae_muscle, left_iliopsoas_muscle, right_iliopsoas_muscle pancreas pancreas small intestine small_bowel, duodenum spleen spleen stomach stomach Table 3. Merging tissues from TotalSegmentator and its corresponding tissue names in HTP models. 389 Acta Polytechnica 63(6):383–389, 2023 1 Introduction 2 Materials and methods 2.1 Segmentation of the CT images 2.2 Hyperthermia treatment planning setup 2.3 Electromagnetic field simulation 2.4 Optimization and evaluation 3 Results 4 Discussion 5 Conclusion Acknowledgements References A Appendices