Multiphysical Characterization of a Tissue-Mimicking Phantom: Composition, Thermal Behavior, and Broadband Electromagnetic Properties from Visible to Terahertz and Microwave Frequencies

Abstract

A water-rich muscle-equivalent tissue-mimicking phantom within a polymeric matrix was experimentally evaluated through a multimodal characterization methodology to determine whether it reproduces the coupled dielectric–thermal behavior of hydrated biological tissue under exposure to electromagnetic waves. The material was analyzed using thermogravimetric analysis, microwave dielectric spectroscopy from 1.5 to 4.0 GHz, VIS–NIR spectroscopy between 350 and 1200 nm, and terahertz time-domain reflection. The thermogravimetric results confirmed dominant water content, with primary mass loss below 200 °C, establishing hydration as the governing factor of its thermal response. Next, the microwave dielectric measurements show that the phantom exhibits a relative permittivity of 37.4 and an electrical conductivity of 2.4 S/m. On the other hand, the VIS–NIR spectra show smooth broadband absorption with limited spatial variability, and principal component analysis reveals macroscopic optical homogeneity without structural spectral distortion. In the THz regime, strong broadband attenuation characteristic of water-rich matrices is observed, and reflection-mode measurements enable robust assessment of temporal stability through time- and frequency-domain signatures. Finally, a microwave thermal validation demonstrates stable behavior under low-power excitation, whereas under hyperthermia-level irradiation, a significant thermal drift of −3.985 °C/h was reached under non-adiabatic conditions, identifying hydration-mediated moisture redistribution as the principal limitation during prolonged high-power exposure. Collectively, these results demonstrate cross-regime dielectric–thermal consistency while explicitly defining the hydration-driven constraints governing long-term stability, providing a validated reference material for broadband electromagnetic and thermal biomedical experimentation.

1. Introduction

In biomedical engineering, tissue-mimicking phantoms are not merely experimental surrogates but engineered functional materials whose performance is governed by well-defined structure–property relationships. By reproducing the electromagnetic and thermal behavior of real biological tissues, these materials enable safe, controlled, and repeatable studies for dosimetric assessment, such as the evaluation of the specific absorption rate (SAR) in microwave hyperthermia, as well as the establishment of exposure limits for near-field and therapeutic devices [1,2]. From a materials-science perspective, however, the fidelity of such studies depends critically on rigorous characterization of phantom composition and multiphysical properties, ensuring that dielectric permittivity, electrical conductivity, thermal conductivity, and specific heat capacity remain consistent with those of the target tissue under realistic operating conditions, as reported in optical phantom studies [2,3], microwave, and thermal hyperthermia analysis [4,5,6]. Only through such validation can phantoms be regarded as reliable material platforms capable of predicting in vivo behavior.

Over the past two decades, phantom design has advanced in deliberate and meaningful steps to meet the growing demands of biomedical research. Early efforts focused on reconstructing internal temperature changes using capacitance-based measurements in brain-mimicking media [7]. Building on that foundation, just a year later, Ito and colleagues developed a solid, muscle-equivalent phantom for microwave dosimetry, laying out the first recipes to match tissue permittivity and conductivity [8]. Subsequent developments expanded phantom functionality, including reusable heat-sensitive materials for focused ultrasound hyperthermia that enabled spatially resolved thermal mapping and improved understanding of treatment dynamics [5]. Gelatin-based models with adaptable Bloom resistance were a significant advancement in phantom fabrication because they allowed for the adjustment of gel stiffness and network structure to mimic multiple soft tissues while systematically characterizing the resulting acoustic, magnetic resonance, and thermal responses within a single material platform [9]. Collectively, these studies illustrate a gradual transition from empirical phantom recipes toward more rational material design, where composition and microstructure are tailored to achieve reproducible multiphysical responses.

By the early 2020s, phantom research had focused on the evaluation of a diverse tapestry of specialized materials and measurement techniques. Researchers started with liquid optical phantoms designed to cover the visible-near-infrared spectrum (500–1000 nm). They used an enhanced single-integrating sphere setup, meticulously prepared dozens of absorption, scattering, and mixed component samples to benchmark the absorption coefficient (${\mu }_a$) and reduced scattering coefficient (${\mu }_s$), demonstrating the stability and repeatability of the system [10]. At the same time, both Zerdine (a commercial material) and agar phantoms were illuminated at 635 nm to extract their core optical parameters, such as transmittance, absorbance, reflectance, refractive index, and attenuation, via direct power comparisons in a homogeneous setup and with a single integrating sphere system. The integrating sphere observations most nearly matched expected tissue values, confirming its role as the gold standard for optical phantom characterization [11]. Complementary efforts extended dielectric characterization to higher microwave and submillimeter-wave frequencies; for example, water–oil–agar blends were employed to investigate D-band (150 GHz) permittivity in breast tissue mimics [12], while heterogeneous solid skin-equivalent phantoms were validated for diagnostic microwave applications across 0.5–26.5 GHz using coaxial probe measurements and vector network analyzers [8,13]. Agar–SiO₂ tumor phantoms tailored for MRI-guided focused ultrasound demonstrated realistic lesion formation and thermal dose deposition by combining intermittent thermometry with multimodal imaging contrast [14].

These characterization advances established the basis for optimizing hyperthermia therapies [15]. For example, Nizam-Uddin et al. paired a flexible monopole applicator with a patient-mimicking breast phantom and fiber-optic thermometry to demonstrate precise, localized heating of multiple lesions [16]. In parallel, Rahpeima and Lin used anatomically realistic breast phantoms loaded with magnetic nanofluids to simulate nanoparticle diffusion, heat generation via Pennes’ bioheat equation, and resulting tissue necrosis [17]. Meanwhile, Fakhradini et al. optimized microwave antenna power in liver-cancer phantoms loaded with superparamagnetic nanoparticles, achieving maximal tumor heating with minimal collateral damage [18]. On the monitoring front, Gaffoglio et al. introduced a real-time 3D temperature reconstruction method combining sparse fiber-optic readings with a simulation library, delivering full-field maps with sub-degree accuracy [19], while Dietrich et al. validated 3D MRI thermometry in agar phantoms during microwave heating, reporting RMS errors below 0.7 °C without RF interference [20]. Complementary reviews and parametric studies have further clarified the influence of applicator design, frequency selection, and antenna configuration on penetration depth, heating uniformity, and hotspot control in hyperthermia treatments [21,22], while microwave radiometry has been demonstrated as a non-invasive technique for detecting sub-degree internal temperature gradients in anatomically realistic breast phantoms [23].

Beyond the microwave regime, optical and terahertz (THz) characterization has emerged as a complementary approach to probe phantom composition, microstructure, and hydration-dependent behavior from a materials-science perspective. In the optical domain, solid tissue phantoms with adjustable subdiffusive scattering parameters have demonstrated that controlled incorporation of scattering inclusions enables not only precise tuning of optical response but also long-term stability and mechanical robustness [24]. These studies highlight that optical properties in tissue-mimicking materials are intrinsically linked to microstructural organization and constituent dispersion, rather than being merely spectral fitting parameters [24,25]. At higher frequencies, THz time-domain spectroscopy has gained increasing attention as a sensitive tool for interrogating soft and hydrated materials, owing to its strong interaction with water content, weak intermolecular forces, and mesoscale structural features. Recent advances in THz-based material characterization have emphasized the importance of reflection-mode approaches, which overcome the limitations of conventional transmission geometries when dealing with optically thick, opaque, or highly absorbing samples [26]. Such developments significantly extend the applicability of THz spectroscopy to realistic tissue-mimicking materials, which can potentially enable the extraction of frequency-dependent dielectric properties.

Importantly, the application of THz techniques to hydrated phantoms has been systematically investigated in recent years. Jayasankar et al. demonstrated, using THz time-domain spectroscopy, that both the effective refractive index and absorption coefficient of hydrogel-based skin phantoms strongly depend on hydration level and water-binding mechanisms [27]. Similarly, Saxena et al. reported that gelatin-based phantoms exhibit pronounced sensitivity of time-domain signatures and frequency-dependent attenuation to moisture redistribution and structural organization of bound water [28]. These studies show that the THz response of hydrated phantoms cannot be accurately described using simplified effective-medium models. In particular, assuming invariant bound-water contributions leads to systematic deviations, as both the fraction and organization of bound water evolve with hydration state.

Despite the substantial progress outlined above, most existing studies still address dielectric, optical, terahertz, or thermal characterization—or hyperthermia validation—in isolation. Consequently, the relationship between phantom composition, broadband electromagnetic dispersion across multiple frequency decades, and thermal stability under sustained electromagnetic excitation remains insufficiently explored, particularly for muscle-equivalent materials intended for microwave hyperthermia and SAR validation.

In this study, we address this gap by fabricating a muscle-equivalent phantom following the formulation proposed by Ito et al. [8] to reproduce real-tissue permittivity and conductivity. A multimodal characterization strategy was implemented to evaluate its electromagnetic and thermal behavior across multiple spectral regimes. The dielectric response was measured from 1.5 to 4 GHz using a coplanar waveguide method, optical absorption was analyzed over the 350–1200 nm range, and the THz response was investigated in reflection mode using THz-TDS to examine time- and frequency-domain signatures and their temporal stability. In parallel, thermal conductivity and volumetric heat capacity were experimentally determined. Finally, the phantom was exposed to 2.45 GHz microwave irradiation under hyperthermia conditions, and its temperature evolution was monitored using infrared thermography. By combining broadband dielectric characterization with controlled thermal validation, this work establishes a quantitative link between hydration state, dielectric dispersion, and thermal response and explicitly identifies hydration-driven dynamics as the key factor governing both electromagnetic performance and long-term stability under RF loading.

2. Materials and Methods

2.1. Materials

The muscle-equivalent phantom was prepared using deionized water, agar powder, polyethylene (PE) powder, sodium chloride (NaCl), TX-151 emulsifier, and sodium azide. The composition and suppliers are listed in

Table 1. Composition and suppliers of the muscle-equivalent phantom.

Ingredient	Quantity (g)	Supplier (City, Country)
Deionized water	3375	BioQuím JD (Bogotá, Colombia)
Agar agar	104.6	McKenna Group S.A.S. (Bogotá, D.C., Colombia)
Polyethylene powder (∼500 μm)	337.5	Alfa Aesar (Ward Hill, MA, USA)
Sodium chloride	37.6	Todo Piscinas y Aseo ISA (Medellín, Colombia)
Sodium azide	2	ITW Reagents (Castellar del Vallès, Barcelona, Spain)
TX-151 emulsifier	84.4	Balmar LLC (Lafayette, LA, USA)

.

2.2. Phantom Fabrication

The sample preparation procedure was based on research from [8], utilizing proportions that create a phantom with electrical properties similar to those of human muscle tissue. First, sodium chloride (NaCl) and sodium azide were dissolved in deionized water using a magnetic stirrer while the solution was heated. Then, the mixture was heated to 70 ºC to facilitate the incorporation of agar.

Once the agar was fully incorporated, the magnetic stirrer was replaced with a mixing motor to combine the polyethylene powder and the TX-151 solidifier simultaneously. Finally, the mixture was allowed to rest in a sealed container covered with plastic wrap for 24 h. It was then fully wrapped in the same film for storage at room temperature, ensuring it was not exposed to direct sunlight. The phantom was stored for two months before characterization. Any variations observed in the following measurements correspond to intra-sample changes associated with gradual moisture loss within the same specimen, and not to differences between independently fabricated samples. Figure 1 shows the summary of the main steps for the development of the phantom.

2.3. EDS-Based Chemical Characterization

The chemical composition of the phantom was analyzed by energy-dispersive X-ray spectrometry (EDS) using an X-MaxN silicon drift detector (SDD, Oxford Instruments, Abingdon, Oxfordshire, UK) coupled to a field-emission scanning electron microscope (FE-SEM, JEOL JSM-7100F, JEOL Ltd., Tokyo, Japan). The microscope was operated in high-vacuum mode at an accelerating voltage of 20 kV and a working distance of approximately 10 mm. The analysis was performed at multiple locations across the phantom surface to assess compositional homogeneity. A total of ten independent EDS measurements were acquired under identical acquisition parameters, and the reported elemental composition corresponds to the averaged values obtained from these measurements.

2.4. Thermogravimetric and Thermal Properties Measurements

Thermogravimetric analysis (TGA) was conducted on a Discovery 550 Analyzer (TA Instruments, New Castle, DE, USA) to assess the thermal stability of the muscle-equivalent phantom under hyperthermia-relevant conditions. Samples were heated in ambient air from room temperature up to 800 °C at a constant 10 °C/min, allowing us to pinpoint any mass-loss transitions and confirm that the phantom matrix remains structurally intact well above therapeutic temperature ranges [1]. In addition, to analyze the thermal behavior of the phantom at body temperature (37 °C), it was heated at a rate of 1 °C/min up to 37 °C. Finally, an isothermal process was conducted for 70 min.

Furthermore, the phantom’s capacity to conduct and store heat was evaluated using a KD2 Pro Thermal Properties Analyzer (Decagon Devices, Inc., Pullman, WA, USA). The KD2 Pro features a dual-needle sensor (30 mm active length, 6 mm needle spacing) that captures both thermal conductivity and volumetric heat capacity simultaneously. To reduce ambient fluctuations, the sample was placed inside a climatic chamber (Memmert, model HPP 110). The samples were kept at temperatures of 25 °C, 35 °C, and 45 °C, with a constant relative humidity of 65%. The sensor was placed 3 cm into the phantom and allowed to equilibrate for 15 min. This time was enough to ensure that the temperature was uniform across the surrounding sample before taking a reading. In addition, each measurement was performed four times to verify accuracy and reproducibility. The experimental setup for these measurements is shown in Figure 2.

2.5. Optical Characterization

Optical characterization of the tissue-mimicking phantom across the visible and near-infrared (VIS–NIR) spectral range was carried out using the experimental setup illustrated in Figure 3. The setup was based on a laser-driven plasma white light source (Energetiq EQ-99-FC, Wilmington, MA, USA), which provides a broadband and spectrally stable emission spanning from 190 nm to 2500 nm, and the reflected optical signal was measured using an optical spectrum analyzer (OSA) (Yokogawa AQ6373, Tokyo, Japan). Both the light source and the OSA were coupled to the measurement system through a bifurcated fiber-optic probe (Ocean Optics, QR200-7-UV-VIS). The probe was positioned at an incidence angle of 45° relative to the phantom surface, as is illustrated in Figure 3. This angular configuration was selected to minimize specular reflections [29].

The VIS–NIR measurements were performed on phantom slabs with dimensions of 8 × 8 × 1 cm. In this experiment, the fiber-optic probe was positioned at different locations randomly distributed across the sample surface. The objective of this analysis was to assess the macroscopic spatial homogeneity of the fabricated phantom using independent measurement points. In this way, the study aimed to identify potential localized variations associated with mixing, gelation, or solvent redistribution processes. In addition, a custom-designed mechanical holder was used to fix both the probe orientation and the probe–sample distance to ensure repeatability and spatial consistency.

Spectral measurements were conducted using a linear wavelength scale with a nominal resolution of 5 nm, covering the spectral range from 350 nm to 1200 nm with a wavelength increment of 1 nm. Before phantom measurements, the system was calibrated using a certified diffuse reflectance standard (USRS-99-010, Labsphere, Palo Alto, CA, USA). Following calibration, diffuse reflectance spectra of the tissue-mimicking phantom samples were acquired under identical experimental conditions.

2.6. Microwave Dielectric Spectroscopy

The dielectric response of the phantom was characterized using a coplanar waveguide (CPW) transmission-line approach, which provides a robust and broadband method for extracting complex permittivity from calibrated scattering parameters. The CPW was fabricated on Rogers RO3003 laminate (relative permittivity ${\epsilon }_{r1}\approx 3.0$, loss tangent $\tan \delta <0.003$), with a substrate thickness of $1.52\mathrm{mm}$, a center conductor width of $1.27\mathrm{mm}$, and a slot gap of $1.515\mathrm{mm}$. A uniform phantom slice with a thickness of $18\mathrm{mm}$ was carefully prepared and placed directly on the CPW as a superstrate (Figure 4). Two-port measurements were performed using a vector network analyzer (ZNA26, Rohde & Schwarz, Munich, Germany) over the frequency range 1.5–4.0 GHz. A full two-port SOLT calibration was applied at the probe reference planes prior to data acquisition. Each measurement was repeated seven times under identical conditions to assess repeatability.

From the calibrated S-parameters and the known physical length of the CPW section, the complex propagation constant $\gamma =\alpha +j\beta$ was extracted, where $\alpha$ and $\beta$ denote the attenuation and phase constants, respectively. Under the quasi-TEM approximation, the phase constant is related to the effective permittivity ${\epsilon }_{\mathrm{eff}}$ as

$$\beta \left(f\right)={\displaystyle \frac{2\pi f}{c}}\sqrt{{\epsilon }_{\mathrm{eff}}\left(f\right)},$$

(1)

where c is the speed of light in free space, and ${\epsilon }_{\mathrm{eff}}$ is the efective relative permittivity of the CPW. To mitigate constant phase offsets associated with calibration and connector effects, ${\epsilon }_{\mathrm{eff}}$ was evaluated from the frequency derivative of $\beta \left(f\right)$ rather than from its absolute value.

The relative permittivity of the superstrate (${ϵ}_{r2}$) is determined based on its relationship with the effective permittivity of the CPW [30], as expressed in Equation (2).

$${ϵ}_{eff}=1+q_1({ϵ}_{r1}-1)+q_2({ϵ}_{r2}-1),$$

(2)

where $q_1$ and $q_2$ are geometry-dependent filling factors. These coefficients depend on the CPW dimensions and the thicknesses of the involved layers and were computed following the formulation reported in [30].

Dielectric losses in the phantom were quantified from the attenuation constant by isolating the dielectric contribution associated with the superstrate. Within the quasi-TEM approximation, the dielectric attenuation term can be written as:

$${\alpha }_d={\displaystyle \frac{\pi }{{\lambda }_0\sqrt{{\epsilon }_{\mathrm{eff}}}}}\left({\epsilon }_{r2}{q}_2\tan {\delta }_2\right),$$

(3)

where ${\lambda }_0$ is the free-space wavelength and $\tan {\delta }_2$ is the loss tangent of the phantom material. The effective conductivity of the phantom was finally obtained from $\tan {\delta }_2$ using the standard relation $\sigma =\omega {\epsilon }_0{\epsilon }_{r2}\tan {\delta }_2$.

2.7. Reflection-Mode THz-TDS Setup and Acquisition Protocol

The experimental configuration illustrated in Figure 5 schematically represents the reflection-mode THz-TDS arrangement used in this work for the characterization of skin-mimicking phantom samples. Measurements were performed with a benchtop THz spectrometer (TDS1008, BATOP GmbH, Jena, Germany) based on internal photoconductive antennas (Tx/Rx) driven by a femtosecond laser source at 780 nm (pulse duration 100 fs, repetition rate 50–100 MHz, average power 40 mW). According to the instrument specifications, the TDS1008 provides a bandwidth of 0.05–4.5 THz, dynamic range of ≥85 dB, a maximum scan range of up to 500 ps, and a spectral resolution of 2 GHz at the maximum scan range.

All measurements were carried out in reflection mode using the SHR (Sample Holder Reflection) mount installed in the internal sample compartment. In this geometry, the THz beam emitted by the Tx antenna is redirected by an internal reflective plate toward the sample plane under oblique incidence; in the SHR configuration, the angle of incidence is 30° by default. The field reflected by the sample is redirected back into the detection path and collected by the Rx antenna, yielding the reflected waveform in the time domain.

A three-step acquisition sequence was used to ensure a consistent reference across the complete measurement campaign. First, a baseline trace was recorded by closing the spectrometer’s metallic shutter to block the THz radiation from reaching the Rx antenna; this baseline captures the system background under identical acquisition settings. Next, under the same conditions, the reference signal was acquired by opening the shutter and placing a protected silver mirror (PF30-03-P01, Thorlabs, Newton, NJ, USA) at the sample position, using the sample area of the SHR mount to reflect the THz beam. The mirror was positioned in the same plane as the phantom samples. Finally, the phantom was placed in the same plane after the mirror was removed. Two phantom slabs with lateral dimensions of 2.7 cm × 2.7 cm and thicknesses of 0.5 cm and 0.3 cm were evaluated. The reference trace acquired with the mirror was used as a fixed reference for all subsequent measurements, since no changes were made to the spectrometer configuration or acquisition parameters.

Time-domain signals were acquired by scanning the optical delay line within a restricted temporal window around the reflected pulse. For all reflection measurements, the acquisition parameters were kept constant: a stepwidth of 0.050 ps, a start position of 245 ps, an end position of 290 ps, and a time constant of 1.5 s. Here, the start and end positions define a 45-ps temporal window selected to capture the main reflected pulse and its immediate oscillations without extending the scan unnecessarily, which would increase acquisition time. The stepwidth sets the temporal sampling of the waveform, while the time constant corresponds to the effective integration time at each delay position, selected to improve measurement stability and signal-to-noise ratio for weak reflections from lossy, water-rich samples. The time constant (1.5 s) and fixed delay sampling (0.05 ps) improve short-term stability for weak reflections, so peak-to-peak and RMS metrics are used here as robust relative observables to track temporal evolution under constant measurement conditions.

To evaluate the time-dependent evolution of the phantom response associated with drying, the spectrometer was programmed to acquire one measurement every 30 min over 7.5 h, resulting in 15 consecutive measurements per sample. After completing the sequence for the first thickness, the phantom was replaced with the second slab, and the same protocol was repeated under identical acquisition settings.

The measurement data were exported using the T3DS software, which stores the baseline, reference, and sample traces in a single dataset together with the acquisition metadata. For consistency, basic preprocessing consisted of removing the baseline contribution from the reference and sample traces and subsequently comparing the phantom response with respect to the mirror reference under unchanged acquisition conditions. This methodological strategy was adopted to minimize systematic variability and to ensure that the observed changes across repeated scans predominantly reflect temporal evolution of the phantom rather than changes in the experimental settings.

2.8. Microwave Heating and Thermal Validation Setup

Microwave thermal validation was performed to evaluate the thermophysical stability of the tissue-mimicking phantom under controlled RF exposure. The experimental setup, shown in Figure 6, combines continuous-wave excitation at 2.45 GHz and non-contact infrared thermography for surface temperature monitoring. This setup and acquisition methodology were previously implemented to study the behavior of organic and inorganic materials [31].

In the experimental setup, the RF excitation was generated using a USRP platform (National Instruments NI USRP-2922, Austin, TX, USA) operating at 2.45 GHz in continuous-wave mode. Two excitation regimes were used. For low-power thermal stability tests, the USRP output was set to 14 dBm and routed through an external RF power amplifier operated at a nominal gain of 28 dB. For hyperthermia-level experiments, the USRP output was set to 20 dBm and amplified by 33 dB to reach therapeutic temperatures. Moreover, a three-port circulator was incorporated to protect the amplifier from reflected power, with a matched 50 Ω load connected to the isolated port. Microwave energy was delivered to the phantom using a unidirectional antenna positioned parallel to the phantom surface, as schematically illustrated in Figure 6. The antenna–sample separation was fixed at 5 mm using a mechanical support to maintain consistent electromagnetic coupling. The phantom was placed on a flat plastic-coated holder to reduce parasitic reflections and minimize conductive heat losses.

Surface temperature was monitored using a calibrated infrared thermal camera (Optris PI 450i, Optris GmbH, Berlin, Germany) operating in radiometric mode, which enables pixel-wise absolute temperature measurements. The camera was positioned normal to the phantom surface at a fixed distance of approximately 40 cm. Before the experiments, the camera was calibrated following the manufacturer’s procedure. The surface emissivity was set to 0.98. Ambient reflections were minimized by shielding the experimental area and avoiding high-reflectivity objects within the camera field of view.

Two operating temperature conditions were evaluated. In the low-power regime, the applied excitation resulted in a stable surface temperature of approximately 26 °C, corresponding to near-ambient conditions. In the hyperthermia-level regime, the excitation power was gradually increased until the phantom reached an initial steady-state surface temperature of 44.6 °C, representative of therapeutic hyperthermia. Once thermal stabilization was reached, infrared images were acquired at predefined time intervals during continuous excitation. At each time point, multiple consecutive thermal frames were recorded under identical conditions to quantify short-term variability and compute mean surface temperature values and associated standard deviations. To avoid cumulative heating or dehydration effects, consecutive experiments were performed on previously unexposed regions of the phantom by rotating the sample between tests.

2.9. Multivariate Analysis

To evaluate the spatial reproducibility of the VIS–NIR apparent absorbance spectra, principal component analysis (PCA) was applied as a multivariate analysis tool. In this case, each spectrum can be considered as a vector containing absorbance values at different wavelengths. When spectra are measured at several positions on the phantom surface, the full dataset can be organized as a matrix in which each row corresponds to one measurement position and each column corresponds to a wavelength.

Before applying PCA, the dataset was mean-centered in order to analyze variations with respect to the average spectral response. This was performed by subtracting the mean spectrum from each individual spectrum:

$$X_c=X-\overline{X},$$

where X represents the original spectral dataset and $\overline{X}$ is the average spectrum calculated across all positions. This step ensures that the analysis focuses on differences among measurement locations rather than on absolute absorbance levels. No scaling process was implemented because all variables correspond to absorbance values expressed in the same units. PCA determines a new set of orthogonal components, called principal components (PCs), that describe the main directions of variability within the dataset. The first principal component (PC1) accounts for the largest fraction of the total variance, while subsequent components capture progressively smaller and independent contributions. The PCA analysis was carried out using Python (v3.14.0).

3. Results

3.1. Chemical Characterization Results

The EDS elemental distribution map in Figure 7 reveals a predominantly carbonaceous matrix (Figure 7b), with dispersed sodium (Figure 7c) and chloride (Figure 7d).

Table 2. Elemental composition of the phantom expressed as weight percentage (wt%) obtained by EDS analysis. The reported values correspond to the mean and standard deviation of measurement number = 10 spatially distinct spot measurements acquired from a single phantom sample.

Element	Weight Percentage (wt.%)
	Mean	Deviation
Carbon	92.81	1.62
Sodium	1.61	0.46
Chloride	1.34	0.43

reports the elemental composition of the phantom expressed as the weight percentage (wt%) obtained by EDS analysis. This elemental composition is consistent with the phantom formulation, where agar and polyethylene constitute the carbon-rich polymeric backbone, while dissolved NaCl salt provides ionic content to adjust dielectric conductivity towards values representative of muscle tissue in the microwave band.

The relatively large standard deviations reflect microscale heterogeneity inherent to manual mixing and gel casting, as localized salt clustering and slight variations in polymer network density lead to spot-to-spot fluctuations in measured elemental content. It should be noted that the reported standard deviations arise from spatially distinct EDS spot measurements performed on a single phantom sample (measurement number = 10), rather than from multiple independently fabricated samples. Because EDS cannot detect light elements like oxygen and hydrogen, the remaining ∼6 wt % likely corresponds to oxygen in the agar backbone and residual moisture retained in the gel matrix.

3.2. Thermal Characterization

The TGA and differential thermal analysis (DTG) of the phantom are shown in Figure 8. The phantom was analyzed to evaluate its chemical composition (Figure 8a) and behavior at body temperature (Figure 8b). The weight loss of (76.76%) between 25 °C and 200 °C is attributed to the evaporation of water from the phantom material [5]. The weight loss of 22.05% between 200 °C and 800 °C is associated with the combustion of polyethylene, agar, and TX-151 (Super Stuff) [32,33]. Finally, the ash content of 1.19% corresponds to the inorganic components such as sodium, which is confirmed by EDS results.

Skeletal muscle is the largest organ in the human body, accounting for approximately 40% of total body mass, and is composed of about 76% water [34]. This high water content plays a key role in metabolic activity, mechanical response, and thermal transport [35]. According to the TGA analysis, the developed phantom exhibits a water content of 76.76%, closely matching that of real muscle tissue, which supports the biomimetic relevance of the proposed formulation. At the same time, these results indicate that the thermophysical properties of the phantom are inherently sensitive to temperature-induced water loss. As a consequence, thermal measurements must be performed under well-controlled temperature conditions, since even modest temperature fluctuations can promote partial water evaporation and lead to measurable variations in thermal conductivity and volumetric specific heat capacity.

Figure 9 shows the response of thermal conductivity and volumetric heat capacity as a function of temperature variations. An approximate reduction of 57.4% is seen in the thermal conductivity of the phantom material, which drops from 0.588 to 0.25075 W/(m·K). According to TGA analysis, the reduction can be related to water evaporation with rising temperature. In addition, the volumetric heat capacity decreases with increasing temperature. An analogous pattern was observed in which the volumetric heat capacity reduces from 3.1935 MJ/(m³·K) to 1.664 MJ/(m³·K) with the rise in temperature from 25 °C to 45 °C. The decrease in volumetric heat capacity as the temperature rises can be attributed to the evaporation of water. Water possesses a high specific heat capacity of 4.2 J/g K, allowing it to absorb thermal energy to break its hydrogen bonds [36,37]. As the temperature rises, water evaporates, which reduces the energy required to raise the temperature of the phantom.

The Tukey method was used to identify significant differences among the means of the thermal conductivity tests. The results showed the following p values: 35 °C vs. 25 °C ($p=1.09\times {10}^{-3}$), 45 °C vs. 25 °C ($p=9.95\times {10}^{-6}$), and 45 °C vs. 35 °C ($p=9.95\times {10}^{-6}$). All pairwise comparisons were statistically significant at the 0.05 significance level. Additionally, the Tukey method was also applied to evaluate significant differences among the means of the volumetric specific heat capacity. The obtained p values were 35 °C vs. 25 °C ($p=5.87\times {10}^{-4}$), 45 °C vs. 25 °C ($p=9.95\times {10}^{-6}$), and 45 °C vs. 35 °C ($p=9.95\times {10}^{-6}$). All differences among the temperature pairs were statistically significant, even at the 0.001 level. These results suggest that temperature significantly influences thermal conductivity and volumetric specific heat capacity.

3.3. VIS–NIR Optical Spectral Analysis of the Tissue-Mimicking Phantom

The mean normalized relative reflectance spectrum of the tissue-mimicking phantom in the VIS–NIR range is shown in Figure 10a. The curve corresponds to the average of eight independent measurements, and the shaded region represents the associated standard deviation. The obtained results reveal that the reflectance decreases from the near-ultraviolet toward the visible region, exhibits a peak around 590–610 nm, and then gradually decays toward longer wavelengths. Beyond 650 nm, the spectral profile remains smooth across the measured range. The dispersion indicated by the shaded band (±SD) remains moderate over most of the spectrum, indicating limited spatial variability across the phantom surface.

The obtained apparent absorbance spectrum is shown in Figure 10b. The results show that the absorbance of the fabricated phantom increases gradually from the visible to the near-infrared region, with a shallow minimum near 580–600 nm that mirrors the reflectance maximum. The narrow standard deviation band confirms good reproducibility across different measurement locations.

To further quantify spatial reproducibility, the coefficient of variation (CV) of the apparent absorbance was computed as a function of wavelength (Figure 10c). Over most of the reliable spectral window, the CV remains below 15%, evidencing a homogeneous optical response across different measurement points. A pronounced increase in CV is observed only near the upper wavelength limit (1050–1100 nm), which is attributed to the reduced signal-to-noise ratio of the optical setup combined with the stronger absorption of water-rich media, both effects amplifying normalization uncertainty.

The absorbance was integrated over representative VIS–NIR intervals (

Table 3. Integrated apparent absorbance of the tissue-mimicking phantom over selected VIS–NIR spectral bands. Values are reported as mean ± standard deviation across all measurement points.

Spectral Band	Wavelength Range (nm)	Integrated Absorbance (a.u.)	CV (%)
VIS	400–700	$368.56\pm 39.10$	10.61
NIR-I	700–900	$265.91\pm 30.38$	11.42
Water band	900–1000	$140.52\pm 17.71$	12.60
NIR-II	1000–1100	$142.06\pm 19.78$	13.93

). The visible range exhibits the highest integrated absorbance, exceeding that of the NIR-I band, confirming that optical attenuation is dominated by shorter wavelengths where scattering and absorption accumulate. In contrast, the water and NIR-II bands present nearly identical integrated absorbance values ($140.5\pm 17.7$ and $142.1\pm 19.8$ a.u.·nm), consistent with a smooth, water-governed absorption baseline rather than discrete spectral features. The corresponding coefficient of variation remains between 10 and 14%, consistent with the wavelength-resolved reproducibility analysis.

To give a complementary and quantitative assessment of spectral repeatability, the PCA was applied to the VIS-NIR apparent absorbance spectra collected from eight points on the phantom surface. Figure 11 presents the PCA score plot, where PC1 accounts for 90.8% of the total spectral variance, and PC2 explains an additional 6.9%. The dominant contribution of PC1 indicates that most of the variability among measurement positions is attributable to a global intensity-related variation spanning the entire spectral range. This pattern shows that modest variance between locations is driven primarily by minor variations in global absorbance magnitude, which may be attributed to small surface heterogeneities or local thickness fluctuations rather than to structural changes in the spectral profile itself.

The results shown in Figure 11 reveal that only one measurement position (Position 7) appears clearly separated from the main cluster in PCA space. This separation indicates a localized deviation relative to the dominant optical behavior of the phantom. To quantify this observation, the root-mean-square error (RMSE) of each spectrum relative to the mean spectral profile was computed. As summarized in

Table 4. Spectral reproducibility metrics and robust outlier detection across measurement positions used in the PCA analysis. The Point ID corresponds to the numeric labels shown in Figure 11.

Point ID	Measurement Position	RMSE vs. Mean (a.u.)	Robust Z-Score	Outlier
1	Position 1	0.2158	0.69	No
2	Position 2	0.2245	0.79	No
3	Position 3	0.0981	−0.66	No
4	Position 4	0.1465	−0.10	No
5	Position 5	0.1640	0.10	No
6	Position 6	0.1340	−0.24	No
7	Position 7	0.4686	3.59	Yes
8	Position 8	0.0848	−0.81	No

, Position 7 exhibits the largest deviation (RMSE = 0.4686 a.u.) and is identified as an outlier using a robust Z-score criterion (3.59). All remaining positions show substantially lower RMSE values and cluster closely in PCA space, confirming a high degree of spectral consistency across independent measurement locations. Importantly, the PCA results agree with the structural and compositional analyses reported in the preceding sections.

3.4. Microwave Dielectric Properties

Figure 12 shows the frequency-dependent microwave dielectric properties of the tissue-mimicking phantom measured between 1.5 and 4.0 GHz using the CPW-based characterization method described in Section 2.6. This range includes the 2.45 GHz ISM frequency used for hyperthermia and SAR applications. As observed in Figure 12, the relative permittivity exhibits a monotonically decreasing trend with increasing frequency, dropping from approximately 37.9 at 1.5 GHz to about 36.5 at 4.0 GHz. This trend is consistent with water-dominant soft tissue analogs and reflects the expected reduction in dipolar polarization response as frequency increases.

In contrast, the electrical conductivity shows a systematic increase with frequency, rising from approximately 1.6 S/m at 1.5 GHz to nearly 4.5 S/m at 4.0 GHz. This trend is consistent with enhanced dielectric losses and increased ionic conduction mechanisms at higher microwave frequencies. At the target frequency of 2.45 GHz, the phantom exhibits a relative permittivity of ${\epsilon }_r$ = 37.4 and an electrical conductivity of $\sigma$ = 2.4 S/m. These values fall within the range reported for muscle-equivalent tissues in the reference study [8] and in established human tissue databases [38,39]. In fact, they are highlighted in Figure 12 by the dashed vertical marker. The smooth frequency dependence of both ${\epsilon }_r$ and $\sigma$, without abrupt dispersion features, indicates a stable and homogeneous electromagnetic response across the investigated band. This behavior supports the suitability of the phantom for broadband microwave applications, particularly for controlled exposure and hyperthermia validation studies. Figure 13 shows the statistical dispersion of the seven repeated microwave dielectric measurements at 2, 3, and 4 GHz using a box plot. The interquartile range remains narrow across all frequencies, with a maximum relative variation below 9% for relative permittivity and 8% for conductivity. Although dispersion slightly increases at higher frequencies, the absence of extreme outliers and the limited spread of the data confirm measurement repeatability and stability of the CPW-based extraction method.

3.5. THz Temporal Stability Analysis

Figure 14 and Figure 15 summarize the reflection-mode THz response of the phantom (0.5 cm and 0.3 cm) compared with the silver-mirror reference. Relative to the mirror, the phantom produces a strongly attenuated reflected waveform, consistent with the high losses expected for a water-rich matrix in the THz range (Figure 14a and Figure 15a). This strong attenuation reduces the signal-to-noise ratio and compromises phase stability across repeated scans. Under these conditions, inversion of the complex reflection response to reliably retrieve frequency-dependent refractive index and absorption coefficient becomes ill-conditioned; therefore, the THz analysis is restricted to robust observables to track temporal evolution. An uncertainty estimate based on the noise floor (from the shutter-closed baseline acquisition) indicates that the measurement fluctuations are smaller than the observed temporal decay/drift trend.

The most relevant observation is a progressive temporal shift of the reflected pulse over sequential acquisitions, indicating a cumulative drift in effective time-of-flight. Importantly, this drift is inferred from the temporal location of the dominant reflected feature, which is less sensitive than amplitude to alignment or coupling variations. For the 0.5 cm specimen, the main pulse remains within the analysis window across all 15 measurements. For the 0.3 cm specimen, the shift occurs faster, and the pulse exits the analysis window at later times; therefore, amplitude metrics and drift estimation are reported only for the initial subset (first 8 measurements, 0–3.5 h) where the same reflected feature can be consistently tracked (Figure 15c,d).

In the frequency domain (Figure 14b and Figure 15b), the useful spectral content is mainly confined below ∼2 THz and preserves a similar overall envelope, while the spectral magnitude decreases gradually with time, consistent with the decay of the time-domain amplitude metrics. Drift quantification expressed as an equivalent optical path-length change (Figure 14d and Figure 15d) yields a moderate slope for 0.5 cm (∼−0.117 mm/h) and a larger magnitude for 0.3 cm (∼−0.334 mm/h), confirming a faster temporal evolution for the thinner specimen. This thickness dependence is consistent with hydration-related temporal dynamics under ambient conditions (e.g., water redistribution and/or gradual drying) and highlights thickness as a practical factor governing temporal stability during THz characterization.

3.6. Microwave Thermal Validation

The experiments were conducted under non-adiabatic conditions, allowing natural convective and radiative heat transfer to the surrounding environment. This configuration reflects realistic hyperthermia scenarios rather than thermally isolated systems. The analysis focused on the temporal evolution of the maximum temperature within a fixed region of interest (ROI), extracted from radiometric infrared images acquired during RF exposure at 2.45 GHz.

Figure 16a shows the temporal evolution of T_max,ROI under low-power microwave excitation (forward power = 25 mW), remaining close to ambient temperature (≈26 °C). At each acquisition time, five consecutive measurements were performed, and the reported values represent the mean temperature. Likewise, the standard deviation is shown as error bars. The obtained results demonstrate a stable thermal response over the entire acquisition period, approximately 390 min. In this case, the retrieved values of T_max,ROI are close to ambient temperature with minor changes. Furthermore, the total coefficient of variation was computed, resulting in a CV of 3.41%, which is low, confirming the strong short-term repeatability and the lack of cumulative thermal effects under these conditions. These findings show that, in the absence of power at hyperthermia levels, the phantom provides a suitable baseline for thermal validation tests.

In contrast, Figure 16b presents the evolution of $T_{\max ,\mathrm{ROI}}$ under high-power microwave excitation (forward power = 100 mW), with the initial temperature reaching approximately 44.6 °C, representative of therapeutic hyperthermia conditions. Unlike the ambient-temperature regime, the hyperthermia experiment reveals a clear and systematic temporal decay in $T_{\max ,\mathrm{ROI}}$. After initial stabilization near the target temperature, the maximum ROI temperature decreases progressively during the analysis interval, which was $120\min$. To quantify the temporal trend, a linear regression model was applied to the temperature–time data, yielding a negative slope of −3.985 °C/h, with a high coefficient of determination ($R^2=0.975$) and a statistically significant p-value ($p=1.71\times {10}^{-3}$). Within the time window analyzed, the temperature decrease follows an approximately linear trend. It is important to note that the observed thermal drift arises from the combined influence of sustained microwave loading and non-adiabatic boundary conditions in a water-dominant matrix. Therefore, the progressive temperature decrease should not be interpreted as intrinsic instability of the phantom, but rather as a response under realistic energy exchange conditions.

In addition, some representative infrared images acquired during both low- and high-power experiments are shown in Figure 17. Under low-power excitation, the thermal maps exhibit weak spatial gradients and minimal temporal variation, consistent with the stable $T_{\max ,\mathrm{ROI}}$ values and low CV reported above. In contrast, hyperthermia-level excitation produces a rapid temperature increase followed by the development of more pronounced thermal gradients within the phantom. At later time points, localized regions of elevated temperature and subtle surface changes become visible in the high-power images. These spatial features qualitatively support the quantitative ROI analysis, indicating progressive thermal alterations during prolonged exposure.

4. Discussion

The purpose of this work was to verify whether the fabricated phantom reproduces the coupled dielectric–thermal behavior of hydrated biological tissue across multiple spectral regimes. The characterization was intentionally multimodal to ensure that composition, dielectric response, optical absorption, and microwave thermal performance remain physically consistent. First, the thermogravimetric analysis confirms that the phantom is dominated by a water-rich phase embedded in a polymeric matrix. The main mass loss below 200 °C corresponds to water evaporation, establishing hydration as the governing factor of its electromagnetic and thermal behavior. This finding is critical to the study because the water content of the phantom influences dipole polarization at microwave frequencies and broadband absorption in the optical and THz regimes. Similarly, water increases the material’s heat storage capacity when subjected to the hyperthermia applicator.

The microwave dielectric comparison presented in

Table 5. Comparison of electromagnetic and thermophysical properties of the tissue-mimicking phantom at 2.45 GHz.

Property	This Work	Ref. [8]	Tissue Databases a
Relative permittivity, ${\epsilon }_r$	37.4	47.0	52.7
Electrical conductivity (S/m)	2.4	2.25	1.74
Volumetric heat capacity (MJ/(m3K))	3.19	3.63	3.73
Thermal conductivity (W/(mK))	0.59	0.55	0.495

^a Values reported in human muscle tissue databases [38,39].

reveals a more discriminative and physically informative contrast. The close match in electrical conductivity with the reference phantom [8] suggests that the ionic concentration and connectivity within the water-rich matrix were accurately replicated. This demonstrates that charge transport processes, dominated by free-ion mobility, are resilient to modest hydration fluctuations and can be successfully regulated by controlling the NaCl content. The relative permittivity differs significantly from both the reference phantom and database values [38,39]. This pattern reflects the inherent sensitivity of dipolar polarization to the fraction, condition, and spatial distribution of bound and free water, as well as polymer-water interactions within the solid matrix. As a result, tiny changes in hydration or microstructural organization have a disproportionate effect on permittivity while keeping conductivity relatively steady. This asymmetric sensitivity between permittivity and conductivity is not a limitation of the phantom but rather a physically expected feature of hydrated polymer systems. It explains the simultaneous agreement in $\sigma$ and deviation in ${\epsilon }_r$ observed in Table 5 and is fully consistent with the monotonic frequency dispersion reported in the microwave characterization. From an electromagnetic perspective, this dielectric contrast governs near-field distribution, energy confinement, and microwave power deposition, which are central to hyperthermia and SAR validation. These same mechanisms form the physical basis for a wide range of microwave sensing and characterization techniques, in which materials with dielectric properties different from those of air induce measurable field perturbations. Such near-field perturbations have been extensively exploited in resonant and non-resonant microwave structures to induce frequency shifts and characterize dielectric variations in biological tissues and tissue-mimicking materials [38,39,40,41,42].

On the other hand, the VIS–NIR optical characterization shows a smooth broadband spectral response with limited spatial variability. The coefficient-of-variation analysis and PCA confirm that variability is dominated by global intensity scaling rather than spectral reshaping, indicating macroscopic homogeneity. Importantly, the integrated NIR absorbance follows the expected water-dominant baseline. While no explicit absorption model was fitted, the spectral behavior is consistent with the high water fraction identified by TGA. In hydrated media, overtone and combination bands of O–H vibrations dominate the NIR region; therefore, the broadband increase toward longer wavelengths is physically coherent with the measured water content.

In addition, the THz measurements provide a higher sensitivity probe of hydration dynamics. Strong broadband attenuation is observed, consistent with water-dominated dielectric relaxation in this frequency range. Time-lapse reflection experiments reveal a progressive temporal shift of the dominant pulse, more pronounced in thinner slabs. This behavior indicates a gradual modification of the effective propagation conditions within the phantom during repeated acquisition. Given the strong dependence of THz absorption and dispersion on water content, the magnitude of attenuation and its temporal sensitivity are consistent with the quantified water fraction obtained from TGA. In water-rich systems, even minor hydration variations produce measurable changes in THz absorption and effective time-of-flight. Therefore, the observed THz drift is attributed to gradual moisture redistribution rather than structural instability of the phantom matrix.

Microwave thermal validation further supports this interpretation. Under low-power excitation, the phantom maintains a stable surface temperature over prolonged periods, demonstrating that RF exposure at ambient conditions does not induce measurable hydration changes or structural alteration. This confirms that the dielectric and thermophysical properties remain invariant under baseline electromagnetic loading. In contrast, under hyperthermia-level excitation, a gradual decrease in surface temperature is observed during sustained exposure. Within the first 30–40 min, however, the temperature remains close to the initial steady state, indicating that the material maintains operational stability during typical short-duration hyperthermia validation experiments. Under these conditions, prolonged microwave loading in a water-dominant matrix promotes gradual moisture redistribution and partial evaporation, which in turn modifies effective microwave absorption and heat retention. The observed thermal drift, therefore, results from the coupled interaction between dielectric losses, energy deposition, and realistic environmental exchange.

For this reason, the phantom demonstrates macroscopic compositional uniformity and dielectric reproducibility under baseline conditions. However, like hydrated biological tissue, it exhibits moisture-dependent evolution under sustained thermal stress. This response does not invalidate its use; rather, it reflects the intrinsic coupling between hydration, dielectric losses, and heat diffusion that characterizes real soft tissues in hyperthermia scenarios.

5. Conclusions

In this work, a multimodal experimental framework was implemented to assess whether the fabricated phantom reproduces the coupled dielectric–thermal behavior of hydrated muscle tissue across microwave, optical, and THz regimes. Thermogravimetric analysis confirms dominant water content, which governs the dielectric dispersion, broadband absorption, and thermal storage capacity of the material.

Likewise, the fabricated phantom exhibits a relative permittivity of 37.4 and an electrical conductivity of 2.4 S/m at 2.45 GHz. While the conductivity closely matches reference formulations and remains within the range reported in tissue databases, the permittivity is lower than literature muscle values, which typically range between 47 and 52.7. This deviation reflects the known sensitivity of dipolar polarization to hydration state and polymer–water interactions, whereas ionic transport mechanisms remain accurately replicated. From an electromagnetic perspective, the phantom preserves the correct order of magnitude and dispersion behavior required for controlled hyperthermia and SAR validation in the 2.4 GHz ISM band.

Optical and THz analyses further demonstrate macroscopic compositional homogeneity. VIS–NIR measurements show smooth broadband absorption with limited spatial variability, and PCA confirms that spectral differences are governed by global intensity scaling rather than structural distortion. In the THz regime, strong broadband attenuation consistent with a water-rich matrix is observed together with measurable temporal sensitivity to hydration redistribution, highlighting the dynamic role of water in high-frequency electromagnetic response. Finally, the microwave thermal validation establishes both functional reliability and operational limits. Under low-power excitation, stable surface temperatures confirm compositional stability under baseline RF exposure. Under hyperthermia-level excitation, a significant thermal drift of −3.985 °C/h is observed during sustained irradiation. This drift arises from hydration-mediated redistribution under non-adiabatic conditions and represents the principal limitation of the material for prolonged high-power experiments. Nevertheless, this behavior remains physically coherent with the thermodynamic response of hydrated soft tissues subjected to sustained energy deposition. Thus, the proposed phantom reproduces the coupled dielectric–thermal behavior of hydrated muscle tissue at clinically relevant microwave frequencies while explicitly defining the hydration-driven constraints that govern its long-term thermal stability.