To achieve broadband impedance matching with biological tissues while maintaining excellent biosafety [25], appropriate probe materials must be selected and engineered. Biological tissues exhibit frequency-dependent dielectric properties, where low relative permittivity minimises transmission loss at high frequencies and high relative permittivity enhances signal integrity at lower frequencies [26]. Accordingly, ADT- ZrO 2 -HO01 zirconia ( ϵ r = 15.68) was chosen as the structural substrate owing to its high permittivity, mechanical robustness, and established biocompatibility in medical implants [27]. Its elevated permittivity enhances field confinement within the probe and facilitates deeper tissue penetration at lower frequencies. In contrast, copper foil ( ϵ r = 1) was employed as the signal transmission medium to optimise high-frequency performance.

To determine the optimal geometric configuration, three probe structures were designed and evaluated: rectangular, conical, and pyramidal. The final design features a contact area of 25.1 × 25.1 mm² and a height of 12.5 mm. The cross-sectional size of the tip is 0.1 × 0.1 mm². As shown in Figure 1, a multilayer biological tissue model (skin, fat, and muscle) with planar dimensions of 40 × 40 mm² was constructed in HFSS. The probe was positioned on the skin surface, and frequency-domain analysis was performed from 100 MHz to 20 GHz across 201 discrete frequency points. Figure 1 presents the resulting S 11 parameter curves for the different geometric configurations. Comparative analysis revealed that the pyramidal structure achieved both broadband tissue–impedance matching and a lower minimum matching frequency than the alternative designs, leading to its selection for final implementation.

Following structural finalization, Figure 2a details the dimensional specifications of the probe assembly and its connector interface. The sintered zirconia fabricated in a truncated square conical shape exhibits exceptional mechanical rigidity. This substrate features 35 μ m electrodeposited copper cladding on orthogonal surfaces, establishing both signal transmission pathways through a connector (a semirigid cable with an outer conductor diameter of 3.58 mm [28]), conductor integration, and a reference ground plane. The probe was impedance-matched to a 50 Ω VNA by adjusting the zirconia core cross-section and copper cladding so that, in tissue contact, reflections at the SMA interface were minimized and the measured S 11 was primarily sensitive to tissue-dependent impedance. Table 1 presents the specific parameters of the probe and biological tissue model. The operational principle involves transmitting excitation signals via the copper-zirconia and detecting subsequent electromagnetic echoes. The trapezoidal profile confers inherent broadband characteristics, enabling the acquisition of a wideband response. Through optimized impedance matching, this architecture achieves depth-specific tissue differentiation.

As shown in Figure 2b, the complete measurement system consists of the following core components. The vector network analyzer (VNA) generates broadband signals that propagate through a coaxial RF cable to the probe, which electromagnetically couples the energy to the biological tissue surface. Reflected signals from tissue interfaces are subsequently captured by the probe and routed back to the VNA for derivation of the reflection coefficient ( S 11 ) . To ensure measurement accuracy and repeatability, experimental controls included the placement of specimens atop a calibrated pressure scale to maintain a consistent contact force and the application of an 80 × 80 mm² copper ground plane overlay to standardize electromagnetic boundary conditions.

Before conducting experimental measurements, it is essential to rigorously calibrate the VNA. The calibration process involves several sequential steps. First, an RF line is used to connect the SMA interface to the VNA. Next, a calibration kit, which includes short, open, load, and through standards, is employed. During this phase, the configuration is systematically set, defining the measurement format and frequency range from 10 MHz to 20 GHz. A total of 2,001 discrete frequency points are designated. The output power is optimized to 7 dBm. All acquired data are archived in CSV format to ensure compatibility across different platforms. Finally, a comprehensive Short-Open-Load-Through (SOLT) calibration procedure is performed to eliminate systematic measurement errors.

To ensure system stability, the probe is rigidly affixed to a high-precision 3D platform, suppressing mechanical vibration artifacts and eliminating measurement errors from positional variations. Throughout the experiments, a precisely calibrated contact pressure of 5.0 ± 0.2 g was systematically maintained. This contact pressure was chosen based on pilot experiments in which increasing loads were applied until visible macroscopic deformation or fluid extrusion from the tissue occurred. Loads above 6 g produced appreciable indentation and changed the local tissue thickness, whereas loads below 4 g resulted in unstable Radio Frequency contact and higher variance in S 11 . A target of 5 g, therefore, represented a compromise between stable electrical contact and minimal mechanical distortion of the specimen. Triplicate measurements were performed at each sampling point to derive statistically robust averaged data. Before each tissue penetration, the zirconia-based probe underwent comprehensive sterilization with medical-grade ethanol, effectively preventing cross-contamination from residual biomaterials. All experimental protocols were executed under controlled laboratory environmental conditions, satisfying stringent requirements for electromagnetic characterization of biological tissues.

This study was approved by the Institutional Review Board of Chinese PLA General Hosptial approved the study protocol (Approval No. S2024-023-02). All participants were adults and provided written informed consent prior to participation. No minors were included in this study. The consent process included a detailed explanation of study objectives, potential risks, and data usage. Participants signed consent forms, which are securely archived by the research team. The ethics committee did not waive the requirement for informed consent.

To establish the probe’s effective operational bandwidth, fresh porcine abdominal tissue was procured commercially and dissected into skin, fat, and muscle tissue sections. Each specimen was precision sectioned to 40 × 40 mm² surface dimensions conforming to simulation parameters, with validated thicknesses maintained at 2 mm for the skin layer and 20 mm for both the fat and muscle layers.

Figure 3 presents the simulated and measured S 11 results of the probe positioned beneath biological tissue surfaces. During the measurement process, the operating frequency of the VNA was set to 100 MHz to 20 GHz to obtain S 11 within the frequency band. The measured return loss S 11 consistently remained below −10 dB throughout the entire operational frequency range spanning from 2.8 GHz to 15.1 GHz. Due to differences in actual biological tissues and the connection between the connector and probe compared to the simulation model, the data is different, but the probe still exhibits wideband performance.

To evaluate the feasibility of tissue discrimination via characterization, we first characterized the simulated S 11 parameter responses of skin, fat, and muscle tissues. Figure 4a revealed distinct resonance signatures across tissue types, indicating strong energy absorption and impedance matching at specific frequencies. Muscle tissue exhibited the lowest resonance frequency at 8.0 GHz, corresponding to its S 11 minimum, which is significantly lower than that of skin (8.8 GHz) and fat (10.5 GHz). This frequency progression directly correlates with dielectric properties. Muscle’s higher water content confers greater relative permittivity and downward frequency shifts, whereas fat’s lower dielectric constant yields higher frequency responses. Skin demonstrates intermediate permittivity between these extremes. These unique spectral signatures establish resonant frequency positioning as a reliable distinguishing feature for tissue differentiation.

Experimental validation of simulated resonance behavior was subsequently performed. Figure 4b confirms that measured S 11 curves exhibit comparable resonant characteristics. Primary resonance frequencies occurred at 8.0 GHz for muscle, 8.3 GHz for skin, and 10.0 GHz for fat. Discrepancies between simulated and measured values remain within an acceptable range of ± 300 MHz. Critically, the relative ordering and frequency spacing of resonance peaks maintain consistency across simulation and experimental results, thereby validating the electromagnetic model while establishing its predictive reliability for biological tissue characterization.

Figure 4c presents a comparative analysis of three distinct resonance points derived from both simulation and experimental datasets across tissue types. Results consistently demonstrate muscle exhibiting the lowest resonant frequencies, fat maintaining the highest, and skin occupying intermediate values across all measurement points. This consistent hierarchical pattern manifests in both datasets, confirming the reproducibility of tissue-specific electromagnetic responses. Precise alignment of multiple resonant peaks between simulated and experimental results establishes that these spectral features originate from intrinsic tissue dielectric properties rather than measurement artifacts. Collectively, these findings verify that biological tissues’ electromagnetic resonance behavior can be reliably characterized through both simulation and measurement, with fundamental resonances providing a robust discriminatory signature for tissue classification.

The penetration depth simulation for the probe was systematically investigated using a stratified tissue thickness reduction methodology, as presented in Figures 5a,b. Initial evaluations employed a 42 mm muscle layer, incrementally reduced in 5 mm steps to characterize S 11 variations under decreasing penetration conditions. Subsequent high-resolution analysis employed 1 mm decrements from 22 mm to 16 mm to precisely determine maximum penetration depth.

Results demonstrate that penetration depth is quantifiable through S 11 attenuation characteristics, revealing a strong depth-frequency correlation. In the 1 GHz–3 GHz range, consistently low S 11 values averaging −14 dB indicate superior electromagnetic penetration. Conversely, frequencies exceeding 4 GHz exhibit sharp S 11 increases to −5 dB, signifying enhanced signal reflection and substantially reduced penetration capacity.

Figure 5a confirms negligible penetration effects during the 42mm to 22 mm thickness reduction phase, evidenced by invariant S 11 curves. However, when the thickness decreased below 22 mm in 1 mm increments Figure 5b, discernible curve variations emerged. A critical threshold occurred at 19 mm.

Characterized by a 200 MHz downward resonance shift. Progressive reduction experiments conclusively established 19 mm as the maximum effective penetration depth.

Experimental assessment of the probe’s penetration depth characteristics is presented in Figures 5c,d, employing a two-phase protocol. Initial evaluation progressively reduced a 42 mm muscle layer in 5 mm increments to identify penetration-induced curve variations. To delineate the maximum penetration depth with high precision, a subsequent analysis was performed using 1 mm decrements across a range of 22 mm–16 mm.

Analysis of measured S 11 curves reveals frequency-dependent penetration performance. Stable S 11 below −20 dB was maintained in the 1 GHz–4 GHz range, indicating exceptional low-frequency penetration. Frequencies exceeding 5 GHz exhibited sharply increased reflection ( S 11 = −8 dB), establishing the probe’s penetration limit. A characteristic 200 MHz resonance shift occurred specifically at 19 mm depth. Further thickness reduction produced progressive leftward resonance migration.

The probe demonstrated significant performance advantages: excellent signal stability ( S 11 < -10 dB) throughout a 19 mm depth range, 7% measurement error at maximum penetration depth. Complete agreement with the multilayer tissue simulation results was observed.

Through systematic simulation and experimental validation, this study demonstrates the reliability of the probe in characterizing the dielectric properties of biological tissues within its effective penetration depth of 19 mm. The measured maximum depth shows excellent agreement with simulation results. Minor deviations attributable to unavoidable experimental interference factors fully validate the simulation model’s accuracy.

Biological safety assessment is paramount for intraoperative tissue identification applications. Electric field distribution analysis elucidates signal attenuation and transmission dynamics during probe operation. We quantified tissue exposure using Specific Absorption Rate (SAR), defined as electromagnetic power absorbed per unit tissue mass, through simulation models with 7 dBm input power at the probe’s operational frequency of 10 GHz.

As shown in Figures 6a,b, a peak SAR of 2.64 × 1 0 − 5   W/kg , significantly below IEEE C95.1-2019 and ICNIRP safety thresholds (2 W/kg averaged over 10 g tissue). SAR distribution remained confined to superficial tissue beneath the probe, exhibiting rapid decay with depth. Complementary electric field analysis confirmed localized field strength at the probe-tissue interface without substantial deep-tissue penetration. These results demonstrate excellent electromagnetic safety compliance at 7 dBm input power, satisfying clinical application requirements.

To systematically evaluate the detection sensitivity and lateral spatial resolution of the proposed probe, we constructed the tumor phantom model shown in Figure 7a. Previous studies have demonstrated that the dielectric properties of bacon closely resemble those of solid tumors and that its electromagnetic response can adequately represent real tumor tissue; bacon has therefore been widely used in tumor-related experimental studies [29]. Accordingly, bacon was selected in this work as a surrogate tumor lesion. Specifically, the bacon sample was embedded between a 2 mm skin layer and a 20 mm fat layer to mimic a superficial tumor located within subcutaneous adipose tissue, and the vertical distance between the bacon and the probe contact surface was fixed at 18 mm. As shown in Figure 7c, seven tumor phantoms with lateral dimensions ranging from 25 mm × 25 mm down to 4 mm × 4 mm and a uniform thickness of 1 mm were prepared. The proposed probe system was then used to acquire the corresponding S 11 spectra at a fixed position to determine the minimum detectable lesion size.

The experimental results were quantified by monitoring the resonance frequency shift and magnitude variation of the S 11 response. Figure 7b shows the S 11 curves of bacon phantoms with different sizes over 2.8–15.1 GHz. In the absence of a tumor, the main resonance occurred at approximately 3.5 GHz with | S 11 | ≈ 15.6 dB. When a 4 mm × 4 mm phantom was introduced, the resonance shifted to about 3.2 GHz with | S 11 | ≈ 16.3 dB. For lateral sizes of 5, 7, 10, and 15 mm, the dominant resonance remained at 3.2 GHz, while the corresponding | S 11 | minima deepened to approximately 18.6, 19.7, 19.4, and 22.0 dB, respectively. When the phantom size increased to 20 mm × 20 mm, the main resonance moved to around 3.4 GHz with | S 11 | ≈ 18.9 dB, and for the largest 25 mm × 25 mm phantom, the resonance shifted further to about 2.9 GHz with | S 11 | ≈ 23.3 dB. These data indicate a progressive leftward shift and deepening of the resonance notch as the lateral tumor size increases, reflecting a pronounced frequency- and amplitude-dependent response of the probe to lesion size.

Figure 7d further summarizes the changes in Δ | S 11 | for the seven phantoms relative to the no-tumor condition. For phantoms with intermediate lateral sizes of 5 mm × 5 mm, 7 mm × 7 mm, 10 mm × 10 mm, 15 mm × 15 mm, and 20 mm × 20 mm, Δ | S 11 | was approximately 3.0, 4.1, 3.8, 4.4, and 3.3 dB, respectively, falling within an overall range of 3.0–4.4 dB. These differences indicate that, when the lesion size reaches 5 mm × 5 mm, the | S 11 | curves for tumor and no-tumor conditions already exhibit clearly distinguishable contrast. For the largest 25 mm × 25 mm phantom, Δ | S 11 | increased to about 7.7 dB, producing a markedly stronger perturbation of the resonance. Taken together, these results demonstrate that the proposed probe is highly sensitive to changes in tumor lateral size and can reliably detect lesions with lateral dimensions of at least 5 mm.

To verify the rapid response and accuracy of the proposed probe in clinical detection, fresh breast cancer specimens were obtained from patients undergoing surgery at the Chinese PLA General Hospital, in accordance with approval by the institutional ethics committee and with written informed consent. As shown in Figure 8, the specimens were placed on a three-dimensional positioning stage, and the probe was connected to a vector network analyzer (VNA) via an RF cable. Based on the tumor-responsive frequency band established in the phantom experiments, the system was configured to scan from 100 MHz to 8 GHz with 1,001 sampling points and a transmit power of 7 dBm. The contact force between the probe and the tissue surface was precisely controlled at 5 ± 0.2 g using the integrated pressure sensor of the positioning stage, and each tissue site was measured three times to ensure repeatability under controlled conditions.

This study selected three representative regions from clinical samples (primarily consisting of adipose–glandular tissues) for measurement: the skin–adipose–glandular composite region, the adipose–glandular region, and the adipose–glandular–tumor composite region. All experiments were conducted in collaboration with clinical and pathological teams, and the results are summarized in Figure 9a. Due to the excellent broadband impedance matching of the probe and its effective coupling with the tissues, each region exhibited significantly distinct resonant characteristics in the frequency spectrum: the skin–adipose–glandular composite region showed a resonant peak at 2.18 GHz ( − 18.87 dB); the adipose–glandular region displayed a resonant response at 2.77 GHz ( − 16.67 dB); while the tumor-containing region presented a specific resonant peak at 2.95 GHz ( − 21.47 dB). Compared with the tumor-free adipose–glandular region, the resonant frequency of the tumor region shifted downward, with an amplitude difference of approximately 4.8 dB. These results indicate that tumor tissue possesses clearly identifiable electromagnetic spectral characteristics, thereby validating the effectiveness and reliability of the probe for clinical tissue classification and tumor detection.

Utilizing a 2.5 cm step resolution. The 3D control system executed raster scans across the surfaces of the clinical sample through a 4 × 4 measurement. Acquired S 11 signatures underwent quantitative comparison against tumor-characteristic reference profiles illustrated in Figure 9a. Normalized similarity indices computed via Euclidean distance metrics between interrogation points and confirmed tumor loci enabled malignancy classification, with grid positions exceeding the 0.9 similarity threshold designated as tumor-positive regions. This algorithmic processing yielded the diagnostic spatial mapping presented in Figure 9b. By comparing the probe detection results with the tumor locations annotated by professional physicians, the accuracy of the probe detection results was validated.

While the proposed broadband bio-matching probe demonstrates promising potential, several limitations merit consideration. Firstly, the current validation relies on a limited sample size. Expanded clinical trials are required to verify the probes’ rapid identification accuracy across diverse clinical samples. Secondly, although the probes exhibit favorable penetration depth and spatial resolution, evaluations have been confined to specific tissue types and geometries, without accounting for pathological alterations such as fibrosis or necrosis. Finally, the existing single-channel measurement configuration necessitates future development of multi-channel probe arrays to enable spatial mapping and imaging capabilities.