Gulsah Yildiz

This study addresses the computational challenges of estimating two-dimensional electric field (EF) distributions within heterogeneous breast tissue models, a critical task in medical applications such as microwave imaging and hyperthermia. Traditional numerical simulation methods are accurate but computationally expensive, often requiring minutes of processing time. To overcome this limitation, we propose a deep learning approach that predicts EF distributions directly from dielectric property maps, specifically electrical conductivity and permittivity, significantly reducing inference time to a matter of seconds. Two convolutional neural network architectures are evaluated: a U-Net model with a ResNeXt50 encoder (R50-UNet) and a standard ResNeXt50 model (R50). Additionally, we introduce a masking-based loss function that emphasizes learning in regions of highest relevance within the domain. Quantitative evaluation demonstrates that the R50-UNet model outperforms the standard R50 model, achieving up to a 14.66 dB improvement in signal-to-noise ratio (SNR). The application of the masking method further enhances performance, with an additional gain of up to 3.24 dB in SNR. Data efficiency analysis reveals that while the R50 model reaches performance saturation with only 25% of the available training data, the R50-UNet architecture combined with masked loss continues to improve as more data are utilized. These findings support the feasibility of using deep learning for fast and accurate EF prediction in biomedical scenarios where computation time and spatial precision are critical.

Salinity is a key indicator in environmental, agricultural, and marine systems, requiring accurate, real-time monitoring to support sustainable resource management. While traditional measurement methods offer many advantages, they are impractical due to preparation requirements, contact-based operation, and sensitivity to environmental factors. In contrast, microwave sensing has emerged as a promising alternative for salinity detection,providing contactless operation, robustness, and sensitivity across wide frequency bands. In this study, we present a non-contact microwave sensing system integrated with a deep neural network (DNN) model for high-precision salinity prediction in water samples. The system, based on two Vivaldi antennas and PA6 plates matching layer, collects reflection and transmission coefficients (RCs and TCs) from water samples with salinity ranging from 0 to 30,000 ppm in increments of 1,000 ppm. A dataset of 620 measurements was acquired at 751 frequency points (6759 features) and was used to train and test the DNN model. Furthermore, four different feature selection methods, T-test, Lasso-L1, mutual information(MI), and correlation analysis (CA), were employed to improve predictive accuracy. When the CA method was applied to the full frequency band (0.5–8 GHz), the Mean Absolute Error (MAE) decreased from 299.91 ppm to 119.57 ppm, while the coefficient of determination (R²) increased from 0.998632 to 0.999674, as the feature size was reduced by half. In addition, multiple narrow frequency bands were analysed to identify optimal bands, and the CA method selected 680 features in the 0.5-2 GHz range, resulting in an MAE of 106.27 ppm and an R² of 0.999702. In addition, when the MI method was applied only to the TC, the best performance was achieved in the 0.5–2.5 GHz band, yielding an MAE of 126.37 ppm and an R²: 0.999695, while utilizing only 302 features.

Breast cancer detection and treatment have advanced significantly with imaging technologies, but challenges remain in distinguishing the type and stage of tumors. Microwave imaging (MWI) offers a promising alternative due to its non-ionizing nature and its ability to exploit dielectric property (DP) contrast. This study investigates the effectiveness of MWI in detecting and characterizing tumors using a phantom for breast tissue and tumor-mimicking NaCl solutions with various DPs (0.1 M, 0.2 M, 0.4 M and 0.8 M). First, the Cole-Cole parameters of these materials were calculated using DP measurements obtained from the open-ended coaxial probe method in order to provide broadband frequency analysis. Furthermore, the developed MWI system was utilized to evaluate tumor detectability and differentiation based on these DP changes. The MWI experiment was performed with 12 Vivaldi antennas between 0.6-2.6 GHz and the results were analyzed for two different positions. The results indicate that MWI system can effectively distinguish tumors with different DPs from each other using quantitative differential imaging due to its sensitivity to variations. To this end, the inverse time migration (RTM) method was employed to compare reference-target pairs (RTP) to generate an image of a tissue-mimicking phantom with tumors. The results show high correlation between RTP image contrast and the target-reference DP difference.

Determining the electric field distribution using electromagnetic solvers requires significant computational effort, a high time consumption, and operator training. In this paper, a deep learning (DL)-based method is specifically developed to reconstruct the electric field distribution generated by antennas, based on the dielectric properties of the breast. ResNet and UNet were employed to predict the electric field within complex breast structures. Nine distinct breast dielectric models were utilized to create the initial dataset. A novel data augmentation technique was applied and the size was increased by three to address the challenge of limited dataset size. The electric field distribution data were generated using commercially available finite element method-based electromagnetic simulations. DL architectures were customized to fit the study and trained over 4374 breast slices. ResNet demonstrated an average Signal-to-Noise Ratio (SNR) of 23.7±6.7 dB and an average Normalized Mean Squared Error (NMSE) of 0.0029±0.0039 across 504 test breast slices, while UNet averaged 24.55±6.64 dB and 0.0024±0.0031 respectively. Additionally, a masked loss calculation was incorporated to enhance the learning accuracy of the implemented models. This approach yielded an increase of 0.9 dB in average SNR for ResNet and a 0.8 dB SNR increase for the UNet architecture on average. These results indicate that the proposed DL models effectively reconstruct electric field distributions with substantial accuracy, highlighting its potential for practical applications in real-time microwave medical imaging and treatment planning. Further refinement and validation with a more diverse set of breast models could enhance the model's applicability and performance.

In microwave hyperthermia tumor therapy, electromagnetic waves focus energy on the tumor to elevate the temperature above its normal levels with minimal injury to the surrounding healthy tissue. Microwave hyperthermia applicator design is important for the effectiveness of the therapy and the feasibility of real-time application. In this study, the potential of using fractal octagonal ring antenna elements as a dipole antenna array and as a connected array at 2.45 GHz for breast tumor hyperthermia application was investigated. Microwave hyperthermia treatment models consisting of different fractal octagonal ring antenna array designs and a breast phantom are simulated in COMSOL Multiphysics to obtain the field distributions. The antenna excitation phases and magnitudes are optimized using the global particle swarm algorithm to selectively increase the specific absorption rate at the target region while minimizing hot spots in other regions within the breast. Specific absorption rate distributions, obtained inside the phantom, are analyzed for each proposed microwave hyperthermia applicator design. The dipole fractal octagonal ring antenna arrays are comparatively assessed for three different designs: circular, linear, and Cross—array. The 16-antenna dipole array performance was superior for all three 1-layer applicator designs, and no distinct difference was found between 16-antenna circular, linear, or cross arrays. Two-layer dipole arrays have better performance in the deep-tissue targets than one-layer arrays. The performance of the connected array with a higher number of layers exceeds the performance of the dipole arrays in the superficial regions, while they are comparable for deep regions of the breast. The 1-layer 12-antenna circular FORA dipole array feasibility as a microwave hyperthermia applicator was experimentally shown.

Microwave hyperthermia (MH) requires the effective calibration of antenna excitations for the selective focusing of the microwave energy on the target region, with a nominal effect on the surrounding tissue. To this end, many different antenna calibration methods, such as optimization techniques and look-up tables, have been proposed in the literature. These optimization procedures, however, do not consider the whole nature of the electric field, which is a complex vector field; instead, it is simplified to a real and scalar field component. Furthermore, most of the approaches in the literature are system-specific, limiting the applicability of the proposed methods to specific configurations. In this paper, we propose an antenna excitation optimization scheme applicable to a variety of configurations and present the results of a convolutional neural network (CNN)-based approach for two different configurations. The data set for CNN training is collected by superposing the information obtained from individual antenna elements. The results of the CNN models outperform the look-up table results. The proposed approach is promising, as the phase-only optimization and phase–power-combined optimization show a 27% and 4% lower hotspot-to-target energy ratio, respectively, than the look-up table results for the linear MH applicator. The proposed deep-learning-based optimization technique can be utilized as a protocol to be applied on any MH applicator for the optimization of the antenna excitations, as well as for a comparison of MH applicators.

Microwave hyperthermia (MH) requires the selective focusing of microwave energy on the targeted region while minimally affecting the healthy tissue. Emerging from the simple nature of the linear antenna arrays, this work demonstrates focusing maps as an application guide for MH focusing by adjusting the antenna phase values. The focusing of the heating potential (HP) on different density breast models is performed via the proposed method using Vivaldi antennas. The effect of the tumor conductivity on the focusing is discussed. As a straightforward approach and utilizing the Vivaldi antennas, the system can be further combined with MH monitoring application.

Purpose Convection–reaction equation‐based magnetic resonance electrical properties tomography (cr‐MREPT) provides conductivity images that are boundary artifact‐free and robust against noise. However, these images suffer from the low convective field (LCF) artifact. We propose to use dielectric pads to alter the transmit magnetic field (B₁⁺), shift the LCF region, and eliminate the LCF artifact. Methods Computer simulations were conducted to analyze the effects of pad electrical properties, pad thickness, pad height, arc angle, and thickness of the pad–object gap. In 3T MR experiments, water pads and BaTiO₃ pads were used with agar–saline phantoms. Two data sets (e.g., with the pad located on the left or on the right of the object [phantom]) were acquired, and the corresponding linear systems were simultaneously solved to get LCF artifact‐free conductivity images. Results A pad needed to have 180° arc angle and the same height with the phantom for maximum benefit. Increasing the pad thickness and/or the relative permittivity of the pad increased the LCF shift, whereas excessive amounts of these parameters caused errors in conductivity reconstructions because the effect of neglected B_z terms became noticeable. Conductivity of the pad, on the other hand, had minimal effect on elimination of the LCF artifact. Combining 2 data sets (i.e., with 2 different dielectric pad positions) resulted in more accurate conductivity maps (low L²‐errors) as opposed to no pad or single pad cases in experiments and simulations. Conclusions Using the proposed technique, LCF artifact is significantly removed, and the reconstructed conductivity values are improved.