Prediction of Maximum Temperature Rise on Skin Surface for Local Exposure at 10–90 GHz

There is a chronic lack of standardized data sets that would enable seamless development and testing the performance of novel computational techniques and thus promote the transparency and reproducibility of computational dosimetry research. In this paper, the limited available data on the incident power density (IPD) and resultant maximum temperature rise for different local exposure scenarios at 10–90 GHz have been statistically modeled to capture the marginal distribution of each variable and their interdependence. The synthetic data are then sampled from the fitted model with respect to predetermined dosimetric constraints. We provide free access to this comprehensive synthetic data set, compiled of the high-fidelity numerical data, the collection of which would traditionally require running computationally demanding simulations within often unattainable commercial electromagnetic (EM) software. As a demonstration practice, surrogate models for predicting maximum temperature rise on skin surface are fitted using the synthetic data, whereas their predictive performance is obtained on the originally available data. The combination of simple polynomial/tensor-product spline surrogates achieves the lowest mean absolute error (MAE) of $0.058^{\circ}\mathrm{C}$, demonstrating an accurate approximation, but also a significant reduction of computational resources.