A Deep Learning Based Prediction of Specific Absorption Rate Hot-Spots Induced by Broadband Electromagnetic Devices

Abstract

The rapid growth of wearable electromagnetic devices has raised concerns about the potential health effects of electromagnetic fields, particularly due to their interaction with biological tissues. The key parameter for assessing these effects is the specific absorption rate (SAR), which serves as the standard for evaluating energy absorption and associated thermal effects on the human body. However, traditional numerical methods for SAR estimation are computationally expensive, limiting their application to real-time scenarios. This study addresses this limitation by using a deep learning approach to predict the positions of SAR hotspots efficiently and accurately. A convolutional neural network model was developed to predict hotspot locations with minimal computational effort, using tissue distribution and operating frequencies. The dataset includes tissue images augmented with physical properties such as density and permittivity, the latter being frequency dependent, to enhance the model precision. The proposed method demonstrates robust performance of data-driven approaches in predicting SAR hotspots in real time, providing a foundation for safer and more effective deployment of electromagnetic devices, including wearable and medical applications. The source code used in this study is openly available at https://github.com/ShayanDodge/DL-SAR-Hotspots.