Speaker
Description
This study compares segmentation-based and segmentation-free deep learning models for assessing localized radiofrequency (RF) exposure in human head models. Unlike conventional segmentation-based models, the segmentation-free model estimates tissue dielectric and thermal properties directly from MRI using deep learning, ensuring smooth transitions between different tissues and capturing the transition of properties within intra-tissue. Finite-difference time-domain method and bioheat transfer equation were used to compute power absorption and temperature rise above 6 GHz. The segmentation-free model showed good consistency with the segmentation-based model in terms of power absorption and temperature rise. Moreover, the segmentation-free model exhibited reduced inter-subject variability. These findings highlight the potential of deep learning-based segmentation-free models in improving RF dosimetry accuracy and reducing computational uncertainty above 6 GHz.
