Power absorption and temperature rise in deep learning based head models for local radiofrequency exposures

Area of Science:

• Computational electromagnetics
• Medical imaging
• Biophysics

Background:

• Accurate modeling of radiofrequency (RF) exposure in humans is crucial for safety.
• Challenges exist in precisely defining head tissue properties for computational models.
• Variability in power absorption and temperature rise impacts human protection assessments.

Purpose of the Study:

• To compare segmentation-based and segmentation-free models for RF exposure assessment.
• To evaluate the impact of deep learning on head modeling for dosimetry.
• To analyze inter-subject variability and dosimetric uncertainties across frequencies.

Main Methods:

• Developed two computational head models: one segmentation-based, one segmentation-free using deep learning.
• Estimated tissue dielectric and thermal properties directly from MRI using deep learning.
• Solved finite-difference time-domain and bioheat transfer equations for temperature rise.
• Analyzed inter-subject variability and dosimetric uncertainties at various frequencies.

Main Results:

• Both models showed strong consistency, with peak temperature rise differences of 7.6 ± 6.4%.
• Segmentation-free models demonstrated reduced inter-subject variability, especially at higher frequencies.
• Maximum relative standard deviation in heating factor variability was 15.0% at 3 GHz, decreasing with frequency.

Conclusions:

• Segmentation-free deep learning models offer advantages in RF dosimetry, reducing variability and enhancing efficiency.
• These models provide a promising avenue for refining individual-specific RF exposure assessments.
• Findings support improved accuracy and consistency in human protection guidelines against RF exposure.