Uncertainty Quantification in SAR Induced by Ultra-High-Field MRI RF Coil via High-Dimensional Model Representation.

Area of Science:

• Medical Imaging Physics
• Computational Electromagnetics
• Biomedical Engineering

Background:

• Increasing magnetic field strength in MRI (e.g., 7 T) challenges Specific Absorption Rate (SAR) safety limits.
• Uncertainty in human head tissue dielectric properties complicates SAR calculations in ultra-high-field MRI.
• Standing wave formation at higher frequencies exacerbates SAR concerns.

Purpose of the Study:

• To develop and validate a computational framework for quantifying the impact of dielectric property uncertainties on induced SAR in UHF-MRI.
• To assess the safety implications of these uncertainties for human head tissues during MRI scans.
• To compare the efficiency and accuracy of the proposed framework against traditional and machine learning-based methods.

Main Methods:

• A surrogate model-assisted Monte Carlo (MC) technique was employed.
• High-dimensional model representation (HDMR) with generalized polynomial chaos expansions was used to build surrogate models.
• The framework efficiently computed SAR statistics by approximating MRI observables (electric fields, SAR).

Main Results:

• The proposed framework achieved high accuracy in SAR statistics with an average relative error of 0.28% for local SAR.
• It required significantly fewer simulations (289) compared to traditional MC and ML-based surrogate methods.
• Results indicated potential SAR value fluctuations of up to 30% in specific head regions due to dielectric property uncertainties.

Conclusions:

• The developed computational framework is highly efficient and accurate for assessing SAR uncertainties in UHF-MRI.
• Considering dielectric property variations is crucial for ensuring patient safety in 7 T MRI systems.
• The findings underscore the need for robust safety assessments that account for biological variability.