Individualized SAR calculations using computer vision-based MR segmentation and a fast electromagnetic solver.

Area of Science:

• Medical Physics
• Biomedical Engineering
• Computational Electromagnetics

Background:

• Accurate estimation of specific absorption rate (SAR) is crucial for MRI safety.
• Generic anatomical models (e.g., Duke model) may not fully capture individual patient variability.
• Reliance on generalized models can lead to inaccurate SAR predictions and potentially conservative safety margins.

Purpose of the Study:

• To develop and validate a fast, patient-specific workflow for on-line SAR supervision in MRI.
• To create individualized electromagnetic models for rapid SAR estimation.
• To reduce the discrepancy between patient anatomy and computational models.

Main Methods:

• A 3D fat-water 3T acquisition was used to create patient-specific electromagnetic models.
• Computer vision algorithms automatically segmented acquired images into key tissue classes (air, bone, fat, soft tissue).
• A fast electromagnetic integral equation solver computed individual EM field exposure and SAR matrices.

Main Results:

• The on-table workflow averaged 7 minutes and 44 seconds.
• Simplified models using essential tissue classes estimated global and local SAR with low error (6.7% and 2.7%).
• Individual volunteers exhibited significant population variability (16.0% global, 20.3% local SAR), often underestimated by the Duke model.

Conclusions:

• Patient-specific modeling for SAR estimation is computationally feasible and timely.
• Individualized SAR estimates effectively address population heterogeneity, outperforming generic models.
• This approach can enhance MRI electromagnetic safety and potentially reduce overly conservative safety margins.