Deep Learning-Based Local SAR Assessment for Body Imaging at 3T

Motivation: Developing methods for local SAR estimation in patients undergoing clinical MRI could significantly enhance RF safety assessments. Goal(s): To develop and evaluate the performance of a deep learning-based method for real-time local SAR assessment in clinical 3T MRI scans. Approach: We trained a convolutional neural network on male and female trunk models, including arms, to map the relation between subject-specific complex B1+-maps and the corresponding local SAR distribution at 3T. Results: The proposed approach demonstrates the potential for reliable assessment of local SAR distribution at 3T. However, considerable underestimations of peak SAR values were observed in some cases. Impact: Recent simulation studies suggest clinical MRI scans may cause greater local tissue heating than previously anticipated. Measurement-based methodologies for local SAR estimation in clinical MRI systems could contribute to safety assessments and may facilitate agreement between thermal simulations and measurements.