Evaluating Rigid Motion-Affected, U-NET-Estimated Q-Matrices with Parallel RF Transmission

Motivation: Patient motion may exacerbate SAR exposure, and may necessitate large corrective safety factors. A position adaptive safety model would facilitate high-performance scanning without compromising safety. Goal(s): We test the efficiency of using neural networks for estimating the effect of patient motion on local SAR for ultrahigh-field MRI. Approach: We trained U-Nets to estimate the effect of patient motion on Q-matrices, and compared network-estimated SAR with ground-truth after-motion SAR for realistic parallel-transmit pulses. Results: Patient motion has a statistically-significant effect on local SAR, but network-estimated safety models can recover a faithful representation of the ground-truth after-motion local SAR. Impact: The proposed approach needs a smaller corrective safety factor, which may enable higher-performance scanning without compromising safety, when using ultrahigh-field MRI for subjects who may not remain still.