Deep Learning Based Ultrafast and Robust RF Shimming Using Quadrature B1+ Maps

Motivation: To address B1+ inhomogeneities at ultrahigh fields, RF shimming is commonly used and requires a lengthy per-channel B1+ calibration scan, making it sensitive to motion and potentially affecting image quality. Goal(s): We aim to develop a faster RF shimming method that predicts RF shims directly from quadrature B1+ maps, bypassing the need for per-channel B1+ maps. Approach: In-vivo B1+ maps were acquired and used to train a deep neural network, with its performance evaluated in vivo. Results: Preliminary findings demonstrate the feasibility of deep learning based RF shimming with performance similar to standard RF shimming. Impact: The deep learning based RF shimming may offer a faster workflow for ultrahigh field imaging, greatly reducing B1+ calibration scan time with less sensitivity to motion and more consistent improvements in image quality.