Fast-RF-Shimming: Accelerate RF Shimming in 7T MRI using Deep Learning

TL;DR

This work tackles $B_{1}^{+}$ inhomogeneity in 7T MRI by replacing slow MLS shimming with a learning-based pipeline called Fast-RF-Shimming. It combines a random-initialized Adam step to produce reference shimming weights, a ResNet that maps $B_{1}^{+}$ fields directly to RF shim outputs, and an optional Non-uniformity Field Detector (NFD) to identify artifacts. Compared with MLS and a similar deep-learning approach, the method delivers a ~5000× speed-up, and achieves lower RMSE (≈$9.04 ext{ extcent}$ of Target FA) with robust performance across folds, while NFD provides near-perfect classification of uniform vs non-uniform outputs. The approach enables real-time, large-scale RF shimming and shows promise for extending to other field strengths, with future work focusing on real-data validation, SAR constraints, and motion robustness.

Abstract

Ultrahigh field (UHF) Magnetic Resonance Imaging (MRI) offers an elevated signal-to-noise ratio (SNR), enabling exceptionally high spatial resolution that benefits both clinical diagnostics and advanced research. However, the jump to higher fields introduces complications, particularly transmit radiofrequency (RF) field ($B_{1}^{+}$) inhomogeneities, manifesting as uneven flip angles and image intensity irregularities. These artifacts can degrade image quality and impede broader clinical adoption. Traditional RF shimming methods, such as Magnitude Least Squares (MLS) optimization, effectively mitigate $B_{1}^{+}$ inhomogeneity, but remain time-consuming. Recent machine learning approaches, including RF Shim Prediction by Iteratively Projected Ridge Regression and other deep learning architectures, suggest alternative pathways. Although these approaches show promise, challenges such as extensive training periods, limited network complexity, and practical data requirements persist. In this paper, we introduce a holistic learning-based framework called Fast-RF-Shimming, which achieves a 5000x speed-up compared to the traditional MLS method. In the initial phase, we employ random-initialized Adaptive Moment Estimation (Adam) to derive the desired reference shimming weights from multi-channel $B_{1}^{+}$ fields. Next, we train a Residual Network (ResNet) to map $B_{1}^{+}$ fields directly to the ultimate RF shimming outputs, incorporating the confidence parameter into its loss function. Finally, we design Non-uniformity Field Detector (NFD), an optional post-processing step, to ensure the extreme non-uniform outcomes are identified. Comparative evaluations with standard MLS optimization underscore notable gains in both processing speed and predictive accuracy, which indicates that our technique shows a promising solution for addressing persistent inhomogeneity challenges.

Figures (4)

• Figure 1: The $B_{1}^{+}$ field inhomogeneity problem addressed in RF shimming design and main advantages of our proposed strategy over conventional MLS method. The block to the left presents the 8-coil $B_{1}^{+}$ data we started with. The two central blocks show our method and conventional MLS method with measured runtime over one subject based on MPRAGE Mugler1990 which is estimated from testing prediction. On the right are two optimization results on the same slice picked from testing cases, which shows a contrast of uniformity over the $B_{1}^{+}$ field between the two methods.
• Figure 2: The procedures of training and prediction of our proposed optimization strategy. The strategy starts with simulations to get the desired $B_{1}^{+}$ as inputs of the Residual Network (ResNet18). Data augmentation is then applied. Adaptive Moment Estimation 17Kingma2014 is used to calculate the reference weights of coils as targets in training. Next, predictions are made by applying the testing data into the trained DL model. Finally, an optional post-processing step with the NFD is designed to classify non-uniform and uniform output fields.
• Figure 3: The RMSE results [% of target FA] of the upper bound, MLS method, our method and uCNN are illustrated using box plots and violin plots. Five folds of testing results are drawn independently in five groups in the figure. Illustrated by $*$, a significant difference with $p < 0.05$ is ensured.
• Figure 4: (Left) The top section illustrates the classification performance of the NFD for non-uniform and uniform cases, while the bottom displays examples of the corresponding output fields. The table shows the mean confidence scores and accuracy percentages for each group. (Right) The confusion matrix visualizes the frequency of predictions for non-uniform and uniform cases.