Fast-RF-Shimming: Accelerate RF shimming in 7T MRI using deep learning.

Area of Science:

• Medical Imaging
• Biophysics
• Machine Learning in Healthcare

Background:

• Ultrahigh field (UHF) Magnetic Resonance Imaging (MRI) provides superior signal-to-noise ratio (SNR) and spatial resolution for clinical and research applications.
• Transmit radiofrequency (RF) field (B1+) inhomogeneities are a major challenge in UHF MRI, causing artifacts that degrade image quality.
• Traditional RF shimming methods are effective but time-consuming, hindering wider clinical adoption.

Purpose of the Study:

• To develop a novel, fast, and accurate learning-based framework for mitigating B1+ inhomogeneity in UHF MRI.
• To introduce a holistic framework, Fast-RF-Shimming, that significantly accelerates the RF shimming process.
• To evaluate the performance of the proposed method against traditional optimization techniques.

Main Methods:

• A learning-based framework, Fast-RF-Shimming, was developed, achieving a 5000x speed-up over traditional Magnitude Least Squares (MLS) optimization.
• The framework utilizes Adaptive Moment Estimation (Adam) for initial shimming weights and a Residual Network (ResNet) for direct mapping of B1+ fields to RF shimming outputs.
• A Non-uniformity Field Detector (NFD) was incorporated as an optional post-processing step to identify extreme non-uniform outcomes.

Main Results:

• Fast-RF-Shimming demonstrated significant gains in processing speed compared to standard MLS optimization.
• The method achieved notable improvements in predictive accuracy for RF shimming outputs.
• Comparative evaluations confirmed the efficacy of the proposed technique in addressing B1+ inhomogeneity.

Conclusions:

• The proposed Fast-RF-Shimming framework offers a promising and efficient solution for mitigating B1+ inhomogeneity in UHF MRI.
• This technique has the potential to enhance image quality and facilitate broader clinical adoption of UHF MRI.
• The combination of deep learning and optional post-processing provides a robust approach to improving MRI diagnostics.