Automated Human Head Model Construction and Fast RF Shimming Design in 7T MRI using Deep Learning

Abstract

The design of RF coils for MRI depends on customized head models with extensive tissue labels, significantly enhancing electromagnetic (EM) simulations for RF coil validation and optimization. Open-source segmentation tools have generated head models from MRI scans, but the most advanced automatically segmented brain model for EM simulations lacks muscle, fat, skin, and detailed brain labels, limited by the tissue types detectable in T1-weighted (T1w) images. This study presents an advanced head segmentation suite with 14 distinct tissue labels. Using state-of-the-art segmentation on paired T1w MRI and CT scans, our models demonstrate improved field uniformity at 7 Tesla. Moreover, Ultrahigh field (UHF) Magnetic Resonance Imaging (MRI) offers an elevated signal-to-noise ratio (SNR), which benefits clinical diagnostics and advanced research. However, the jump to higher fields introduces complications, particularly transmit radiofrequency (RF) field (B1+) 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 B1+ inhomogeneity but remain time-consuming and typically require the patient’s presence to compute solutions. Although these methods show promise, challenges such as extensive training periods, limited network complexity, and practical data requirements persist. Here, we introduce a holistic learning-based framework called Fast-RF-Shimming, which achieves a 5000× speed-up in RF shimming compared to traditional MLS. It first uses random-initialized Adaptive Moment Estimation (Adam) to derive reference shimming weights from multi-channel B1+ fields, then employs a Residual Network (ResNet) to map these fields to ultimate shimming outputs, incorporating a confidence parameter into its loss function. A Non-uniformity Field Detector (NFD) optionally identifies extreme non-uniform outcomes. Comparative evaluations with standard MLS underscore notable gains in both processing speed and predictive accuracy. As a result, this technique presents a faster, more efficient RF shimming framework for UHF MRI, offering a promising solution for persistent inhomogeneity challenges.