Graduation Semester and Year
Fall 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Biomedical Engineering
Department
Bioengineering
First Advisor
Dr. Anke Henning, Ph.D.
Second Advisor
Dr. Jae Mo Park, Ph.D.
Third Advisor
Dr. Ananth Madhuranthakam, Ph.D.
Fourth Advisor
Dr. Christos Papadelis, Ph.D.
Fifth Advisor
Dr. George Alexandrakis, Ph.D.
Abstract
High and ultra-high field MRI and MRSI offer markedly improved signal-to-noise ratio and spectral dispersion, enabling in-vivo characterization of tumor metabolism with unprecedented detail. However, these benefits are tightly coupled to high demands on static magnetic field (B0) homogeneity, particularly in the brain where susceptibility interfaces near the skull base and paranasal sinuses generate complex, higher-order field perturbations. In glioma patients, additional susceptibility variations arise from surgical cavities, hemorrhage, calcifications, and cystic components, further degrading B₀ homogeneity. As a result, shimming often become the main bottleneck limiting robust, whole-brain spectroscopic imaging and, consequently, our ability to map metabolic and macromolecular changes across heterogeneous tumor habitats.
This work has two overarching aims. First, to systematically characterize and improve B0 shimming at 7T by combining detailed shim hardware calibration with a vendor-independent, image-based shimming framework that explicitly accounts for higher-order fields and cross-terms. Second, to apply the resulting shimming pipeline to fast FID-MRSI in patients with low- and high-grade gliomas, enabling quantitative mapping of both metabolite and macromolecular signals across lesions and contralateral brain, and to evaluate how improved shimming translates into better spectral quality and more sensitive metabolic readouts.
To address the first aim, the static shim fields of two 7T systems were experimentally calibrated using 3D multi-echo GRE B0 mapping in both a head-and-shoulder phantom and in vivo. Vendor 1 provides a full set of third-order spherical harmonic (SH) shim channels, whereas Vendor 2 supports only a partial third order set in addition to second-order shims. For each system, we measured the full shim calibration matrix relating shim currents to B0 changes and decomposed the resulting fields into an SH basis. This allowed us to quantify shim purity (how closely each channel approximates an ideal SH term) and to reveal cross-term interactions that are not captured by vendor-supplied coefficients. Based on these measurements, we implemented a vendor-independent, image-based shimming tool that uses the empirically measured shim fields as its basis, rather than assuming ideal SH shapes. The tool ingests whole-brain B₀ maps, allows flexible ROI definition (single voxel, MRSI slab, whole brain, or multi-region), and performs regularized least-squares optimization to compute optimal shim currents up to third order for each system. The framework supports both static and slice-wise solutions and can be driven by either phantom or in-vivo calibration datasets.
The B0 shimtool was applied to study in glioma application, patients with histologically confirmed low-grade (WHO II/III) and high-grade (WHO IV) gliomas who underwent comprehensive 7T MRI including structural imaging (e.g., MP2RAGE and FLAIR for anatomical localization and lesion segmentation), B₀ mapping, and 2D FID-MRSI. Spectroscopic imaging was performed with ultra-short TE (TE ≈ 1.21 ms), short TR (≈ 270–320 ms), a 50×50 in-plane matrix over a ~220×220 mm² FOV, and a 12 mm slice thickness, yielding a nominal resolution of ~4.4×4.4×12 mm³. B0 shimming for MRSI was performed using the image-based shim tool, with ROIs covering the full MRSI slab a explicitly including tumor, peritumoral tissue, and contralateral brain. For comparison, vendor-standard shimming (typically limited to global or low-order SH fits and rectangular shim voumes) was also evaluated in a subset of cases.
Spectral analysis was performed using a two-step LCModel-based pipeline. In the first step, metabolite-only basis sets were used to quantify key metabolites including total N-acetylaspartate (tNAA), total creatine (tCr), myo-inositol (mI), glycine (Gly), glutamate (Glu), glutamine (Gln), glutathione (GSH), and choline-containing compounds. In the second step, the
residual spectra were refitted with a dedicated macromolecular (MM) basis set to obtain spatial maps of individual MM components (e.g., MM09, MM12, MM14, MM17, MM20, MM22, MM26, MM27, MM30, MM32, MM36, MM37, MM38, MM39). Voxel-wise quality control was performed using Cramér-Rao lower bounds, SNR, and linewidth. Region-of-interest (ROI) analyses were carried out in tumor core, peritumoral/edematous tissue, and contralateral normal-appearing white matter, using segmentations derived from anatomical imaging and, where available, clinical radiologic reports. Group comparisons were made between low- and high-grade tumors, and correlation analyses were performed between metabolic/macromolecular amplitudes and tumor grade-related features.
Shim calibration demonstrated that nominal higher-order channels often exhibit substantial admixtures of lower- and other higher-order SH terms, particularly in the third-order set, and that these cross-terms differ substantially between Vendor 1 and Vendor 2 hardware implementations. By explicitly using the empirically measured fields instead of idealized SH shapes, the vendor-independent shim tool reduced residual B0 inhomogeneity relative to vendor-standard shimming, particularly in frontal and temporal lobes where higher-order terms are critical. For MRSI ROIs, the optimized shim solutions consistently yielded narrower spectral linewidths and more spatially uniform B0 across the slab. This translated into a larger fraction of voxels meeting pre-defined quality criteria (e.g., linewidth below a clinically acceptable threshold and low CRLBs) compared to vendor shimming alone. In practical terms, more voxels in and around the tumor could be included in the metabolic analysis, particularly near air–tissue interfaces or surgical cavities where conventional shimming typically fails.
In the glioma cohort, improved B0 shimming enabled robust mapping of both metabolite and macromolecular signals across heterogeneous tumor regions. High-grade gliomas showed the expected pattern of decreased tNAA and elevated choline-containing compounds, consistent with neuronal loss and increased membrane turnover, respectively. In addition, high-grade tumors exhibited marked increases in Gly and mI, which may reflect enhanced proliferation, altered osmoregulation and microglial activation, and increased glial content. The macromolecular analysis revealed spatially varying increases in specific MM components within the enhancing tumor core and, in some cases, extending into the peritumoral region. These MM alterations are consistent with reports of upregulated branched-chain amino acid and protein/peptide metabolism in aggressive gliomas and may provide complementary information beyond conventional metabolite ratios.
Low-grade gliomas showed more heterogeneous metabolic profiles. In many cases, tNAA depletion was less severe, and choline elevation was moderate compared to high-grade lesions, with largly preserved Glu and Gln levels in surrounding cortex. Macromolecular changes were also present but generally less pronounced. Across subjects, ROI-based comparisons indicated (and in several cases statistically significant differences) between high- and low-grade tumors for combinations of metabolite ratios (e.g., tCho/tNAA, Gly/tCr, mI/tCr) and selected MM amplitudes. Importantly, these contrasts remained detectable even when the analysis was restricted to voxels that passed conservative quality thresholds, highlighting that the B0 shimming improvements did not merely increase voxel count but improved the reliability of tumor-specific spectral information.
Taken together, the results demonstrate that (i) careful experimental calibration of shim hardware and (ii) vendor-independent, image-based B₀ shimming can substantially improve field homogeneity at 7T, especially when higher-order terms are available and fully characterized. When combined with ultra-short-TE FID-MRSI and a two-step metabolite/macromolecule fitting strategy, these shimming advances enable robust, high-resolution metabolic mapping of gliomas, including spatially resolved macromolecular signatures that are often inaccessible at lower fields or with conventional shim workflows. The integrated framework developed here lays the groundwork for more standardized and transferable B0 shimming across platforms and for incorporating detailed metabolic and macromolecular imaging into clinical and translational studies of brain tumors. In future work, the same toolkit can be extended to dynamic shimming, multi-slab coverage, and multi-time-point studies to further improve longitudinal monitoring of treatment response and progression in glioma patients.
Keywords
Ultrahigh field, MR Spectroscopic Imaging (MRSI), Brain Tumor, Metabolic Imaging of Glioma, B0 shimming, Brain Tumor Segmentation, Macromolecule, Amnio Acid Synthesis, WarBurg effect
Disciplines
Biological Engineering | Systems and Integrative Engineering
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Jani, Mahrshi, "Vendor-Independent B0 Shimming Framework with application to metabolic MRI in Human Gliomas" (2025). Bioengineering Dissertations. 209.
https://mavmatrix.uta.edu/bioengineering_dissertations/209