Advancing Ultra-High Field MRSI: Faster, Smarter, and Broader Metabolic Imaging

Abstract

Magnetic Resonance Spectroscopic Imaging (MRSI) is a non-invasive imaging method capable of mapping tissue metabolism with spatial specificity. However, its clinical adoption has been hindered by technical barriers such as long acquisition times, lipid contamination, low signal-to-noise ratio (SNR), and limited spectral bandwidth, particularly at ultra-high field strengths like 7T. This thesis addresses these challenges through fast acquisition sequences, advanced reconstruction algorithms, and application-specific strategies, enabling high-resolution, clinically viable MRSI across various anatomical regions. Chapter 2 introduces a two-dimensional FID-EPSI sequence for ¹H MRSI in the brain, featuring an external crusher coil and ℓ²-regularized reconstruction to suppress extracranial lipid signals without increasing SAR. Phase correction methods were also implemented, improving spectral fidelity and enabling robust metabolite quantification. Chapter 3 presents a silent EPSI sequence using an ultrasonic gradient insert coil to reduce acoustic noise by ~19 dB while maintaining a wide spectral bandwidth (20.34 kHz). This approach enhances patient comfort and scan tolerability and achieves a 4.5-fold reduction in scan time—key for clinical feasibility. Chapter 4 extends MRSI to the human tongue at 7T using a custom-built RF coil and a Hamming-weighted k-space strategy. This allowed for the first spatially resolved in vivo ³¹P MRSI of the tongue within a 10-minute scan at 15 mm isotropic resolution. The method enabled detection of key phosphorous metabolites, demonstrating potential for oral cancer assessment. Chapter 5 shifts focus to ²H MRSI in the liver. After oral ingestion of deuterated glucose, hepatic glucose uptake was tracked using a ²H EPSI sequence. Hamming-weighted acquisition improved SNR by 2.4-fold compared to uniform sampling, facilitating clear mapping of deuterated water, glucose, and lipid signals in vivo. Chapter 6 explores advanced low-rank subspace reconstruction for 3D ²H MRSI, inspired by the SPICE framework. Central Hamming-weighted k-space enabled subspace estimation directly from the data, eliminating the need for separate training scans. Emerging quantum-simulated spectral basis methods were also discussed as a way to further reduce acquisition time. Chapter 7 integrates these findings into a practical implementation guide for MRSI on a clinical 7T Philips platform. It covers protocol development, reconstruction workflows, and strategies for extending EPSI to ²H and ³¹P, making the transition from research to clinical use more accessible. Overall, this thesis introduces innovations such as non-RF lipid suppression, silent EPSI, Hamming-weighted sampling, and low-rank subspace reconstruction. These advancements collectively improve acquisition speed, resolution, and SNR, and were validated across brain, tongue, and liver, translating MRSI innovations into clinical practice.