New methods in quantification and RF pulse optimisation for magnetic resonance spectroscopy

Magnetic Resonance Spectroscopy (MRS) is a powerful medical diagnostic and research tool that enables us to identify metabolite concentrations in a region of interest (ROI) in-vivo. This non-invasive diagnostic technique provides a large amount of information about a certain region in the body, such as the brain or spinal cord, with no impact on patient wellbeing. MRS is readily available in many clinical units across the UK with an MRI machine and no additional hardware is needed. MRS has a number of challenges, including the requirement of a much higher level of magnetic field calibration compared to MRI, and detecting and analysing a substantially weaker signal per metabolite. To complicate the matter, there is a broad range of metabolites found in-vivo with overlapping proton spectra, obscuring signals and making spectral analysis very challenging. The primary focus of this thesis is to explore methods to aid quantification of metabolites by exploring two ends of the issue, focusing specifically on GABA, NAA, Creatine quantification, of interest to a range of neuroscience studies. Firstly, the focus is on the analysis of the acquired spectral data utilizing the MEGA-PRESS pulse sequence, specifically aimed at GABA. Comprehensively benchmarking the current state-of-the-art spectral quantification methods with experimental data from phantoms of known composition lays the foundation for devising an improved quantification technique. This novel quantification method utilises a convolutional neural network for MEGA-PRESS spectra and can outperform the state-of-the-art. Secondly, an optimisation method to find RF pulses that create specific excitations in the metabolites is devised, leading to spectra that are simpler to analyse. Such customisation of the spectra allows the removal of overlapping or obscuring features, creating chemically selective spectral acquisition methods. Moreover, the RF pulses are optimised over a range of scanner uncertainties to improve robustness. Simulations demonstrate that this approach can separate GABA, NAA and Creatine as well as Glutamine and Glutamate at 3 Tesla