Learning based quantification and removal of lipid signals from MR spectroscopic imaging of the brain

Magnetic resonance spectroscopic imaging (MRSI) has long promised to be a powerful noninvasive molecular imaging technique for metabolites and nuero-transmitters. Recently proposed fast J-resolved MRSI has further enabled the separation of J-coupled overlapping molecules, elevating the molecular MRSI to a new level. However, the overwhelming nuisance signal such as the lipid signal has made it extremely difficult to separate the desired molecular signal from the nuisance signal which is usually orders of magnitude higher than the desired molecular signal. Therefore, it is necessary to remove the nuisance signal. The nuisance removal problem is especially more challenging in the case of J-resolved MRSI due to its additional encoding dimension thus limited spatial resolution. To solve this problem, reference-based approach is necessary to help lipid removal since estimating the lipid signal from the data itself is an ill-conditioned problem. However, reference-based approach usually requires precise quantification of the reference signal, which is rather challenging due to the complicated structure of the lipid signal. In this work, we present a novel lipid quantification method to improve quantification performance by utilizing refined prior information, a restrictive lipid signal model, and machine learning. The proposed method not only enables incorporation of tissue dependent prior, but are also accelerated with the help of machine learning, thus achieves improved quantification performance with good robustness. With the improved performance of lipid quantification of the reference signal, the reference-based method can better help lipid removal from the J-resolved MRSI data. Eventually, a higher quality molecular map can be obtained. The propose method has been evaluated on both simulation data and in-vivo data, producing encouraging results. The proposed method could further provide a potential framework for spectral quantification of MRSI data.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Zhang, Yuanbo

Contributors dc:contributor

• Liang, Zhi-Pei

• Magnetic Resonance Spectroscopic Imaging (mrsi)
• Spectral Quantification
• Lipid Removal
• Bayesian Estimation
• Machine Learning

Statement dc:rights

• Copyright 2023 Yuanbo Zhang

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/122267

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F122267