synMARSS-An End-To-End Platform for the Parametric Generation of Synthetic In Vivo Magnetic Resonance Spectra

Area of Science:

• Magnetic Resonance Spectroscopy (MRS)
• Computational Chemistry
• Machine Learning in Science

Background:

• Synthetic magnetic resonance spectra (MRS) are crucial for validating analysis methods and training machine learning models.
• Existing tools like Magnetic Resonance Spectrum Simulator (MARSS) are limited in generating spectra for arbitrary sequences.

Purpose of the Study:

• To extend the MARSS tool to generate synthetic MRS for any NMR sequence, named synMARSS.
• To incorporate advanced physical parameters like relaxation, line shape distortion, and diffusion into spectral simulation.
• To enable simulation for non-proton nuclei (e.g., 2H, 13C, 31P) and investigate specific spectroscopic phenomena.

Main Methods:

• Developed synMARSS by extending the MARSS basis set simulation tool.
• Implemented a parametric model to generate synthetic spectra from basis sets and sequence parameters.
• Incorporated T1 and T2 relaxation, arbitrary line shape distortions, and diffusion effects.
• Extended capabilities to non-1H nuclei and investigated heteronuclear coupling approximations.

Main Results:

• synMARSS successfully generates high-quality synthetic MRS for arbitrary sequences.
• The tool efficiently produces large datasets for machine learning applications.
• Approximating 14N heteronuclear coupling as homonuclear coupling showed minimal impact on major metabolite concentrations at short echo times, but increased effects at longer echo times.

Conclusions:

• synMARSS is a versatile tool for generating synthetic MRS, supporting diverse sequences, non-proton nuclei, and advanced physical modeling.
• The findings highlight the importance of echo time in assessing the impact of coupling approximations on metabolite quantification.
• This advancement facilitates MRS data analysis, experimental design, and machine learning model development.