Abstract

This work introduces MRS-Sim, an open-source, publicly available in vivo-like data simulator for generating synthetic magnetic resonance spectroscopy data. Current literature shows inconsistencies and varying parameters in simulating data, making it difficult to generalize and reproduce results. MRS-Sim is a powerful tool for modeling the complexities of MRS data for various clinical scenarios. It uses high fidelity basis functions to simulate sequence- and vendor-specific acquisitions. The underlying physics model includes all spectral components commonly found in standard spectral fitting routines and some novel components. The first is a 3D B0 field map simulator to model B0 field inhomogeneities, ranging from slight variations to severe distortions. The second is a novel semi-parametric simulator that generates poorly characterized residual water region signal and baseline offset contributions. This framework can simulate everything between raw multi-coil transients and preprocessed coil-combined data.

A repository of information has been compiled to help non-expert users with general simulation of MRS data. When simulating clinical-like datasets, it is important to be able to study the underlying ranges and statistical distributions of the simulation parameters of the clinical data. Therefore, accompanying software can analyze the distributions and ranges of parameters in fitted datasets, allowing simulations to be tailored to specific clinical data.

The modularity of this framework allows for easy customization of simulations to suit a variety of clinical scenarios and encourages continued community development. By simulating in vivo-like data, this framework can aid in many tasks in MRS, including verifying spectral fitting protocols and reproducibility analyses. The availability of readily available, synthetic data will also assist deep learning research for MRS, particularly in cases where clinical data is insufficient or unavailable for training. Providing easy access to high-quality synthetic data will help address reproducibility issues and make MRS research more accessible to a wider audience.

GitHub: 

https://github.com/JohnLaMaster/MRS-Sim 

*Note: While this work is still under review, the repository is being updated and finalized with a few new details. 

1H Spin Database: 

Metabolites, Macromolecules, Citations

Need help simulating data?

If you are looking to simulate synthetic data and need some help, please feel free to reach out.