LaMaster John, Oeltzschner Georg, Li Yan
Munich Institute of Biomedical Engineering, Technical University of Munich, Munich, Bavaria, Germany.
School of Computation, Information, and Technology, Technical University of Munich, Munich, Bavaria, Germany.
NMR Biomed. 2025 Oct;38(10):e70130. doi: 10.1002/nbm.70130.
Realistic, in vivo-like synthetic data is increasingly needed to develop and validate methods in magnetic resonance spectroscopy. MRS-Sim is a powerful, open-source framework for simulating such data while providing known ground truth values. Its modularity enables modeling the complexities of MRS data for various in vivo scenarios. The underlying physical equations include both commonly used spectral components of linear-combination fitting routines and two novel components. The first is a 3D field map simulator that models field inhomogeneities, ranging from slight variations to severe distortions. The second is a novel semi-parametric generator that mimics signals from poorly characterized residual water regions and spectral baseline contributions. This framework can simulate scenarios ranging from raw multi-coil transients to preprocessed, coil-combined multiaverage data. Simulating realistic in vivo-like datasets requires appropriate model parameter ranges and distributions, best determined by analyzing the fitting parameters from existing in vivo data. Therefore, MRS-Sim includes tools for analyzing the ranges and statistical distributions of those parameters from in vivo datasets fitted with Osprey, allowing simulations to be tailored to specific datasets. Additionally, the accompanying repository of supplemental information assists nonexpert users with general simulations of MRS data. The modularity of this framework facilitates easy customization in various in vivo scenarios and promotes continued community development. Using a single framework for diverse applications addresses the inconsistencies in current protocols. By simulating in vivo-like data, MRS-Sim supports many MRS tasks, including verifying spectral fitting protocols and conducting reproducibility analyses. Readily available synthetic data also benefits deep learning research, particularly when sufficient in vivo data is unavailable for training. Overall, MRS-Sim will promote reproducibility and make MRS research more accessible to a wider audience.
在磁共振波谱学中，开发和验证方法越来越需要逼真的、类似体内的合成数据。MRS-Sim是一个强大的开源框架，用于模拟此类数据，同时提供已知的真实值。其模块化能够对各种体内场景下MRS数据的复杂性进行建模。基础物理方程既包括线性组合拟合程序中常用的光谱成分，也包括两个新的成分。第一个是3D场图模拟器，用于模拟场不均匀性，范围从轻微变化到严重畸变。第二个是一种新型的半参数发生器，用于模拟来自特征不明确的残留水区域和光谱基线贡献的信号。该框架可以模拟从原始多线圈瞬态到预处理的、线圈组合的多平均数据等各种场景。模拟逼真的类似体内数据集需要合适的模型参数范围和分布，最好通过分析现有体内数据的拟合参数来确定。因此，MRS-Sim包括用于分析来自用Osprey拟合的体内数据集的那些参数的范围和统计分布的工具，从而使模拟能够针对特定数据集进行定制。此外，随附的补充信息存储库可帮助非专业用户进行MRS数据的一般模拟。该框架的模块化便于在各种体内场景中轻松定制，并促进社区的持续发展。使用单个框架进行多种应用解决了当前协议中的不一致问题。通过模拟类似体内的数据，MRS-Sim支持许多MRS任务，包括验证光谱拟合协议和进行再现性分析。现成的合成数据也有利于深度学习研究，特别是在没有足够的体内数据用于训练时。总体而言，MRS-Sim将促进再现性，并使更广泛的受众更容易进行MRS研究。
