Leveraging large publicly available datasets to benchmark new 1H-MRS modeling methods against established algorithms

Contemporary linear-combination modeling (LCM) methods for MRS data are usually implemented in compiled, ‘black-box’ fashion. The ability to modify underlying algorithms and/or introduce novel quantification approaches may improve the transparency, robustness and accuracy of metabolite estimation, but there is no established framework for rapid prototyping of modeling algorithms and benchmarking their performance. Here, we use a large, publicly available 3T PRESS dataset from multiple sites and vendors to assess the performance of a new open-source LCM algorithm, featured in a new MRS data analysis toolbox (‘Osprey’, github.com/schorschinho/osprey). Quantification of four major metabolites is compared to the widely used LCModel algorithm.