We are developing novel approaches for post-processing of quantification of MRS(I) and CEST MRI metabolite detection. In particular, we will build on "Osprey", our new open-source platform (see Software and Databases Section, under the Resources page), as the framework to accommodate the new acquisition techniques developed in this TRD.
Osprey currently follows recent expert consensus on pre-processing, linear-combination modeling and quantification strategies of MRS data to ensure reproducible, transparent analysis and reporting. Compared to many other MRS analysis packages, Osprey’s modular structure allows each part of the analysis pipeline to be modified. In particular, we will implement novel ways of modeling, going beyond conventional single-spectrum linear-combination modeling or peak fitting methods. We will implement simultaneous multi-spectrum modeling to improve T1/T2 estimation in multi-TR/TE series, to increase the robustness of multiplexed spectral editing, and to explore the modeling of non-averaged data. Furthermore, we will incorporate new and more advanced reconstruction of MRSI and CEST methodology as soon as we complete their development (see Multi-metabolite editing and CEST metabolic imaging sections).
The Osprey environment further allows to develop and incorporate novel classes of deep-learning-based methods for data reconstruction, artefact rejection, and spectral quantification. In particular, we will design a framework to generate large amounts of synthetic MRS(I) and CEST data. These artificial datasets will be used to train network-based methods, for example to separate short-T2 (macromolecule) components from long-TE (metabolite) components. Finally, we will develop methods that mutually inform MRS(I) and CEST acquisitions with the goal of increasing the spectral and spatial resolution of metabolic imaging, for example by predicting the macromolecular background in short 1H-MRS spectra from CEST Z-spectra.