Endpoint-driven measurement designs for hyperpolarized Magnetic Resonance

Hyperpolarized carbon-13 magnetic resonance (MR) is a powerful technique that has opened a window into in vivo metabolism through real-time, non-invasive observation of biochemical pathways in living systems. By hyperpolarizing a substrate and introducing it to a biological system of interest, significant magnetization enhancement is achieved for both the substrate and its metabolites, which would otherwise be undetectable. Both nuclear magnetic resonance (NMR) and magnetic resonance imaging (MRI) can be used for data acquisition. This dual capacity allows the investigation of cell samples or entire organs and organisms, providing insights into fundamental metabolism as well as disease diagnostics.

To convert hyperpolarized MR data into meaningful metabolic information, pharmacokinetic models are employed that quantify metabolic flux through modeling of conversion rate parameters. The elevated polarization of the substrate introduced by hyperpolarization is a finite and short-lived resource. The enhanced magnetization is lost due to relaxation and measurements as these involve excitations that cost magnetization. Both relaxation and excitation processes need to be accounted for in the pharmacokinetic model.

An initial focus of this study was to evaluate the appropriateness of pharmacokinetic models across multiple hyperpolarized experimental settings. Initially, a two-metabolite unidirectional enzyme-driven reaction was successfully modeled for a pulse-and-acquire type sequence, and the model was verified through in vitro experiments. Subsequently, a system involving up to three metabolites was measured using a balanced steady state free precession (bSSFP) sequence to detect low-yield metabolites. A novel model was developed and validated through in vivo studies and was compared to existing GRE models. Finally, an attempt was made to model a complex seven-metabolite system. Initial efforts to apply the same logic as with previous settings were unsuccessful, leading to the exploration of an alternative framework, D-CODE, which also did not yield success.

The set of flip angles applied for excitation constitutes a measurement parameter that can be freely chosen, and which directly affects the generated signal that, in turn, influences the uncertainty of estimated pharmacokinetic model parameters. Therefore, a framework was developed to optimize variable flip angle (VFA) schemes, with the goal of minimizing the variance of pharmacokinetic model parameter estimates. This was achieved through maximizing Fisher information about the model parameters. The framework was made flexible such that any model parameter can be in focus. Through in silico Monte Carlo simulations and experiments in vitro, it was demonstrated that VFA schemes outperformed optimized constant flip angle (CFA) schemes, providing conversion rate estimates with reduced uncertainty. Additionally, VFA schemes consistently reproduced a range of underlying parameters and accurately estimated additional model parameters, including B1-field strength scaling. Wrongly assumed flip angles resulting from unknown B1-field variations can be detrimental for conversion rate estimates. However, by application of a prior, robustness towards this was incorporated.

In conclusion, optimized VFA schemes significantly improve conversion rate estimates, which can aid quantification of metabolism and, in the future, potentially also improve human disease monitoring.