A Bayesian hierarchical model for radiopharmaceutical 18F-choline in prostate cancer imaging

Objectives The radiopharmaceutical ¹⁸F-choline is increasingly being used for imaging of primary and recurrent prostate cancer. Our aim is to develop a hierarchical model to describe the pharmacokinetics of ¹⁸F-choline after an intravenous injection and to use Bayesian analysis for the estimation of parameters in the proposed model.

Methods By using singular perturbation theory, we simplify the full compartmental model to a reduced model in which the compartment of blood is separated from the organ compartments. We use Bayesian posterior simulation to estimate the parameters in the blood compartment and sequential Monte Carlo to estimate the parameters in the organ compartments. We apply our method in both the individual level - estimating the parameters of each patient, and on the population level - estimating the parameters of the population from which the patients came from.

Results Results show a higher uncertainty in the estimated concentration in the organ compartments shortly after injection. By pooling information from all patients, we are able to estimate the parameters and build a confidence interval in the population level, thus making the model useful for new patients.

Conclusions We propose a Bayesian parameter estimation method in pharmacokinetic models by combining model reduction techniques with Monte Carlo Markov Chain and sequential Monte Carlo methods to quantify uncertainty in pharmacokinetic parameters and reveal their correlation structure. The results can be used to improve the sampling times for pharmacokinetic investigations by analyzing the different sources of uncertainty.