Nonstationary compartmental modeling: application to receptor binding in PET

Objectives Nonstationary compartmental models (NCMs) allow kinetic parameters to change during the scan and are thus useful to characterize physiological response to a stimulus administered either before or during the PET examination. This abstract presents a general method for analysis of nonstationary models. We show an example of application of the technique to modeling of D2 receptor binding with C11 raclopride and amphetamine stimulus.

Methods We assume classical 1-, 2-, and 3-tissue compartment kinetic models. In general, any of the kinetic parameters describing the model k_x, x=1,…,6 is allowed to change during the scan and therefore they are functions of time k_x(t). These functions are modeled by splines with adjustable numbers of knots to model the complexity of k_x(t)'s (no functional model is assumed). The Markov Chain Monte Carlo algorithm is used to determine probable knot locations and knot values. Based on estimated knot characteristics, functions k_x(t) or macroparameters (e.g. volume of distribution or binding potential) are determined. We apply the technique to full reference 2-tissue compartment modeling of C11 raclopride in two monkeys with nonstationary regimes induced by administration of amphetamine during the scan. We assume that only k₃ (reflecting availability of receptors not occupied by the tracer or neurotransmitter, which is modulated by the drug challenge) was affected and modeled it as k₃(t) using the first order spline and 3 knots. Other parameters were assumed constant and unknown.

Results We obtained an excellent agreement of the Bayesian-model prediction with the data. The algorithm correctly identified time of stimulus injections by finding the first knot location very close to 30 min. The results were in agreement with previously published analysis using simplified models.

Conclusions The NCMs provide a useful, robust, and flexible tool to extend classical kinetic modeling.

Research Support R01CA165221, R21NS072148, R01MH100350