Quantitative analysis of dynamic PET data holds out the possibility of providing objective criteria for detecting coronary disease and for monitoring its progress over time. A common clinical procedure is to compare images following exercise or pharmacologic vasodilation to resting state images to identify regions with reduced flow reserve. Although the diagnostic utility of these images has been demonstrated, they are limited in their ability to provide comparisons between subjects or in the same subject over time because they do not account for differences in the amount and time course of the blood concentrations of the radioisotope. Quantitative criteria based on kinetic PET data have the potential to overcome this limitation, but detailed analyses of the sources of variability in these criteria are necessary before their advantage over image interpretation can be validated. Uncertainties from physical effects such as scatter, dead time, and crystal penetration which represent limitations of the PET data collection technology have been studied extensively at this laboratory and elsewhere. It is the purpose of this proposal to investigate the major non measurement sources and effects of uncertainty and variability. The statistical quality of dynamic PET is generally not good enough to provide estimates of enough parameters to describe physiological models as they are presently understood. For this reason, models are usually simplified by lumping several compartments together. We have shown that model reduction without careful analysis can produce misleading results. This previous work has prompted us to examine the issue again, combining theoretical analysis with simulations and improved tools for data analysis. Our primary interest in looking at lumped models will be to relate their parameters to the physiological parameters of a complex model and assess the bias introduced in their estimates. Additional sources of uncertainty to be studied include variation of kinetic parameters for a physiological region due to uncertainty about the boundary of the region, heterogeneity of kinetic parameter values in a region of interest, and uncertainty in the diagnostic criterion.
The specific aims of this proposal are: (1) To compare pharmacokinetic models of flow tracers such as 82Rb, 13NH3, H215O, 68Ga-BAT-TECH, and 62Cu-PTSM in the context of lumping analysis, time-scale analysis, and statistical error analysis. (2) To estimate the statistical uncertainty of kinetic parameters due to variability in region of interest placement, and to assess our ability to reduce this uncertainty by gating, modeling spillover, and specific guidelines for placement of regions. (3) To undertake a systematic investigation of the spatial heterogeneity of model parameters. (4) To assess the sensitivity of clinical data interpretation to PET measurement error, natural heterogeneity, region of interest placement, and physiological differences between subjects. A collection of over 160 dynamic PET previously acquired data sets from injections of H215O, 82Rb, and 68Ga-BAT-TECH will be analyzed in regions of interest designed to facilitate inter- and intra-subject comparisons.
