Preliminary evaluation of a kinetic parameter estimator with application to direct parametric reconstruction

We present a ML-EM estimator for kinetic parameters from list-mode and histogram mode (hist-mode) dynamic PET data, based upon the observation that emissions originating from each contributing exponential mode in the compartment model are identically and independently distributed samples drawn from an inhomogeneous Poisson distribution. An approximation formula for the covariance of the estimator is developed based on the Cramer-Rao bound, validated for 1- and 2-compartment models, and compared with multiple noise realizations. 1D experimental data were simulated using various count levels and rate constants typical of metastatic colorectal cancer and glioma FDG uptake. We conclude that estimation of kinetic parameters from list-mode data is theoretically achievable and that estimate covariance can be usefully approximated from a single realization.