Computation of variance in compartment model parameter estimates from dynamic PET data

This paper investigates the validity of the analytical framework for bias and variance in kinetic parameter estimations. Analytical computation of bias and variance is compared against Monte Carlo simulations for two different compartment models at different noise levels. Difference between the estimated and measured variance increases with the level of noise and complexity of the compartment model. The standard deviation of the computed variance also increases with the increasing noise-level and model complexity. The difference between the estimated (from the formulation) and measured variance (from Monte Carlo simulations) is less than 1.5% for 1-tissue (1T) compartment model and less than 15% for 2-tissue (2T) compartment model at all noise levels. In addition, the standard deviation in the computed variance is less than 1% for 1T compartment model and less than 10% for 2T compartment model at all noise levels. These results indicate that the proposed framework for the variance in the kinetic parameter estimations can be used for 1-T and 2-T compartment models even in the existence of high noise.