ABSTRACT
We present a probabilistic methodology to determine the heterogeneity distribution of model parameters within a region of interest in a positron emission tomography study using compartment models. We apply the least-squares fit in its probabilistic form by maximizing the corresponding likelihood for each voxel-wise time-activity curve separately yielding parameter estimates for each voxel. The distribution of the estimates is computed from the voxel-wise estimates with kernel technique and applying the maximum a posteriori estimation. The nuisance variation of the estimator is deleted. With simulated data, the algorithm found accurately even the sources of the bimodal distributions.
