vPET-ABC: Voxel-wise Approximate Bayesian Inference for Parametric Imaging of Neurotransmitter Release

We recently developed a method, called PET-ABC, that provides complete Bayesian statistical analysis of ROI-based dynamic PET data. The aim of this work was to extend the method to voxel-based analyses (vPET-ABC), with an initial focus on parametric imaging of neurotransmitter (NT) release in PET activation studies. In this study, the kinetic model used in conjunction with vPET-ABC was the linear-parametric neurotransmitter PET (lp-ntPET) model. This model, which incorporates time-varying kinetic parameters, describes the effect of neurotransmitter changes on dynamic receptor-ligand data during PET activation studies. The vPET-ABC pipeline produces reliable estimates of voxel-based parameters and their associated posterior probability distributions. It also computes model probability at the voxel level, where the two alternative models under consideration are lp-ntPET, which allows for an activation, and the more parsimonious Multilinear Reference Tissue Model (MRTM) which does not. Initials results, obtained from a simulated GATE 4D [11C]raclopride scan with realistic noise, demonstrated that vPET-ABC can provide insightful information about NT release at the voxel level, including the reliability of parameter estimates, which is important for reliably identifying subtle activations in noisy PET data. We believe the technique will be equally useful in the analysis of total body dynamic PET data, where the applicability of a single kinetic model to all tissues cannot be assumed.