Direct EM reconstruction of parametric images from list-mode brain PET using a novel model based on logan graphical analysis

Abstract

Direct reconstruction of parametric images from raw dynamic PET data has the potential of producing lower noise images than obtained using intermediate frame-based reconstructed images, due to the accurately characterized statistical properties of raw PET data. The goal of this study was to extend a previous direct parametric reconstruction algorithm (PMOLAR). PMOLAR uses the Expectation Maximization (EM) algorithm to estimate the parametric maps. Previous versions of PMOLAR were based on the one-tissue (1T) compartment model (PMOLAR-1T). The 1T model is suitable for some PET tracers, but most reversible PET radioligand kinetics are better described by more complex models, mainly the two-tissue (2T) compartment model. Alternatively, Logan Graphical Analysis (GA) can be applied to all reversible PET tracers. In this study PMOLAR was adapted to a new model based on GA. Two versions of the new reconstruction algorithm were investigated. PMOLAR-GA was evaluated on human data acquired on the High Resolution Research Tomograph (HRRT) after injection of [11C]PBR28, a radiotracer used to study neuroinflammation. The new model derived from GA was first compared to previous modeling methods (MA1, LEGA, 2T) suitable for [11C]PBR28 on fits of region of interest time-activity curves. Then the two new versions of PMOLAR were evaluated on a human 4D PET data set. PMOLAR-GA parametric maps were compared to frame-based parametric maps. Using routine reconstruction settings, frame based parametric maps were visually very noisy and biased (between 27±4% and 222±121% on a regional level, depending on the modeling method). While the PMOLAR-GA mean regional values were very close to reference values (11%±9% or 8%±9%), with visually lower noise at the voxel level.