Abstract
313
Objectives To investigate direct parametric imaging using a novel 4D AB-OSEM reconstruction algorithm incorporating graphical modeling of [C-11]PiB.
Methods Direct 4D parametric imaging is an emerging technique that attempts to tackle the noisy voxel-level kinetics of PET images. In graphical modeling of reversibly binding tracers, the intercept term is commonly negative, posing a challenge to closed-form 4D EM formulation. As a solution, we developed a generalized AB-OSEM algorithm in the 4D context, allowing a negative bound for the intercept term, and applied this technique to reference-tissue relative equilibrium (RE) graphical modeling of PiB. To achieve enhanced quantitative performance, the slope and intercept images are initialized using conventional parametric images, from which spatially varying lower bounds are also estimated and incorporated within AB-OSEM. For validation, kinetic parameters from 11 PiB studies on normal controls were estimated, averaged and assigned to a mathematical brain phantom followed by realistic simulations (30 noise realizations). Distribution-volume ratio (DVR) images were reconstructed using indirect vs. direct methods, and were qualitatively and quantitatively compared for 11 regions-of-interest (ROIs). The method was also evaluated on a 90min PiB study on the HRRT scanner.
Results Direct 4D AB-OSEM reconstruction resulted in substantial visual and quantitative accuracy improvements in parametric DVR images (>30% noise reduction from 23% to 15%; matched bias). For patient study, the proposed method was shown across the ROIs to visually and quantitatively outperform the conventional method: for matched DVR values, improved noise levels (~35-40%) were obtained.
Conclusions Direct parametric imaging applied to PiB-PET using a novel closed-form iterative 4D AB-OSEM formulation achieves significant noise reduction compared to conventional parametric imaging