Abstract
318
Objectives To improve quantitative accuracy of parametric images via introduction of MAP reconstruction priors based on voxel kinetics.
Methods Standard 3D dynamic PET imaging consists of independent image reconstructions at individual frames followed by application of appropriate kinetic model to the time activity curves (TACs) at the voxel or ROI level. Here we propose the use of priors generated based on clustering of pre-reconstructed dynamic images to define clustered neighborhoods of voxels with similar kinetics. The cluster-based priors have the advantage of further enhancing noise vs. bias trade-off performance in dynamic PET imaging, because: (a) there are typically more voxels in clusters than in local neighborhoods, and (b) neighboring voxels with distinct kinetics are less likely to be clustered together. Due to high level of noise in reconstructed dynamic frames, we developed dynamic clustering based on graphical integration of reconstructed images. For validation, the means of parameters estimated from 55 human 11C-raclopride dynamic PET studies were utilized for extensive simulations using a mathematical brain phantom. Distribution-volume (DV) images were estimated following MLEM, quadratic prior (QP-MAP) and the proposed cluster-based priors (CP-MAP) as applied to dynamic image reconstruction followed by graphical modeling, and were qualitatively and quantitatively compared for 11 regions-of-interest (ROIs).
Results CP-MAP reconstruction resulted in substantial visual and quantitative accuracy improvements in parametric DV images (compared to MLEM, over 50% noise reduction was observed with matched bias; compared to QP-MAP, over 20% noise reduction with smaller bias).
Conclusions The proposed “3.5D” dynamic PET reconstruction, moving beyond conventional 3D image reconstruction via incorporation of dynamically clustered priors, resulted in significantly enhanced quantification in parametric images.
Research Support This work was supported by the 973 Program of China under Grant No. 2010CB732503 and the NIH grant 1S10RR023623