Bias Evaluation and Reduction in 3D OP-OSEM Reconstruction in Dynamic Equilibrium PET Studies

Abstract

Background: Iterative image reconstruction algorithms are widely used in positron emission tomography (PET). However, they are known to contribute to quantitation bias, which is particularly pronounced during dynamic PET studies such as neuroreceptor binding studies with 11C-labelled radiotracers where count rates become low towards the end of the examination. This problem is relevant in case simultaneous PET/MR studies which apply a bolus-infusion protocol to allow the multimodal comparison between control or resting state and stimulation effects in a single session of e.g. 60 min, i.e. 3 half-lives of 11C. A quantitation bias may interfere with stimulation related changes. In this study, we evaluated the impact of the 3D ordinary Poisson OSEM (3D OP-OSEM) on quantitation accuracy reconstructions and the subsequent propagation into binding potential values using a decay study of a 11C filled phantom and a human brain data set. To evaluate the reconstruction bias, we tested different reconstruction framing schemes and propose a framing scheme that keeps the counts per time frame constant over the full acquisition time. We also compared the vendor’s 3D OP-OSEM image reconstruction method to an in-house developed reconstruction (PRESTO toolkit). Results: In general, a bias for low and high activity concentration regions was observed in the range of ± 3% and - 3% to 5%, respectively. Using the alternative proposed framing scheme, a lower bias was achieved for regions with low activity concentration using PRESTO, and a stabilization of the mean bias for the binding potential was achieved at around 5% with the vendor’s reconstruction throughout the relevant activity curve time interval. Conclusions: The bias in activity concentration propagated into the binding potential values leading to a mean bias of around 5%, thus allowing the detection of changes in binding values during equilibrium of > 5%.