Effect of measurement uncertainty on region of interest based and parametric binding potential estimates for the high resolution research tomograph (HRRT)

High resolution Positron Emission Tomography (PET) imaging leads to very small pixel sizes. Generally the increase in resolution is not paralleled by a corresponding increase in sensitivity which may cause the count density per voxel to be low. Here we are exploring how the statistical quality of the data acquired with the high resolution research tomograph (HRRT) influences the accuracy of the determination of the binding potential (BP) for typical human studies performed with ¹¹middotC-raclopride. Susceptibility to noise was tested for 3 modelling approaches: the Logan graphical model, the simplified reference tissue method (RTM) and the delayed ratio method (DRM). For each approach BP was calculated on a region of interest (ROI) and voxel basis (parametric maps). Using a method based on experimentally defined replicas of time activity curves (TACs) representative of those obtained in human scans we found that for this tracer the contribution of the statistical noise to the BP determination is ~ 5-8 % when the TACs are evaluated on an ROI basis (either ROI TACs used as input, or ROI placed on the BP parametric image) and 9-12 % when calculated on a single pixel basis. The Logan approach was found to suffer from a considerable bias due to statistical noise when the BP was calculated on a single pixel basis, while RTM and DRM showed no such bias. Overall, for this tracer and these scanning conditions the RTM proved to be the least sensitive to statistical noise in the data.