Towards fully automated MR-independent binding potential estimation for PET

Abstract

Estimation of parametric maps of binding potential (BP) for brain imaging with positron emission tomography (PET) is an essential stage of analysis when radioligands are used for neuroreceptor-related studies. A popular approach is the simplified reference tissue model (SRTM), as this avoids the need for arterial blood sampling. For now, in many PET centers, BP computation is neither automated nor standardized, and this can lead to method discrepancies between different research groups. The following work provides some benchmarks for the choices to make when computing the BP. Our results are based on simulated data. We examine the impact of different levels of pre-estimation smoothing on bias and variance, showing that the size of the reference region has little impact on BP estimates and that pre-estimation smoothing of the dynamic images has a better bias-variance tradeoff than post-estimation smoothing of the BP images. Moreover, our results demonstrate that there is scope for leniency in the choice of reference region location, and that this should be further examined with real data. These results can be used to design a fully automated, MR independent kinetic analysis method which is not only useful in its own right, but which also constitutes a very necessary step towards achieving fully automated direct (4D) BP parametric map estimation.