ABSTRACT
One of the many advantages of PET imaging is that it oers quantiable measures of the underlying biology. Yet, in most instances PET images are quantied by normalizing average values in regions of interest (ROIs). is “normalization” scheme takes the form of the “standardized uptake value” (SUV) or related to a reference region presumably devoid of the underlying biological process under investigation. While the application of SUV has been discussed in detail by others (Boellaard 2009, Ito et al. 2009, Lucignani et al. 2004, Ning et al. 2009, Suzuki et al. 2009, Wahl et al. 2009) to name a few, we will discuss quantication of small animal PET images. Quantication in this regards is dened modeling the kinetics of the imaging agents in tissue. Recent examples of kinetic modeling in small animals span many elds, including oncologic imaging (Kim et al. 2008, Pollok et al. 2009), cardiac imaging (Shoghi 2009, Shoghi et al. 2008), and neuroimaging (Guo et al. 2009) among others. By far, the predominant choice of tracer is 18FDG largely owing to its availability. Kinetic modeling oers several advantages in comparison to “normalization”
schemes, in that it oers measures of transport/perfusion rates which can be used to tease out delivery issues from metabolism/binding of ligands. Kinetic analysis also oers means to measure metabolic rates and receptor density, as well as fracture of underlying vasculature which can be used to discern “true” tissue activity especially in situations when one expects changes in vasculature, for example, due to injury. Kinetic analysis also oers means to separate specic from nonspecic signal, for example, when a ligand is found to have a secondary nonspecic-binding site. Finally, with more advanced application of kinetic analysis, spatial information on the kinetics in tissue may be characterized. Quantication of small animal PET images is a process which requires: (1) the derivation of the input function, (2) a notion of a kinetic model for the interaction of the imaging agent in tissue, (3) parameter estimation, and nally (4) validation of the model. To some extent, the latter three steps may be iterated until the appropriate model is conrmed.
