ABSTRACT
This chapter covers the foundations of iterative image reconstruction methodology for emission tomography (PET and SPECT), with a particular focus on PET. Iterative reconstruction methods allow more accurate modelling of the data, in particular the physics of the data acquisition process as well as the noise present in the data. Starting with how to represent radiotracer distributions as images, then considering how to model the scanner data acquisition process, the chapter proceeds to describe statistical methods for iterative reconstruction, making use of a Poisson noise model of the data. These maximum likelihood methods are then extended to consider prior information, such as our presuppositions regarding the radiotracer distribution, in order to compensate for noise. The chapter finishes by touching on more recent research directions, including direct 4D image reconstruction as well as inclusion of deep learning mappings into image reconstruction, in order to use data-informed priors for the reconstructed images.
