An accurate system model for the PET/CT IRIS pre-clinical scanner

Positron Emission Tomography (PET) is a nuclear medicine technique which allows to determine the in-vivo spatial and temporal distribution of a radiotracer inside an examined subject. PET has applications both in clinical and pre-clinical studies, where in the latter it is used as a research tool for drug development and targeting and development on small animal subjects like mice and rats. The process of image formation starting from the data acquired by a PET scan- ner is called image reconstruction and, nowadays, can be performed with the use of iterative algorithms. These algorithms offer superior results at the expense of a high computational cost, in terms of calculation time and memory requirements, and can be performed only on modern workstations. With the use of the itera- tive methods the quality of the reconstructed image relies to a great extent on the accuracy of the so-called system model (or system matrix), which defines the re- lationship between the object space and the measurement space. The creation of the system model is a computational challenge and it is often subdivided into factors that can be computed separately to reduce the time needed to perform its construction. This thesis focuses on the creation and the validation of an accurate factorized system model for the INVISCAN PET/CT IRIS scanner. The model is made of two components, an analytical component called geometric matrix and a second component, called detector matrix, which models the interactions of the photons inside the detector modules in order to take into account the penetration and inter- scatter effects that can occur in the crystals matrix of the detector modules. The geometric, the detector and the system matrices cannot be created as dense matrices due to their huge number of elements. The whole process of data manipulation has been performed with the use of sparse matrix dedicated algorithms. The first part of the thesis focuses on the creation of the system model of the scanner. First the detector matrix is created starting from a GEANT4 Monte Carlo (MC) simulation and then the system matrix is assembled joining the geometric and the detector matrices. Giving the high number of lines of responses (LOR) of the scanner, the MC needs to be optimized to obtain a reasonable total simulation time. The optimization is performed by exploiting the symmetries of the scanner and by tuning the parameters of the simulation for each LOR. The detector matrix can be created with different levels of accuracy that strongly affect the memory footprint of the whole system model, which, in turn, affect the computational requirements of the entire image reconstruction platform. Three models of different accuracy are here proposed. The second part of the thesis focuses on the comparison between the currently used system model of the IRIS scanner and the new models. The comparison has been performed analyzing the Image Quality phantom described in the NEMA NU 4 - 2008 protocol Performance Evaluation of small animals Positron Emission Tomographs. The images reconstruction has been performed on real data of the phantom acquired within the IRIS scanner in possession of the Istituto di Fisiologia Clinica - CNR Pisa. The results showed that the new factorized models improves the noise of the reconstructed images and the so-called recovery coefficients, that are related to the spatial resolution of the scanner