Implementation and evaluation of scatter estimation algorithms in positron emission tomography

In positron emission tomography (PET) the current trend is to use the fully 3D capabilities of the scanner to increase sensitivity, hence improve the quality of data or reduce the scanning time. However, some difficulties have to be resolved. In 3D PET, the largest contributor to image degradation i...

Πλήρης περιγραφή

Λεπτομέρειες βιβλιογραφικής εγγραφής
Κύριος συγγραφέας: Τσούμπας, Χαράλαμπος
Άλλοι συγγραφείς: Νικήτα, Κωνσταντίνα
Μορφή: Thesis
Γλώσσα:English
Έκδοση: 2009
Θέματα:
Διαθέσιμο Online:http://nemertes.lis.upatras.gr/jspui/handle/10889/1800
Περιγραφή
Περίληψη:In positron emission tomography (PET) the current trend is to use the fully 3D capabilities of the scanner to increase sensitivity, hence improve the quality of data or reduce the scanning time. However, some difficulties have to be resolved. In 3D PET, the largest contributor to image degradation is Compton scatter since the scattered photons may comprise more than 50% out of all coincidences in the whole body studies. Much progress has been achieved the last few years by the use of scatter correction algorithms, such as the single scatter simulation (SSS). In this work, a model-based scatter simulation (MBSS) algorithm has been implemented in a software library called STIR (i.e. Software for Tomographic Image Reconstruction) initially based on SSS. The aim of the current work is to validate the MBSS implementation; investigate the influence of several parameters; and, if possible extend the existing algorithm. The results are compared with both SimSET Monte Carlo simulation package and measured data. The comparison shows that SSS is in excellent agreement with the single scatter distribution produced by SimSET and in several cases can also approximate accurately the total scatter. However, SSS is just an attempt to estimate the total Compton scatter effect, as it is possible that both photons may scatter, and potentially more than once. As shown, the single scatter distribution may have different shape from the total scatter distribution. How accurate this approximation is, it depends on how many detected photons are scattered multiple times. Multiple scatter is more likely to happen if the attenuation medium has large volume, hence it is more severe in 3D studies of the torso than the brain. In this work, the methodology used for the single scatter simulation algorithm is extended to handle twice-scattered events. Detailed description on how to implement the double scatter simulation (DSS) together with a preliminary evaluation is included. The results are promising even if the required computational time for DSS is much higher than for SSS, though not being prohibited. Finally, at the end of the thesis, an efficient recursive formula is proposed to estimate the rest multiple scatter distribution.