Evaluation of Recoverable Single Photons in a Silicon Based PET Detector Concept

Sammanfattning

In conventional positron emission tomography (PET), image reconstruction is based on coincident pairs of 511 keV annihilation photons. However, many annihilation events are not used because one photon could be scattered in the patient or not detected. This loss is especially important in larger patients, where attenuation and scatter reduce the number of true coincidences. A high resolution monolithic silicon detector could make it possible to recover part of this discarded information, since single unscattered photons with multiple detector interactions can be directionally constrained using Compton collimation. This thesis quantifies the availability of such single photon events and evaluates the spatial information they provide compared with conventional true coincidences. Monte Carlo simulations were performed for a fluorine-18 source in cylindrical water phantoms with diameters representing different patient sizes. Events were classified using simulation truth information into true coincidences and unscattered single photons with at least two detector interactions. The two event classes were evaluated using midplane backprojection. The results show that Compton collimated singles form a large event class relative to true coincidences, with the singles to coincidences ratio increasing from about 2 to about 22 over the investigated phantom diameter range. The coincidences produced narrow pointsource responses, with the full width at half maximum values being less than around 2.2 mm. However, the Compton collimated singles produced broad point source responses that reached up to around 33 to 34 mm for the largest phantom diameter. For simulated lesions in a background activity, the Compton collimated single photon images appeared smoother and had lower noise, while the coincidence images provided sharper lesion boundaries and better detectability of small structures. These results suggest that Compton collimated singles may be useful as complementary lower resolution information, especially for improving sensitivity or reducing image noise, but that coincidence data remain superior for precise localization and small lesion detection. Future work should investigate how reconstruction frameworks can combine the high spatial precision of coincidences with the larger event availability of singles to improve sensitivity or reduce image noise in PET.