The effects of collimation on PET image noise due to scatter, random counts, and deadtime

The rates for true coincident events, scattered events, and singles in a Positron Emission Tomograph (PET) depend on the collimator size and shape. Monte Carlo techniques were used to compare efficiencies for wanted and unwanted events in multislice PET (MS-PET) and open collimator positron volumetric imaging (PVI) configurations. All systems used cylindrical arrays of bismuth germanate (BGO) or NaI crystals 51 cm in diameter and 10 cm deep suitable for whole brain imaging. The PVI systems detect about five times more true coincident events at low activity concentrations, but their scatter fraction is about three times higher. They are also much more sensitive to activity outside the scan field. As well as causing random counts when they fell within the energy acceptance window, single events are the main cause of deadtime. When the detectors are made from light-encoded blocks deadtime is the major limitation at high count rates. When discrete crystals are used, the efficiency is lower and the random count rates are a more significant source of noise. Noise-effective count rates are used to compare the relative cost in system performance among different systems and sources of noise.