Abstract
In the past few years, the collection of positron emission tomography (PET) data without inter-plane shielding has become a widely accepted technique for significantly increasing the sensitivity of multi-ring scanners. However, the resultant increase in the registration of counts due to scattered events is undesirable for quantitative studies, since it reduces contrast and confounds the linear response of the scanner to activity concentration. This thesis describes the development, implementation and evaluation of a correction for scattered photons based on the simultaneous acquisition of emission data in two energy windows. Initial experiments were performed to characterize the distribution of of scattered photons in data collected with a commercial PET brain scanner operated without inter-plane shielding (septa). In this mode of acquisition, termed 3-D, coincidences between all rings of detectors are accepted. This is in contrast to the conventional 2-D acquisition mode, where data is acquired with a more restricted range of inter-ring combinations. The fraction of scattered photons under standard operating conditions was measured as 35 (+/-2) % for a line source in a 20 cm diameter water-filled cylinder, and the scatter response function found to be shift-variant. A 20% gain in counts from events that do not scatter in the object but in the detectors themselves was achieved by lowering the energy threshold as far as reasonably possible. The correction developed for scattered photons relies on parameters relating two energy windows which were selected to maximize counting statistics and minimize spatial variations. The ratio functions for the selected windows were found to be shift-invariant, and showed little variation with object size. The parameters were however found to be very susceptible to changes in detector efficiency, showing up to 10% variations over time. The correction was implemented with constant values taken for the ratio functions, and integrated into the routine reconstruction sequence with pre-processing steps taken to minimize noise propagation. When evaluated in a range of standard and customized test objects, the correction restored contrast in inactive areas to within 5% of the true value. Relative activity concentrations in different sized phantoms were restored to better than 6%. A means of calibrating the data corrected for scattering was implemented and quantification in a range of activity distributions was accurate to within 7%. The correction method was tested in a phantom which simulates the activity distribution in a human brain. Applying the method to human data confirmed the potential of using this method routinely for quantification in vivo. A limitation of the method for dynamic scanning was identified: high count rate pile-up effects introduce global spatial and spectral distortions which are enhanced in the dual energy window correction for scattering. However, in multi-time frame scanning of test phantoms, the correction consistently restored contrast and maintained linearity. The data acquisition, correction for scattering and reconstruction regimes that have been developed in this work have, thus far, allowed the routine collection of several hundred dynamic ligand studies in patients and normal volunteers, which have been analyzed as part of clinical research projects in a fully quantitative manner.