Quantitative accuracy of time-of-flight PET at high count rates

The benefit of time-of-flight (TOF) in PET reconstruction has been established for clinical imaging, and the signal-to-noise benefit with TOF has been demonstrated to increase with higher randoms fractions. TOF information mitigates and reduces image artifacts arising from errors in the corrections, but it may not account for all changes with count rate. Standard performance measures of image quality are carried out at low or clinical count rates for whole-body scanning, but the dependence of TOF quantitative performance on count rate has not been well-studied. All scanners with detector modules that employ photosensor-sharing experience pile-up of detected events with increasing count rate. We investigated the quantitative accuracy of TOF PET using cardiac and uniform phantoms imaged over a range of activities. At very high count rates the cardiac phantom showed significantly degraded image quality and quantification errors without TOF; TOF reconstruction was more stable to image artifacts and biases at high activities. For the uniform phantom, however, when the same number of prompt-delay events (trues+scatters) was reconstructed at different activities, a systematic negative bias in a large region was observed with increasing activity for TOF reconstructions. This bias was significantly larger than that for non-TOF reconstructions. This finding correlates with a cool center seen in the TOF images and suggests an overcorrection for scatter. At high count rates, object scatter modeled by the TOF-modified single scatter simulation (SSS) does not change, but detector pile-up leads to an increase in detected events with higher energies and long TOF tails than at low activities, neither of which is modeled by TOF-SSS. Deadtime correction factors implicitly include these sources of bias but depend on the algorithm (TOF vs. non-TOF) and likely the activity distribution as well through the inaccurate scatter estimate.