Application of Scatter Events in TOF Offset Calibration for PET

A time-of-flight (TOF) calibration method of PET system using clinical datasets was proposed to partially replace the calibration with phantom scan. This method promises a more stable performance of TOF PET scan and a reduction of manual calibration frequency. In this study, the noise distribution of the TOF drift estimation, and the effects of scatter estimation is studied. In this method, the TOF projection from reconstructed image using non-TOF projections is used as a reference sinogram to estimate the timing drift in each line-of-response (LOR), and crystal-based TOF offsets are derived from the LORs' drift dataset using a maximum-likelihood approach. However, the conventional analytical projection of non-TOF image could not provide accurate information of estimated scatter events in the TOF sinogram, which decrease the accuracy of the drift calculation for each LOR. In this paper, the statistical distribution of the estimated TOF offsets for each LOR was studied using a uniform phantom. The effects of scatter were studied by comparing the results generated from no-scatter estimation (NS), single scatter simulation (SSS), and Monte Carlo simulations (MCS). The variance of the LOR-based timing drift was shown to be linearly correlated with the signal-to-noise ratio in the LOR. The results demonstrated both SSS and MCS improved the accuracy of the TOF offset estimation. It reduced around 30% RMSE between the estimated crystal-based TOF offsets and the known added drifts by applying the scatter estimation. Using the proposed TOF calibration method with MCS, the image quality of a measured uniform phantom with obvious drifts of TOF offsets is improved. The study showed that the inclusion of the scatter events derived from SSS and MCS reduced the estimation error of the data-driven TOF calibration method, and the proposed method could keep the performance of TOF reconstruction.