A new algorithm for scaling of PET scatter estimates using all coincidence events

The currently most popular scatter estimation methods for PET are based on the model-based single scatter simulation (SSS) algorithm. To accommodate for factors that have not been simulated, such as out-of-FOV activity or multiple scattering, the scatter estimate needs scaling. These scale factors are normally determined by fitting the scatter estimate to the ´tails´ of the data. Under certain circumstances, the ´tail´ region can become very small such that the fit can become unstable. In addition, the scatter estimate can become inaccurate in the ´tails´. This paper describes a new methodology to determine the scaling factors which uses all available data, and hence is more robust. The essence of the method is to include simulated data for the unscattered events in the fit such that all measured data (precorrected for randoms) can be included. We investigate the influence of appropriate constraints on the convergence.