A model for multiple scatters in fully 3D PET

Scatter correction algorithms based on the forward-calculation of the distribution of single scatters must be combined with an algorithm to estimate multiple scatters in order to accurately estimate the overall scatter distribution. The multiple scatters can be modeled as the convolution of the single-scatter distribution with a one-dimensional Gaussian kernel. The characteristics of this kernel are determined with Monte Carlo simulations. The width of the kernel is assumed to be invariant. Its magnitude varies with transverse projection angle by a factor related to the path length through the object being imaged. Evaluations using several object geometries show that the model yields accurate results