A comparative study of multiple scatter estimations in 3D PET

We compare two methods with different accuracy in modeling of multiple scatter estimation in 3D PET. In the first approach, the single scatter is obtained from model-based simulation while the multiple scatter projections are estimated by convolving the single scatter projections with a one-dimensional Gaussian kernel. A new convolution kernel model is proposed, in which both the amplitude and width of the kernel are assumed to vary as a function of a filtered attenuation sinograms. In the second approach, both single and double scatter are calculated from model-based simulations using an implementation based on the STIR library (Software for Tomographic Image Reconstruction).