A comparative investigation of scatter correction in 3D PET

In 3D PET scatter degrades image quality and quantification. The currently most popular scatter estimation method is the single scatter simulation (SSS) which accommodates for multiple scattering by scaling the single scatter estimation. However, it has not been clear yet how accurate this approximation is for cases where multiple scatter is significant, raising the specific questions: 'How important double scatter correction is, and how accurately do we simulate the total scatter events by appropriate scaling?' This project aims to clarify the improvements in terms of quantification due to scatter correction, using: (i) single scatter events only, (ii) single and double scatter events, (iii) total scatter events, or (iv) scaled single scatter, and evaluate the analytic scatter estimation algorithm as implemented in the open source reconstruction software STIR. The analytic SSS scatter estimation implemented in STIR is compared with the SimSET Monte Carlo package. Scatter correction accuracy is examined for different levels of scattering and scaling approaches. A large anthropomorphic phantom was reconstructed with FBP. The images have been compared quantitatively: Areas with high scatter fraction are compared with single scatter corrected images and show a 50% bias reduction after performing single and double scatter correction. Scaled single scatter correction results are in good agreement with SimSET true events, less than 10% difference. Total-fit and tail-fit scaled single scatter results in approximately equal mean values. SSS correction with tail-fit scaling in STIR is very close with SimSET true events, 10% difference. The results show that multiple scatter correction improves accuracy and scaling single scatter is an efficient method to compensate for multiple scattering for standard PET scanning acquisitions.