Comparative evaluation of scatter correction techniques in 3D positron emission tomography.

Much research and development has been concentrated on the scatter compensation required for quantitative 3D positron emission tomography (PET). Increasingly sophisticated scatter correction procedures are under investigation, particularly those based on accurate scatter models and iterative reconstruction-based scatter compensation approaches. The main difference among the correction methods is the way in which the scatter component in the selected energy window is estimated. Monte Carlo methods provide further insight and might in themselves offer a possible correction procedure. Five scatter correction methods were compared in this study where applicable: the dual-energy window (DEW) technique, the convolution-subtraction (CVS) method, two variants of the Monte Carlo-based scatter correction technique (MCBSCI and MCBSC2) and our newly developed statistical reconstruction-based scatter correction (SRBSC) method. These scatter correction techniques were evaluated using Monte Carlo simulation studies, experimental phantom measurements and clinical studies. Accurate Monte Carlo modelling is still the gold standard since it allows the separation of scattered and unscattered events and comparison of the estimated and true unscattered component. In this study, our modified version of Monte Carlo-based scatter correction (MCBSC2) provided a good contrast recovery on the simulated Utah phantom, while the DEW method was found to be clearly superior for the experimental phantom studies in terms of quantitative accuracy at the expense of a significant deterioration in the signal-to-noise ratio. On the other hand, the immunity to noise in emission data of statistical reconstruction-based scatter correction methods makes them particularly applicable to low-count emission studies. All scatter correction methods gave very good activity recovery values for the simulated 3D Hoffman brain phantom, which averaged within 3%. The CVS and MCBSC 1 techniques tended to overcorrect while SRBSC undercorrected for scatter in most regions of this phantom. It was concluded that all correction methods significantly improve the image quality and contrast compared to the case where no correction is applied. Generally, it was shown that the differences in the estimated scatter distributions did not have a significant impact on the final quantitative results. The DEW method showed the best compromise between ease of implementation and quantitative accuracy, but entailed a significant deterioration in the signal-to-noise ratio.