I am learning scatter correction from papers with ETM(the estimation of trues method).
Firstly, choose a high energy window.In many paper, ETM is the hypothesis that in the high energy window the scattered events are negligible. According to coincidence counts and scatter counts(Figure 1&2 and Table 1), I chose the whole energy window(WEW:350KeV-650Kev) and high energy window(HEW:550-650KeV).
Then, I choose a scale factor to scale high energy window according to all coincidence counts in WEW and HEW. But I found the delta sinogram (=WEW sinogram - scale(HEW)) slice had negative number(It makes me confuse, should it be all positive number? cuz True events maybe are distributed in different place in WEW and HEW sinogram) . Then, get scatter estimation by filtering with a Gaussian filter(3X3 kernel with sigma=3). Finally, use WEW - scatter estimation and convert sinogram to listmode(cdf).
Result are shown in Figure3-5. It looks like little different between scatter corrected and True+scatter. Is there any mistake in my operation? Or should I use NEMA IQ phantom? Or should I change to learn Single Scatter Simulation?
Hi Hannah,
I worked with ETM more than 20 years ago and I am commenting just from memory, so I may be missing something. But, if I remember well, one of the problems I noticed at the time was that in order to improve the results of ETM I should try to get rid of scatter in the HEW because some was still present and after being scaled by the factor, it messed up the scatter estimate in the WEW (part of your negative values may come from that). One possibility was to use the Single Scatter Simulation (SSS) in the HEW, which I had access to.
Another thing I noticed at the time, was that in order to improve the results, ideally the normalization correction should be adapted to the HEW: normalization factors are slightly different in the two energy windows, so I tried to measure them in the two windows (otherwise, after scaling the HEW, they may produce some patterns related with not-so-good normalization factors being applied). This may help explaining why your scatter estimate seems to have some patterns such as the losangles in the sinogram.
Finally, after those steps, some heavy smoothing was necessary (which is ~ok, the scatter distribution is expected to have mainly low frequency components).
What I did is better explained in this paper from 2000: https://www.researchgate.net/profile/Bernard_Bendriem/publication/11329084_A_hybrid_scatter_correction_for_3D_PET_based_on_an_estimation_of_the_distribution_of_unscattered_coincidences_Implementation_on_the_ECAT_EXACT_HR/links/56d03c1708ae059e375cbccf/A-hybrid-scatter-correction-for-3D-PET-based-on-an-estimation-of-the-distribution-of-unscattered-coincidences-Implementation-on-the-ECAT-EXACT-HR.pdf
Hope this helps,
Nuno
Thanks for your reply! I download your paper and this paper describes many details, such as scale factor. It helps me a lot! I will read this paper carefully! And try to figure out SSS formula.
