TY - JOUR T1 - Making simulated PET images indistinguishable from real PET images for evaluation purposes JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1985 LP - 1985 VL - 52 IS - supplement 1 AU - Simon Stute AU - Thomas Carlier AU - K. Cristina AU - B. Hutton AU - Irene Buvat Y1 - 2011/05/01 UR - http://jnm.snmjournals.org/content/52/supplement_1/1985.abstract N2 - 1985 Objectives Monte Carlo (MC) simulations of PET scans are widely used to assess quantification methods. The simulated activity map (AM) is most often considered as piecewise constant, which may result in overestimation of performance. We describe a method to simulate PET images indistinguishable from real ones. Methods The input of the MC simulation consists of an estimated AM and associated attenuation map obtained from a patient PET/CT scan. A key step is to properly tune the spatial resolution (SR) and noise level (NL) of the input AM so as to reproduce PET images with SR and NL characteristic of clinical PET images. By simulating point sources with 0 to 20mm FWHM, we studied the relationship between SR in the input AM and in the PET images reconstructed from the simulated data. Using replicates of uniform activity distributions with NL (=SD/mean) varying from 0 to 90%, we characterized the propagation of NL from the input AM to the reconstructed images. All MC simulations were performed with GATE, based on a GEMINI GXL PET scanner. Images were reconstructed using OP-OSEM with 10 subsets, varying the number of iterations to change the SR/NL compromise. Results The NL in the images reconstructed from the simulated data was almost independent of NL in the input AM (<8% variation between NL=0 and NL=90%). SR in the reconstructed images varied as a function of SR in the input AM, from 6mm for a 0mm input SR to 20.5mm for 20mm input SR. An 8mm SR associated with a “clinical” NL could be obtained by using as an input AM real PET images obtained with a higher iteration number (corresponding to SR~5.5mm, high NL) than usually considered suitable for clinical use. These results were confirmed by using a real patient scan, for which the real and simulated PET scans could not be visually distinguished. Conclusions Highly realistic clinical PET scans with complex activity distribution can be simulated by using very high resolution but noisy clinical scans obtained with a higher than usual iteration number for reconstruction ER -