Fast analytic simulation toolkit for generation of 4D PET-MR data from real dynamic MR acquisitions

This work introduces and evaluates a fast analytic simulation toolkit (FAST) for simulating dynamic PET-MR data from real MR acquisitions. Realistic radiotracer values are assigned to segmented MR images. PET data are generated using analytic forward-projections (including attenuation and Poisson statistics) with the reconstruction software STIR, which is also used to produce the PET images that are spatially and temporally correlated with the real MR images. The simulation is compared with the GATE Monte Carlo package, which has more accurate physical modelling but it is 150 times slower compared to FAST for ten respiratory positions and 7000× slower, when repeating the simulation. The region of interest for mean values and coefficients of variation obtained with FAST and GATE, from 65 million and 104 million coincidences, respectively, were compared. Agreement between the two different simulation methods is good. In particular, the percentage differences of the mean values are: 10% for liver, and 19% for the myocardium and a warm lesion. The utility of FAST is demonstrated with the simulation of multiple volunteers with different breathing patterns. The package will be used for studying the performance of reconstruction, motion correction and attenuation correction algorithms for dynamic simultaneous PET-MR data.