Analytical simulations of dynamic PET scans with realistic count rates properties

In any domain, simulations are of utmost importance for assessing and predicting the performance of post-processing methods by controlling the truth and all model's parameters. In PET, two simulation methods exist: analytical and Monte Carlo. While not being as realistic as Monte Carlo ones, analytical simulations are fast enough for generating statistical replicates and realistic enough for assessing data-processing methods, especially in regard with noise. In this work, our aim was to produce replicated simulations of dynamic PET brain scans with realistic count rates properties. We extended the methodology of the analytical simulator ASIM [1] to get (i) a different modeling of the partial volume effect (down-sampling and image-based PSF modeling), (ii) a projection model based on real crystal coordinates, (iii) a fan-sum based random distribution and (iv) an intrinsic modeling of the decay and count rates through the dynamic acquisition. Attenuation, normalization (from real crystal efficiencies), scatters (convolution-based) and randoms (fan-sum based) were taken into account. The Zubal brain phantom and the geometry of the Siemens Biograph Hirez 4-rings scanner were used. All other input parameters were extracted from real acquisitions of volunteers with the HRRT: frame durations, time-activity curves, scatter and random fractions and total number of prompts for the whole acquisition. Simulated count rates were compared to those from the acquisitions on a frame-by-frame basis, for three different tracers. Results showed good agreements and realistic reconstructed images. These data will be used for assessing reconstruction and post-processing methods in the context of dynamic PET brain scans.