PET benchmark

The PET benchmark simulates a whole-body scanner that does not correspond to any existing system. Rather, it serves as a simple system that illustrates the majority of the features of GATE. It consists of 8 detector heads arranged in a 88 cm diameter and 40 cm axial length octagonal cylinder. Each head is made of 400 detector blocks and each block is a 5 x 5 array of dual-layer LSO-BGO crystals. The heads are equipped with partial septa and can rotate in a step-and-shoot mode.

The phantom is a 70 cm long water cylinder with an F18 (half-life = 109.8 min) and an O15 (half-life = 2.07 min) line source of 100 kBq. The simulated acquisition is 4 min in duration. The source activities are set such that the PET benchmark runs in less than 10 CPU hours on a 1 GHz processor. The acquisition is divided into two 2 min frames; after the first frame, the gantry rotates by 22.5 deg. Only coincident events are recorded, using a coincidence time window of 120 ns. This large window is used in order to record a large number of random coincidences. The lower and upper energy thresholds are set to 350 and 650 keV, respectively.

The standard electromagnetic package of Geant4 is used. Only the Rayleigh interactions are modeled using the low energy package. To speed up the simulation, X-rays and secondary electrons are not tracked.

Approximately 3.7 x 107 decays occur during the simulated acquisition. About 7.0 x 105 coincidences are recorded and written in a ROOT file. Based on the ROOT output, figures and plots are calculated to check the correct execution of the simulation. The benchmark results include 1) the total number of generated events and detected coincidences, 2) their spatial and time distributions, 3) the fraction of random and scattered coincidences, 4) the average acolinearity between the 2 annihilation gammas.

The PET benchmark has been run on twelve different system configurations. Two operating systems have been tested : Linux and Mac OS (versions 10.2.6 and 10.3). The Linux distributions were RedHat (versions 9.0, 8.0 and 7.3) and Suse (version 8.1). The source code compilation has been performed either with gcc 2.95 or with gcc 3.2. The computing time for the PET benchmark varied between 6 and 10 hours with 1.5 GHz and 3 GHz processors. It grew up to 12 hours with 1 GHz processors.

The results in the table below and the standard deviations show that the main parameters of interest are stable whithin less than 1 %. It is strongly recommended that the user validates a new or updated GATE installation using this table.