Speeding up simulations for PET applications using CPU-based software
7th September 2021
J.J. Scheins, M. Lenz, U. Pietrzyk, N.J. Shah and C. Lerche
Accurate simulation is an essential tool for the investigation and design of many aspects of positron emission tomography (PET) imaging, e.g. scanner design, acquisition protocols, data corrections or quantification and modelling. In this respect, Monte Carlo simulations (MCS) using GATE software represent a fundamental approach to modelling photon interactions. However, simulations are computationally demanding and extremally time-consuming.
In response to this problem, researchers from INM-4 have developed a novel CPU-based software called the PET physics simulator (PPS), which combines several efficient methods to significantly boost performance.
In this work, the benefits of PPS are demonstrated for an elaborated PET scanner with 3-layer block detectors. The results show that all code optimisations yielded an acceleration factor of ≈20 (single core), and multi-threading on a high-end CPU workstation (96 cores) was shown to further accelerate the PPS by a factor of 80. This resulted in a total speed-up factor of ≈1600, which outperforms comparable GPU-based MCS by a factor of ≳2. The novel technique of coincidence multiplexing was also used to further enhance the throughput by an additional factor of ≈15, giving a combined acceleration factor of ≈24 000.
In this way, the PPS can be used to simulate sources and complex PET detector systems with an effective throughput of 106 photon pairs in less than 10 milliseconds. Moreover, the results of the research confirm that PPS is an accurate and highly competitive MCS tool for PET applications that require a maximum event throughput, e.g.MCS scatter correction or system response matrix modelling.
Following on from these initial developments, the team intend to further improve the method and anticipate that it will become a very effective means of aiding more accurate quantification in PET applications in the future.
Original publication
High-throughput, accurate Monte Carlo simulation on CPU hardware for PET applications