A parallel computational model for GATE simulations.

Area of Science:

• Medical Physics
• Computational Science
• Nuclear Instrumentation

Background:

• Monte Carlo simulations, such as those using GATE/Geant4, are crucial for Positron Emission Tomography (PET) but are computationally intensive.
• Traditional parallelization strategies struggle with PET's centralized coincidence processing and high communication overheads.

Purpose of the Study:

• To develop and evaluate a novel parallel computational model for GATE simulations tailored to PET experiments.
• To address the limitations of classical event distribution by decentralizing processing while maintaining coordination.

Main Methods:

• Implementation of a new parallel computational model within GATE, featuring factory classes for sequential and parallel execution.
• Decentralization of event generation and processing, coupled with a centralized event and time coordinator.
• Validation using a Mann-Whitney test to compare output equivalence with sequential models and computational performance evaluation for scalability and balance.

Main Results:

• The parallel model demonstrates equivalent output (in terms of tallies) to the sequential counterpart, confirmed by Mann-Whitney testing.
• Computational performance evaluations indicate that the proposed software is scalable and well-balanced.
• The model efficiently handles event generation and coincidence processing, overcoming limitations of previous approaches.

Conclusions:

• The developed parallel computational model offers an efficient and scalable solution for GATE/Geant4 Monte Carlo simulations in PET.
• This approach effectively manages the complexities of PET data processing in a parallel environment.
• The model enhances the feasibility of performing high-fidelity PET simulations with reduced computational burden.