Hybrid GATE: A GPU/CPU implementation for imaging and therapy applications

Abstract

Monte Carlo simulations (MCS) play a key role in medical applications. In this context GATE is a MCS platform dedicated to medical imaging and particle therapy. Yet MCS are very computationally demanding, which limits their applicability in clinical practice. Recently, graphics processing units (GPU) became, in many domains, a cost-effective solution to access high power computation. The objective of this work was to develop a GPU code targeting MCS for medical applications integrated within the GATE software. An aim was to enhance GATE computational efficiency by taking advantage of GPU architectures. We first developed a GPU framework with basic elements to run MCS for different medical applications. The implementation was based on a GPU adaptation of the Geant4 code. For each main GATE medical application, we developed a specific code from the GPU framework. Some of these GPU codes are currently being integrated in GATE as new features, and users can perform GPU computing in their GATE simulations. The acceleration factor resulting from the implementation of the tracking within the phantom on GPU was 60 for a PET simulation and 80 for a CT simulation. By using GPU architectures, we are also extending GATE to support optical imaging simulations that are heavily demanding in terms of computational resources. Radiation therapy applications currently supported by GATE V6.2 are also being adapted to run on hybrid GPU/CPU architectures.