GPU-based fast Monte Carlo dose calculation for proton therapy

Xun Jia, Jan Schümann, Harald Paganetti, Steve B. Jiang

Research output: Contribution to journalArticle

85 Citations (Scopus)

Abstract

Accurate radiation dose calculation is essential for successful proton radiotherapy. Monte Carlo (MC) simulation is considered to be the most accurate method. However, the long computation time limits it from routine clinical applications. Recently, graphics processing units (GPUs) have been widely used to accelerate computationally intensive tasks in radiotherapy. We have developed a fast MC dose calculation package, gPMC, for proton dose calculation on a GPU. In gPMC, proton transport is modeled by the class II condensed history simulation scheme with a continuous slowing down approximation. Ionization, elastic and inelastic proton nucleus interactions are considered. Energy straggling and multiple scattering are modeled. Secondary electrons are not transported and their energies are locally deposited. After an inelastic nuclear interaction event, a variety of products are generated using an empirical model. Among them, charged nuclear fragments are terminated with energy locally deposited. Secondary protons are stored in a stack and transported after finishing transport of the primary protons, while secondary neutral particles are neglected. gPMC is implemented on the GPU under the CUDA platform. We have validated gPMC using the TOPAS/Geant4 MC code as the gold standard. For various cases including homogeneous and inhomogeneous phantoms as well as a patient case, good agreements between gPMC and TOPAS/Geant4 are observed. The gamma passing rate for the 2%/2 mm criterion is over 98.7% in the region with dose greater than 10% maximum dose in all cases, excluding low-density air regions. With gPMC it takes only 6-22 s to simulate 10 million source protons to achieve ∼1% relative statistical uncertainty, depending on the phantoms and energy. This is an extremely high efficiency compared to the computational time of tens of CPU hours for TOPAS/Geant4. Our fast GPU-based code can thus facilitate the routine use of MC dose calculation in proton therapy.

Original languageEnglish (US)
Pages (from-to)7783-7797
Number of pages15
JournalPhysics in Medicine and Biology
Volume57
Issue number23
DOIs
StatePublished - Dec 7 2012

Fingerprint

Proton Therapy
Protons
Radiotherapy
Uncertainty
History
Air
Electrons
Radiation

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging
  • Radiological and Ultrasound Technology

Cite this

GPU-based fast Monte Carlo dose calculation for proton therapy. / Jia, Xun; Schümann, Jan; Paganetti, Harald; Jiang, Steve B.

In: Physics in Medicine and Biology, Vol. 57, No. 23, 07.12.2012, p. 7783-7797.

Research output: Contribution to journalArticle

Jia, Xun ; Schümann, Jan ; Paganetti, Harald ; Jiang, Steve B. / GPU-based fast Monte Carlo dose calculation for proton therapy. In: Physics in Medicine and Biology. 2012 ; Vol. 57, No. 23. pp. 7783-7797.
@article{d9844bb9b7c049eda89a2b267bf485a8,
title = "GPU-based fast Monte Carlo dose calculation for proton therapy",
abstract = "Accurate radiation dose calculation is essential for successful proton radiotherapy. Monte Carlo (MC) simulation is considered to be the most accurate method. However, the long computation time limits it from routine clinical applications. Recently, graphics processing units (GPUs) have been widely used to accelerate computationally intensive tasks in radiotherapy. We have developed a fast MC dose calculation package, gPMC, for proton dose calculation on a GPU. In gPMC, proton transport is modeled by the class II condensed history simulation scheme with a continuous slowing down approximation. Ionization, elastic and inelastic proton nucleus interactions are considered. Energy straggling and multiple scattering are modeled. Secondary electrons are not transported and their energies are locally deposited. After an inelastic nuclear interaction event, a variety of products are generated using an empirical model. Among them, charged nuclear fragments are terminated with energy locally deposited. Secondary protons are stored in a stack and transported after finishing transport of the primary protons, while secondary neutral particles are neglected. gPMC is implemented on the GPU under the CUDA platform. We have validated gPMC using the TOPAS/Geant4 MC code as the gold standard. For various cases including homogeneous and inhomogeneous phantoms as well as a patient case, good agreements between gPMC and TOPAS/Geant4 are observed. The gamma passing rate for the 2{\%}/2 mm criterion is over 98.7{\%} in the region with dose greater than 10{\%} maximum dose in all cases, excluding low-density air regions. With gPMC it takes only 6-22 s to simulate 10 million source protons to achieve ∼1{\%} relative statistical uncertainty, depending on the phantoms and energy. This is an extremely high efficiency compared to the computational time of tens of CPU hours for TOPAS/Geant4. Our fast GPU-based code can thus facilitate the routine use of MC dose calculation in proton therapy.",
author = "Xun Jia and Jan Sch{\"u}mann and Harald Paganetti and Jiang, {Steve B.}",
year = "2012",
month = "12",
day = "7",
doi = "10.1088/0031-9155/57/23/7783",
language = "English (US)",
volume = "57",
pages = "7783--7797",
journal = "Physics in Medicine and Biology",
issn = "0031-9155",
publisher = "IOP Publishing Ltd.",
number = "23",

}

TY - JOUR

T1 - GPU-based fast Monte Carlo dose calculation for proton therapy

AU - Jia, Xun

AU - Schümann, Jan

AU - Paganetti, Harald

AU - Jiang, Steve B.

PY - 2012/12/7

Y1 - 2012/12/7

N2 - Accurate radiation dose calculation is essential for successful proton radiotherapy. Monte Carlo (MC) simulation is considered to be the most accurate method. However, the long computation time limits it from routine clinical applications. Recently, graphics processing units (GPUs) have been widely used to accelerate computationally intensive tasks in radiotherapy. We have developed a fast MC dose calculation package, gPMC, for proton dose calculation on a GPU. In gPMC, proton transport is modeled by the class II condensed history simulation scheme with a continuous slowing down approximation. Ionization, elastic and inelastic proton nucleus interactions are considered. Energy straggling and multiple scattering are modeled. Secondary electrons are not transported and their energies are locally deposited. After an inelastic nuclear interaction event, a variety of products are generated using an empirical model. Among them, charged nuclear fragments are terminated with energy locally deposited. Secondary protons are stored in a stack and transported after finishing transport of the primary protons, while secondary neutral particles are neglected. gPMC is implemented on the GPU under the CUDA platform. We have validated gPMC using the TOPAS/Geant4 MC code as the gold standard. For various cases including homogeneous and inhomogeneous phantoms as well as a patient case, good agreements between gPMC and TOPAS/Geant4 are observed. The gamma passing rate for the 2%/2 mm criterion is over 98.7% in the region with dose greater than 10% maximum dose in all cases, excluding low-density air regions. With gPMC it takes only 6-22 s to simulate 10 million source protons to achieve ∼1% relative statistical uncertainty, depending on the phantoms and energy. This is an extremely high efficiency compared to the computational time of tens of CPU hours for TOPAS/Geant4. Our fast GPU-based code can thus facilitate the routine use of MC dose calculation in proton therapy.

AB - Accurate radiation dose calculation is essential for successful proton radiotherapy. Monte Carlo (MC) simulation is considered to be the most accurate method. However, the long computation time limits it from routine clinical applications. Recently, graphics processing units (GPUs) have been widely used to accelerate computationally intensive tasks in radiotherapy. We have developed a fast MC dose calculation package, gPMC, for proton dose calculation on a GPU. In gPMC, proton transport is modeled by the class II condensed history simulation scheme with a continuous slowing down approximation. Ionization, elastic and inelastic proton nucleus interactions are considered. Energy straggling and multiple scattering are modeled. Secondary electrons are not transported and their energies are locally deposited. After an inelastic nuclear interaction event, a variety of products are generated using an empirical model. Among them, charged nuclear fragments are terminated with energy locally deposited. Secondary protons are stored in a stack and transported after finishing transport of the primary protons, while secondary neutral particles are neglected. gPMC is implemented on the GPU under the CUDA platform. We have validated gPMC using the TOPAS/Geant4 MC code as the gold standard. For various cases including homogeneous and inhomogeneous phantoms as well as a patient case, good agreements between gPMC and TOPAS/Geant4 are observed. The gamma passing rate for the 2%/2 mm criterion is over 98.7% in the region with dose greater than 10% maximum dose in all cases, excluding low-density air regions. With gPMC it takes only 6-22 s to simulate 10 million source protons to achieve ∼1% relative statistical uncertainty, depending on the phantoms and energy. This is an extremely high efficiency compared to the computational time of tens of CPU hours for TOPAS/Geant4. Our fast GPU-based code can thus facilitate the routine use of MC dose calculation in proton therapy.

UR - http://www.scopus.com/inward/record.url?scp=84870380877&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84870380877&partnerID=8YFLogxK

U2 - 10.1088/0031-9155/57/23/7783

DO - 10.1088/0031-9155/57/23/7783

M3 - Article

VL - 57

SP - 7783

EP - 7797

JO - Physics in Medicine and Biology

JF - Physics in Medicine and Biology

SN - 0031-9155

IS - 23

ER -