Applying graphics processor units to Monte Carlo dose calculation in radiation therapy

M. Reza Bakhtiari, H Malhotra, Md Jones, Vipin Chaudhary, J.l. Walters, D NazarethJournal of Medical Physics 2010 · DOI

TL;DR

This paper explores using Graphics Processing Units (GPUs) to accelerate Monte Carlo (MC) simulations for radiation therapy dose calculations. By leveraging the massive parallel processing power of GPUs, the authors demonstrate a significant speedup compared to traditional CPUs, making complex MC simulations more feasible for clinical use.

What Problem Does This Paper Address?

Radiation therapy relies on accurate dose calculations to effectively treat cancer while minimizing damage to healthy tissue. Monte Carlo (MC) simulations are considered the gold standard for this, as they precisely model the physical interactions of radiation particles with matter. However, MC simulations are notoriously computationally intensive, requiring millions of particle histories to achieve sufficient statistical accuracy. This often translates to hours of computation time on standard CPUs, hindering their routine use in clinical treatment planning. The authors aim to address this computational bottleneck by investigating the use of GPUs, which have emerged as powerful parallel processors, to significantly speed up these MC calculations.

Methods

The core of the paper involves implementing a simplified Monte Carlo photon transport simulation and running it on both a standard CPU and a GPU. The simulation models a photon beam incident on a semi-infinite, uniform medium, tracking individual photons and their interactions (scattering and absorption) based on known coefficients. The key innovation here is the utilization of NVIDIA's CUDA (Compute Unified Device Architecture) programming language to harness the GPU's parallel processing capabilities. Each photon's history is simulated using a separate thread on the GPU, allowing for massive parallel execution. The program was then implemented on a standard PC using C++ for comparison.

Datasets

Not discussed in the paper. The authors created a simplified simulation for demonstration purposes rather than using pre-existing datasets.

Key Insights and Ideas

  • GPUs as a Supercomputing Resource for Clinics: The paper highlights that GPUs offer a low-cost, portable way for clinics to access computational power comparable to large supercomputing clusters, democratizing access to advanced simulation techniques.
  • Massive Parallelism is Key: The inherent nature of MC simulations, where each particle's path can be tracked independently, aligns perfectly with the massively parallel architecture of GPUs. This is the fundamental reason for the observed speedup.
  • CUDA as an Enabler: The availability of GPU programming languages like CUDA is crucial for unlocking this computational potential, making it accessible to researchers and developers.

Key Results

The authors found that their GPU-implemented MC simulation was approximately 150 times faster than the CPU implementation for a large number of histories. This means that in the same amount of time, the GPU could simulate 150 times more photon histories, leading to significantly improved statistical accuracy or drastically reduced computation times.

Limitations and Future Work

The authors acknowledge that the simulation presented is a simplified model and not a physically-accurate MC code for clinical use. Future work would involve implementing a more comprehensive MC code that accounts for secondary particle production and more complex geometries. The ultimate goal is to enable the early adoption of MC for routine clinical planning, which requires further development and validation of these GPU-accelerated simulations.

Summary updated

Abstract

We investigate the potential in using of using a graphics processor unit (GPU) for Monte-Carlo (MC)-based radiation dose calculations. The percent depth dose (PDD) of photons in a medium with known absorption and scattering coefficients is computed using a MC simulation running on both a standard CPU and a GPU. We demonstrate that the GPU's capability for massive parallel processing provides a significant acceleration in the MC calculation, and offers a significant advantage for distributed stochastic simulations on a single computer. Harnessing this potential of GPUs will help in the early adoption of MC for routine planning in a clinical environment.
BibTeX
@article{bakhtiari2010applying,
  author = {M. Reza Bakhtiari and H Malhotra and MD Jones and Vipin Chaudhary and J.L. Walters and D Nazareth},
  title = {{Applying graphics processor units to Monte Carlo dose calculation in radiation therapy}},
  journal = {Journal of Medical Physics},
  year = {2010},
  doi = {10.4103/0971-6203.62198},
}
Figure