Assessing the robustness of proton therapy plans using fast Monte Carlo tools
(2015) BHPA
Files
No attached file found for this publication.
Details
- Authors
- Abstract
- I.Introduction and purpose Intensity modulated proton therapy (IMPT) generally improves healthy tissue sparing during the treatment planning while keeping a target coverage similar to the most advanced photon techniques. However proton therapy is very sensitive to the proton range uncertainty introduced by the patient setup, organ motions, limited imaging and approximations used for the dose calculation. Depending on the heterogeneity level of the geometry, this range uncertainty can reach up to 4.6%+1.2mm (13.2mm for 200 MeV protons) with typical analytical dose calculation methods. This study addresses this issue in two steps. Firstly the range uncertainty is reduced to 2.4%+1.2mm by using Monte Carlo (MC) dose calculation. Secondly the remaining uncertainty is controlled by verifying the robustness of the treatment plan. The impact of these uncertainties on the treatment quality can be assessed by simulating various scenarios of uncertainties. MC simulation is the key element of this project and offers some advantages in comparison to analytical dose calculation methods. MC techniques offer optimal accuracy for dose calculation by simulating particle interactions inside the patient geometry. It also allows for integrating several types of uncertainties (for instance, breathing) within one single calculation, reducing the total number of simulations. Moreover the computation time of MC simulation, which is often too long for a clinical use, can be greatly reduced by employing a very fast and optimized implementation on modern computation architectures such as GPUs or Xeon Phi coprocessors. II.Material and methods The new MCsquare software created for this study implements our optimized physical models and algorithms on the Intel Xeon Phi coprocessor to accelerate the Monte Carlo computation. The Xeon Phi architecture is composed of 60 independent cores, each containing a small vector calculation unit. As a result, the MC simulation of one dose distribution with MCsquare is completed in less than one minute achieving a gain factor up to 700 in comparison the Geant4. The robustness of the treatment plan to uncertainties is verified by simulating various scenarios considering the different sources of uncertainties. Firstly, the daily patient setup uncertainty is modeled by shifting the patient CT image. This rigid translation can be split into two components: a systematic one which represents the mean position error of the patient for all fractions and a random one which represents the position variations around the mean error from one fraction to another. The second type of uncertainties concerns errors induced by the restricted patient data extracted from the CT image. Some inaccuracy directly come from the imaging limitations (noise, resolution, …). However the main uncertainty comes from the conversion of the CT Hounsfield units into physical quantities such as densities and stopping powers. This can be taken into account by applying a uniform scaling of the patient densities, although more detailed models might also used in a MC framework. To assess the impact of these uncertainties, multiple scenarios are therefore created, each one corresponding to one possible realization of the treatment. Multiple scenarios of systematic translation and density scaling are thus simulated. In addition, the CT image is randomly shifted several times for each scenario according to our random error model of the patient position. The experiment is performed on a cylindrical water phantom of 20 cm radius. The target (CTV) surrounds a circular organ (OAR) as shown in the figure. The treatment plan is firstly optimized employing the traditional PTV method by expanding the CTV with a 2.5 mm isotropic margin. The second treatment plan is obtained with a robust optimization for the CTV and OAR considering uncertainties of 3%+2.5mm. Both plans are then recomputed for all scenarios using MCsquare. III.Results and discussion All uncertainty scenarios are shown in a dose-volume histogram (DVH). The robustness of the treatment plan is then easily verified by looking to the DVH curve deviations with respect to the nominal plan (red curve). The robust plan shows small deviations compared to the traditional PTV plan. Consequently, the CTV coverage of the PTV plan is no longer guaranteed in case of uncertainty. More complex uncertainty models such as organ motions will be integrated in the future. IV.Conclusions Fast and accurate MC algorithms are employed to reduce and verify the impact of range uncertainties on the treatment quality. The treatment plan robustness is verified by simulating various uncertainty scenarios. In proton therapy, the traditional PTV margins are no longer sufficient to guarantee the homogeneous coverage of the CTV in presence of uncertainties.
- Affiliations
Citations
Souris, K., Lee, J., & Sterpin, E. (2015). Assessing the robustness of proton therapy plans using fast Monte Carlo tools. BHPA. https://hdl.handle.net/2078.5/229219