Encrypted login | home

Program Information

Development of a Hybrid Stochastic-Deterministic Method for Dose Calculation in Radiotherapy

R Hayward

R Hayward*, F Rahnema, Georgia Institute of Technology, Atlanta, GA

SU-C-500-6 Sunday 1:00PM - 1:55PM Room: 500 Ballroom

Purpose: To develop a hybrid stochastic-deterministic method, COMET-PE, for dose calculation in external beam photon radiotherapy.

Methods: To calculate dose, COMET-PE solves the coupled Boltzmann Transport Equations for photons and electrons. The method uses a deterministic iteration to compose response functions that are pre-computed using Monte Carlo. Thus, COMET-PE takes advantage of Monte Carlo physics without incurring the computational costs typically required for statistical convergence. Dose distributions are calculated for a heterogeneous benchmark problem using both COMET-PE and DOSXYZnrc (Monte Carlo) methods. The benchmark consists of a CT-based lung phantom, composed of air, lung, soft tissue, and bone, irradiated by a 2 cm x 2 cm photon field. The 6 MV source spectrum comes from Monte Carlo simulation of a Varian Clinac 2100. The COMET-PE solution is computed at resolution of 1 mm (27,648,000 voxels) before being reduced to a resolution of 5 mm (221,184 voxels) for compatibility with the DOSXYZnrc reference solution.

Results: The agreement between dose distributions calculated with COMET-PE and Monte Carlo is excellent. Of voxels receiving greater than 10% of the maximum dose, 98.73% pass the 2% (point-wise relative difference) or 2 mm (distance-to-agreement) criterion and 99.38% pass the 3% / 3 mm criterion. Localized discrepancies are observed at the beam corners; these are caused by the difficulty of using a continuous representation for the discontinuous primary fluence. Most of the failures, however, occur where the beam exits the phantom and where Monte Carlo uncertainties are the highest. The COMET-PE calculation is over 10 times faster than the Monte Carlo reference solution.

Conclusion: The COMET-PE method calculates dose with accuracy comparable to Monte Carlo while using only a fraction of the time and providing a solution with orders of magnitude more detail.

Contact Email: