A Comparison of DPM and VMC++ Mont Carlo codes applied to heterogeneous media Jing Cui 1, Scott Davidson 2, Patricia Lindsay 2, Dave Followill 2, Issam El Naqa 1, and Joe Deasy 1 (1) Washington University, St. Louis, MO (2) UT M.D. Anderson Cancer Center, Houston, TX
Acknowledgment The authors would like to thank Jan Seuntjens and Frank Verhaegen for performing BEAM calculations and for their helpful discussions.
Outline Motivation Differences of DPM and VMC++ Case studies Water phantom Air cavity heterogeneity Thorax phantom IMRT Head and Neck cases Conclusions and Future work
Motivation Monte Carlo (MC) technique is physically sound for dose calculation. Takes a large amount of CPU time, particular for the complete MC codes, such as EGS4, ITS, MCNP, or PENELOPE.. Fast MC algorithms: DPM (J. Sempau, 2000) (Dose Planning Method) and VMC++ (Voxel Monte Carlo) (I. Kawrakow, 2001). For fast MC codes, the simplifications of the underlying physics, variance reduction, and random number generation may not be the same. Implementation issues are complex and therefore testing and quality assurance is important. Focus: direct comparison
Differences between VMC++ and DPM Underlying Physics Implementations Availability Speed
VMC++ VMC++ (I. Kawrakow, 2001) is the class II condensed history MC simulation of coupled electron-photon transport The electron transport procedure is taken from VMC (Fippel, 1999). Uses small angle approximation Re-uses electron histories and STOPS variance reduction technique.
DPM DPM (J. Sempau, 2000) employs the standard condensed history model for electron treats large energy transfer collisions in an analog sense Uses the continuous slowing down approximation (CSDA) to model small loss collisions
Implementation Difference in other aspects VMC++ is implemented in C++ DPM is written in Fortran. Availability VMC++: the executable could be acquired from the author DPM: open source. Speed (D. W. Rogers, 2006), VMC++ is between 50 and 100 times faster than a corresponding EGSnrc; agrees within 1% DPM has not obtained the same speed as the VMC++ but is much faster than EGSnrc.
VMC++ validations J. Gardner, J. Siebers and I. Kawrakow (2007), VMC++ vs. DOSXYZnrc Water phantom, bone-lung-bone phantom, patient plans For patient plans, most severe difference is: 1% of Dmax for 2% of voxels with D>0.5*Dmax.
Comparisons Standard open-field water phantom calculation on various field sizes Air cavity heterogeneity Anthropomorphic thorax phantom, to deliver a 5-beam conformal plan IMRT Head and Neck plans
Water Phantom Comparison Using a point source, 6MV spectrum Left: PDD of 5x5cm 2 for BEAM, DPM, and VMC++ Right: Dose profiles of 5x5cm 2 for DPM, and VMC++, at depth 5,10, and 20 cm
Air Cavity Comparison SSD = 100cm, FS = 0.5cm x 0.5cm Air cavity = 1cm 3 5cm AIR 1cm Water PDD
Phantom Studies Heart Tumor (TLD) R L Lung Cord Lung S. Davidson (AAPM 2007): MO-D-AUD-1, TU-FF-A1-4
Phantom Studies (con t) Convert everything to water: DPM and VMC++ agreed within 0.88% ([STD of the difference larger than 5%Dmax]/Dmax);
Phantom Studies (con t) Lateral Anterior-posterior Superior-inferior Convert everything to water; Dose profiles along the isocenter
Phantom Studies (con t) Lateral Anterior-posterior Superior-inferior Heterogeneous media; Dose profiles
Phantom Studies (con t) Using water-equivalent materials: the results of DPM and VMC++ agreed within 0.88% ([STD of the difference larger than 5%Dmax]/Dmax); Use whatever density values obtained from the CT scan; this resulted in an agreement to within STD of 0.95%.
IMRT Head and Neck Plans Patient plans: 3 cases Metrics: PTV D95, brainstem Dmax, RT parotid gland Dmean, Cord D2.
Head and Neck (con t) Typical dose profiles (case 2)
Head and Neck (con t) Case 2, DVH
Head and Neck (con t) Diff = (DPM-VMC)/VMC*100
Conclusions [1/2] Carefully designed tests were conducted for DPM and VMC++ in water phantom and heterogeneous phantoms. They behaved almost identically in water. The calculation showed agreement within 1% for 1cm 3 air cavity case, except at the interface of the water and air, where the large dose gradient occurred. In the lung phantom study, DPM and VMC++ agreed within 1% RMS, where DPM appeared to be more sensitive to material changes.
Conclusions [2/2] However, in H&N IMRT examples, results so far show some significant differences between DPM and VMC++. Even in metrics that are insensitive to noise (e.g., parotid mean dose) Further investigation is warranted More examples More treatment sites Possible implementation problem in IMRT simulations?
Thank you!