Validation of Nuclear Reaction Models to Simulate Proton Therapy Range Verification Using Prompt Gamma-Rays
J Verburg*, H Shih, J Seco, Massachusetts General Hospital and Harvard Medical School, Boston, MAMO-A-213AB-6 Monday 8:00:00 AM - 9:55:00 AM Room: 213AB
Purpose: The impact of nuclear reaction model differences on simulation of prompt gamma-ray imaging for proton therapy range verification was assessed. Four nuclear reaction models were used to simulate gamma emission in proton beams, and were validated against experimental cross-sections.
Methods: Proton-induced nuclear reactions on carbon, oxygen, nitrogen and calcium were investigated with the Monte Carlo toolkits GEANT4 9.5 and MCNPX 2.7, and the dedicated nuclear reaction codes TALYS 1.4 and EMPIRE 3.1. Absolute cross-sections of discrete prompt gamma lines and the total gamma production were obtained for the 1-200 MeV incident proton energy range. They were compared to 34 discrete line measurements reported in literature. Using these cross-sections, we analyzed the gamma production along the path of proton beams passing through various tissues.
Results: The differences in absolute discrete line cross-sections as predicted by the models ranged from almost zero to an order of magnitude, depending on the gamma line and incident proton energy. Overall, the dedicated nuclear reaction codes provided a better fit to most experimental excitation functions. For a 150 MeV proton beam stopping in soft tissue, these differences amount to a variation by a factor of 4 of the gamma emission around the Bragg peak location. The maximum of gamma production near the end of proton range differed by 7 mm, and the change of the 50% emission fall-off position was 4 mm.
Conclusions: There is a clear need for improvement of nuclear reaction models to accurately simulate proton range verification using prompt gamma-rays. Current simulation codes show large uncertainties in both the total gamma yield and the correlation of gamma emission with the proton Bragg peak. GEANT4 and MCNPX in particular appear to have limited predictive power.