Simulation of the head of an accelerator: Calculation optimization and statistical comparison methods

Introduction: The group Developments and Application for Medicine of the LPSC, in collaboration with the Public Hospital of Grenoble, is developing the TraDeRa detector (Transparent Detector for Radiotherapy). This device includes an ionization chambers matrix and provides 2D signal maps for any irradiation field, upstream to the patient. Converting the detector's signal into dose in the patient is a challenge. To do so, Monte Carlo simulation is a powerful tool, and will give accurate results if correctly parameterized. The work described here focuses on the optimization of the treatment head simulation (especially on the target, the source of X-rays) and the determination of the correct nominal energy and radial distribution of electrons on the target. In order to calibrate the detector, one must reach a perfect agreement between the observables from the simulation and from the beam of our reference accelerator. Methods: The interactions of the electrons in the target are numerous and time consuming to simulate. A study was conducted on both parameters of different particles transport methods and variance reduction in PENELOPE Monte Carlo code, which remains a reference for electron transport for the considered energies [1]. We have optimized a set of parameters to keep a reasonable computation time without biasing the physical observables. The determination of the initial characteristics of the electron beam (nominal energy/radial distribution of electrons on the target) is done by trial and error process. Several simulations are performed at various energies and radial electron distributions. The dose deposited in a simulated water phantom is compared with the depth-dose curves and dose profiles acquired under irradiation. We propose two efficient methods of comparison, the calculation of the Kolmogorov-Smirnov test and an original extension with more sensitivity. Results: The set of optimized parameters provides an overall increase on the simulation efficiency of nearly 300%. The useful secondary particles generation rate (stored in the PSF) was also increased by about 250%. Determining the characteristics of the electron accelerator of our reference beam is ongoing, we currently have simulated three energies around 6 MeV. In the appendix are presented the results of two comparative tests for these three energies. These tests including measurement uncertainties are robust and will accurately lead to the beam nominal energy with only five simulated energy sets. To date, the simulation has been running for six months on the IN2P3 computing center. Conclusions: The developed statistical tests are robust and will allow us to accurately determine the characteristics of the beam by comparing the simulated and the measured depth-dose curves and dose profiles. We will then be able to validate the reference accelerator simulation and calibrate the detector in terms of dose. This will allow us to reconstruct the 3D dose matrix in water from upstream information collected on a TraDeRa simulated model.