Session Item

Friday
May 07
14:15 - 15:30
21st century brachytherapy: is it available, affordable and relevant?
0210
Symposium
00:00 - 00:00
A GPU Monte Carlo to support clinical routine in a compact spot scanning proton therapy system
PO-1417

Abstract

A GPU Monte Carlo to support clinical routine in a compact spot scanning proton therapy system
Authors: Gajewski , Jan(1);Schiavi , Angelo(2);Krah , Nils(3);Patera , Vincenzo(2);Vilches-Freixas , Gloria(4);Martens , Jonathan(4);Unipan , Mirko(4);Rucinski , Antoni(1)*;Nijsten , Bas(4);Bosmans , Geert(4);Rinaldi , Ilaria(4)[ilaria.rinaldi@maastro.nl];
(1)Institute of Nuclear Physics, Pan, Krakow, Poland;(2)University of Rome, Sapienza, Rome, Italy;(3)University of Lyon/CNRS CREATIS UMR5220- Centre Léon Bérard, Creatis UMR5220, Lyon, France;(4)Maastricht Radiation Oncology MAASTRO clinic, Radiation Oncology, Maastricht, The Netherlands;
Show Affiliations
Purpose or Objective

The purpose of this work is to implement an independent fast Monte Carlo (MC) dose calculation tool, Fred, to complement our clinical Treatment Planning System (TPS). Current efforts concentrate on making the MC code compatible with our proton delivery system as well as creating the necessary programming interfaces with existing software. Fred achieves high accuracy and computation speed by using physics models optimized for radiotherapy, stream-lined implementation, and extensive use of GPU technology for parallelization.


Material and Methods

We have implemented the beam model of our proton beam and are currently validating it against measured commissioning data and data calculated with the clinical TPS. This step required significant changes in the core of Fred because of the peculiarities of our proton machine. The beam exits the accelerator with a pristine energy of around 230 MeV and then travels through the dynamically extendable nozzle of the device. The nozzle contains the beam monitor system, the range modulation system (RMS), and the multi-leaf collimator system named adaptive aperture (AA). We use a single model to parametrize the longitudinal (energy, energy spread) and transverse (beam shape) phase space of the non-degraded beam in the default nozzle position. The plates in the RMS are moved in and out of the simulation geometry dynamically by Fred. In certain machine configurations, the nozzle overlaps with the patient CT voxel geometry. We extended the MC code to properly handle such situations. We currently focus on implementing the AA system which moves the leaves partially into the beam path. We are finalizing an alternative conversion method from Hounsfield Units (HU) to relative stopping power (RSP) in which the user provides a lookup table between HU and mass density. Finally, we are moving Fred from a stand-alone application into a library which can be controlled dynamically from other software or through a scripting interface such as python.

Results

Figure 1 shows a sketch of the beam nozzle configuration during a breast treatment. The beam line is shown in red and the RMS in blue. Fred tracks the particles until the exit window (dark blue) of the nozzle and then continues in the overlapping CT geometry (light blue). Since the RMS is explicitly simulated, diverging tracks from nuclear interactions can be seen. 

Figure 2 shows the dose distributions for a breast case simulated with Fred (left), with the clinical TPS (middle), and comparison of the dose profiles. Note that these results do not yet use the new adapted implementation of the CT number to RSP conversion. The calculation time of this plan was 2.3 min with overall 10 million primaries simulated.



Conclusion

We have implemented a fast independent MC in our proton facility. We will soon start using Fred for plan verification based on machine and log files and daily (on-the-fly) dose recalculations.