ESTRO 2024 - Abstract Book

S4690

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTR0 2024

Purpose/Objective:

At Cyclotron Centre Bronowice IFJ PAN (CCB), the only proton therapy center in Poland, about 50 patients are treated every day in two gantry treatment rooms (Proteus 235, IBA). The standard patient specific quality assurance (PSQA) of a treatment plan requires between 15 and 40 minutes of beamtime per plan using 2D matrix of ionization chambers (MatriXX PT from IBA Dosimetry). Therefore, an alternative PSQA method is desirable to spare the beam time that could be used for patient treatment. The clinical application of GPU-accelerated FRED Monte Carlo (MC) code at CCB to optimize PSQA procedure will be presented, including beam model tuning and in-house machine log files interpreter development.

Material/Methods:

GPU-accelerated FRED MC code was developed at the Sapienza University of Rome for applications in particle therapy [1]. With two GPU cards (NVIDIA Titan), it reduces the treatment plan recalculation time by a factor of about 1000 compared to general-purpose MC code GATE/Geant4 using 100 CPUs. Commissioning and validation of the clinical proton beam model in FRED for two gantries was performed based on the experimental data [2]. The beam model configuration requires reference dose measurements of 10x10cm 2 monoenergetic fields at 2 cm depth in water to obtain the number of particles per monitor unit calibration factor. However, for this approach, discrepancies between FRED calculations and measurements have been observed, particularly for plans with high range and modulation. Therefore, we introduced a second, alternative beam model exploiting calibration measurements at depths between 3 and 7 cm for energies higher than 140 MeV. It has been done based on experience with commercial treatment planning system Varian Eclipse 16.1, where a similar tendency was observed and such procedure improved the accuracy of the clinical beam model used at CCB. For 46 clinical treatment fields, the 2D gamma analysis of dose distributions simulated using both models was performed, to compare the FRED calculation and MatriXX PT measurements. The measurements at proximal, isocentric, and distal depths were considered in the analysis, i.e.120 planes in total for each of the beam models. Moreover, to further exploit the calculation performance of FRED for PSQA applications, the machine logs interpreter containing beam parameters from treatment plan irradiation has been developed. This allows the use of the FRED MC not only as a secondary dose engine but also for independent dose recalculation based on the beam parameters logged during the treatment plan delivery.

Results:

The beam model adaptation with dosimetric calibration at varying depths allowed to increase the gamma index 2%/2mm passing rate for the most problematic plans by up to 5%. No significant result deterioration was observed for the planes where the passing rate was higher than 95% for the beam model calibrated at a depth of 2 cm. For all analyzed planes recalculated with a modified model, the gamma 3%/3mm passing rate is higher than 95%, while for 82% of them, it is higher than 99%. The developed log file interpreter can be used to extract the information about the beam position and size, monitor units, and nominal energy, and convert it into the RT Plan file. The estimated recalculation accuracy of these parameters are ±0.1mm for position and size, and 0.5% for the monitor units and nominal energy.