Exploitation of HPC for realistic pediatric dosimetry simulations using GATE

Introduction.

The objective of this research concerns the development of a realistic dosimetric database based on the estimation of absorbed dose rate at different organs using Monte Carlo (MC) simulations. Several radiopharmaceuticals with clinically specified biodistributions and pediatric applications in Nuclear Medicine (NM) procedures were studied in the context of this project. High Performance Computing (HPC) resources utilized to achieve low statistical uncertainties.

Materials & Methods

The absorbed dose rate assessment for this study was based on GATE_v9.1 Monte Carlo toolkit. For our purpose 20 computational phantoms (15 XCAT and 5 IT’IS ), with different anatomical characteristics and varying age, gender, mass and height were simulated using 5x10⁸ primary events. Four different radiopharmaceuticals (99mTc-MDP, 123I-mIBG, 131I-MIBG, and 153Sm-EDTMP) were considered, together with their specified activity distribution. The radiation activity of each organ derived from clinical data, at different time points (eg. 4 time points for 99m-Tc was assumed: T=0h, T=1.42h, T=4.11h, T=20.2h), and used for the calculation of dose rate at the respective time points. Simulations were executed on YOTTA HPC centre, to accelerate their execution, having 112 parallel jobs running. YOTTA HPC consist of compute nodes that each one includes 28-Core Intel Broadwell CPUs and 512GB of memory. The “Dose actor” concept was utilized to measure the deposited energy per voxel, producing 112 dose maps for each case. They were then merged to a single file for minimizing uncertainty, during the calculation of the absorbed dose rate per organ.

Results

All simulations were performed in YOTTA HPC reducing the time needed for their execution, more than ~80 times (~9.5h required for the simulation in the PC in contrast to ~7.5mins in the HPC). Dose rate for 31 different organs of each pediatric phantom were calculated (case of Tc-99m MDP). In addition, the relative percentage uncertainty in the dose values per organ was measured, fluctuating between 0.05% and 2.7%, with a median value of 0.11%, indicating that our MC environment reduces the statistical uncertainty. Furthermore, there are fluctuations at the dose rate, for the same organ on different phantoms (up to ~71% difference at male phantoms and up to ~65% at female phantoms).

Discussion & Conclusions

In this abstract the case of 99mTc-MDP is presented and indicative values of dose rates are presented. The main goal of our study concerns the implementation of simulations for a variety of commonly used radiopharmateuticals in NM applications, to develop large database for internal dosimetry in pediatric examinations. Next steps of our objectives include the use of Artificial Intelligence techniques, utilizing ML/DL algorithms, trained on our simulated database, to predict the internal absorbed dose for any new pediatric patient.