Updated computational performance evaluation of ICRP mesh-type reference computational phantoms using PHITS, Geant4, MCNP6.3, and EGSnrc

Abstract

The International Commission on Radiological Protection (ICRP) has adopted mesh-type reference computational phantoms (MRCPs) for the calculation of new reference dose coefficients following the forthcoming general recommendations. In the present study, the computational performance of adult and pediatric MRCPs was systematically evaluated using recently updated Monte Carlo radiation transport codes, including PHITS (version 3.31), Geant4 (version 11.01.p02), MCNP6.3, and EGSnrc (version 2025a). Performance evaluation included initialisation time, memory usage, computation time, and multi-threading efficiency for photon, electron, and neutron transport under selected external exposure scenarios. Among the evaluated codes, EGSnrc exhibited the longest initialisation time, approximately 8-9 times longer than PHITS. Nevertheless, the absolute time remained within a few minutes. In terms of memory usage, PHITS and EGSnrc required the least memory (<2 GB), whereas Geant4 required the largest memory (approximately 13-14 GB). For computation time, EGSnrc showed the fastest performance for photons and electrons with energies <= 10 MeV, while PHITS demonstrated the shortest computation time for neutrons. MCNP6.3 consistently exhibited longer computation times compared with the other codes. When compared with simulations using the existing ICRP voxel-type reference computational phantoms (VRCPs), PHITS and Geant4 showed comparable or shorter computation times for mesh phantoms. EGSnrc exhibited comparable or shorter computation times for pediatric mesh phantoms, whereas the opposite trend was observed for adult phantoms. In contrast, MCNP6.3 consistently showed longer computation times for the MRCPs than for the VRCPs. Despite its relatively longer absolute computation time, MCNP6.3 demonstrated the best multi-threading performance among the evaluated codes, showing near-linear scaling of computation speed with increasing number of threads. The results of this study provide practical guidance for users in selecting an optimal Monte Carlo code for dose calculations using ICRP MRCPs.