Ponente
Descripción
Introduction
Geant4 is a Monte Carlo toolkit widely used in bio-medical applications. As such, validation and performance monitoring focused on physics quantities relevant to this domain are of crucial need for the community. To respond to these needs, we developed G4-Med, a fully automated benchmarking tool and regression testing suite of Geant4 for relevant use cases in the bio-medical field. The first benchmark against reference data was carried out for version 10.5, as described in [1], from which recommendations on physics lists were proposed according to each use case.
Materials & Methods
G4-Med includes 20 testing cases, covering from basic physical quantities (attenuation coefficients, stopping powers, cross sections, dose point kernels, etc.) to tests of more realistic set-ups typical of medical physics applications (hadron therapy, brachytherapy, external electron beam therapy, mammography, etc). The latest tests incorporated, not covered in [1], aim at the validation of Geant4-DNA physics, and of fragment production for hadrontherapy and in-vivo PET, respectively. Each test runs with various reference physics lists to cover all relevant electromagnetic and hadronic physics interaction models. The tests were integrated in the geant-val platform [2] and have been executed for Geant4 development tags and public releases, including the latest public version 11.1. The calculations are compared to reference data for validation, and to previous Geant4 versions for regression testing. Further, execution times measured with various physics lists have been compared for version 11.1. Physics lists used as reference were G4EmStandardPhysics_option3 (“emstd_opt3”) and G4EmDNAPhysics_option2 (“emdna_opt2”), which are the fastest ones for condensed-history and track-structure simulations, respectively.
Results
As for electromagnetic physics, the EPICS2017 dataset was added. In the electromagnetic tests, in general, results remained stable across versions, showing “emstd_opt3”, “emstd_opt4”, “emlivermore” and “empenelope” similar accuracy when compared to experimental measurements. A significant improvement was obtained in the Brachytherapy test, where a better agreement against experimental data was observed with version 11.1, when calculating the dose at larger distances from an I-125 source. All the EM physics lists under study showed a computing performance between 1.5 and 2 times slower than the “emstd_opt3”. As for hadronic physics, charge-changing cross sections improved in Geant4 version 11.1. For the case of ion Bragg peak curves, versions 10.5 and 11.1 gave similar results. In the case of the carbon fragmentation tests, the results showed an improvement of version 11.1 against 10.5 in terms of angular distribution of the emitted light fragments. More results of the G4-Med regression testing can be accessed and downloaded from geant-val web site [2].
References
[1] P. Arce et al (2021), Med. Phys. 48 (1): 19-38.
[2] https://geant-val.cern.ch/
