Optimization of the Patient Phantom and GEANT4 Parameters in the Ion Beam Therapy Planning System

Abstract

Purpose: To develop and describe the architecture of a physical module for a Dose-Anatomical Planning System (DAPS) for ion beam therapy based on the GEANT4 platform, and to optimize its computational efficiency for clinical application under limited computing resources. Material and methods: Simulation of the interaction processes of carbon ion beams with biological tissues was performed using the Monte Carlo method with the GEANT4 toolkit, version 10.7.4. The Schneider method, implemented as a separate software component, was used to convert computed tomography (CT) data into physical characteristics of the medium (density, chemical composition). Computational experiments were conducted on a real patient phantom to assess the influence of the discretization step of Hounsfield units (1, 2, 5, 10, 20 HU) on the accuracy of dose distribution calculation, with particular attention to the Bragg peak region. Additionally, the contribution of secondary particles (neutrons, electrons, positrons, gamma rays) and nuclear fragments to the total energy deposition was analyzed. The effectiveness of applying different GEANT4 physics models (full QGSP-BIC-HP-EMY vs. electromagnetic EMY) at various planning stages was investigated. Results: It was experimentally shown that discretization steps of 1 HU and 10 HU yields statistically indistinguishable results for the energy deposition profile in the Bragg peak region (difference is within Monte Carlo statistical uncertainty, <1%). A step of 20 HU leads to a systematic error in the determination of the Bragg peak position of up to 2 mm, which is unacceptable for clinical planning. It was found that the contribution of secondary particles (neutrons, electrons, positrons, gamma rays) to the total dose is less than 1%, and their exclusion from simulation speeds up GEANT4 calculations by approximately 1.9 times without statistically significant loss of accuracy. It was demonstrated that while hadronic processes must be retained for accurate carbon ion fragmentation modeling, they can be disabled during the preliminary stage of energy map calculation (for bolus design, for example), accelerating computations by a factor 5–6. A method for converting DICOM coordinate systems to the GEANT4 laboratory coordinate system using Tait-Bryan angles was developed and verified, along with an algorithm for constructing an optimized voxel grid oriented along the beam axes. Conclusion: A physical module for DAPS has been developed, optimized for speed and accuracy. The proposed solutions for CT data discretization (10 HU step), construction of an optimized voxel grid, and simplification of the physical model (disabling neutron, electron/positron, and photon transport) allow reducing the full ion beam therapy planning cycle from several days to several hours while maintaining the required clinical accuracy (±1 mm).