Efficient voxel navigation for proton therapy dose calculation in TOPAS and Geant4.

Area of Science:

• Medical Physics
• Computational Physics
• Radiotherapy Physics

Background:

• Monte Carlo (MC) particle transport codes are essential for accurate patient dose calculations.
• Efficient navigation through complex geometries, such as voxelized computed tomography (CT) data, is critical for MC simulation performance.
• Various voxel geometry parameterizations exist within MC simulation packages like Geant4, each with potential trade-offs in speed and memory usage.

Purpose of the Study:

• To evaluate and compare the efficiency of different Geant4 voxel geometry parameterizations for MC-based patient dose calculations.
• To identify the optimal parameterization for handling patient-specific CT data in terms of simulation speed and memory consumption.
• To assess the impact of parameterization choices on dose distribution accuracy.

Main Methods:

• Simulations were performed using the TOPAS (Tool for Particle Simulations) MC package, built upon Geant4.
• Four Geant4 parameterizations were investigated: G4VPVParameterisation, G4VNestedParameterisation, G4PhantomParameterisation (with and without boundary skipping), and an in-house MGHParameterization.
• Runtime and memory usage were analyzed across three patient CT datasets (head and neck, liver, prostate), including a modified dataset with uniform water density.
• Dose distributions were compared against a reference for accuracy assessment.

Main Results:

• G4VPVParameterisation showed significantly slower simulation speeds or substantially increased memory requirements.
• G4PhantomParameterisation without boundary skipping and MGHParameterization demonstrated comparable simulation speeds to G4VNestedParameterisation.
• G4PhantomParameterisation with boundary skipping showed a slight runtime improvement but resulted in dose calculation discrepancies for one patient case.
• Maximum memory usage was generally consistent across parameterizations, except for the high-memory option of G4VPVParameterisation.

Conclusions:

• G4VNestedParameterisation emerged as the preferred parameterization for the studied patient geometries and treatment plans due to its favorable balance of computational speed and memory efficiency.
• The choice of parameterization significantly impacts MC simulation performance and, in some cases, dose calculation accuracy.
• Further optimization of navigation algorithms is crucial for advancing MC-based radiotherapy dose calculations.