ESTRO 2024 - Abstract Book

S4645

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTR0 2024

Monte Carlo (MC) simulation is considered the gold standard for dose calculation in small animals. However, due to the small voxel size (~100 ) in micro-CT images, several millions of particles are required to achieve sufficient statistical accuracy in the dose. This makes MC-based calculations very computationally expensive and time-consuming. Contrary to the clinic, the preclinical workflow requires that imaging, contouring, irradiation planning, and delivery be performed in a single session, wherein the animal is maintained under anesthesia throughout the entire process. To minimize exposure of the animal to anesthesia and to promote high throughput studies, a fast dose engine is desired. In this work, we employ a physics-based deterministic proton dose algorithm [1], which can quickly and accurately compute dose distributions in small animals.

Material/Methods:

We approximated a solution to the Linear Boltzmann transport equation (LBTE), which describes the proton phase space density. The LBTE was simplified into two partial differential equations: the one-dimensional Fokker Planck (FP) and the Fermi-Eyges (FE) equations, which were solved numerically using the discontinuous Galerkin method and analytically (for a Gaussian boundary condition) on a discretized multi-dimensional grid, respectively. To properly model the dose in the presence of heterogeneities, the beam was decomposed into 61 smaller beamlets arranged in concentric circles with optimized weights, distances, and sizes. The splitting scheme used is 1+6+6+12+12+24, which represents 1 central beam and 60 beamlets distributed on five concentric circles wherein 6 beamlets are placed on the innermost ring and 24 beamlets are placed on the outermost ring. The dose distribution for a 2-mm collimated 25 MeV proton beam delivered to the center of a mouse brain was calculated in the MOBY phantom [2] with a resolution of 0.1 × 0.1 × 0.1 mm 3 . The deterministic dose prediction was benchmarked against the TOPAS MC code for the same setup.

Results:

The comparison between the deterministic code and TOPAS dose maps is shown in figure 1. A 3D gamma analysis with dose difference/distance to agreement set to 3%/0.1 mm resulted in a 97.5% passing rate. The validity of the deterministic dose engine was also demonstrated by the good agreement with TOPAS in terms of the proton range, integral depth dose curves, and lateral profiles. Furthermore, the deterministic code was able to calculate the dose for 1×10 7 primary protons in two minutes using a single CPU core, while TOPAS took an hour using 48 parallel CPU cores.