KAHIBARO
34.4. Dose Calculation
Table of Contents
Dose actors
In radiation therapy simulations, dose is usually calculated with a dedicated dose actor that you attach to one or more volumes of interest. Conceptually, the actor observes every step of every particle that passes through its assigned volume and accumulates the deposited energy, then converts this to absorbed dose.
A dose actor defines a 3D grid of voxels inside its parent volume. For each voxel, it stores the total deposited energy and the mass of that voxel, and from these two quantities it computes the absorbed dose. The absorbed dose $D$ in a voxel is defined as the energy deposited per unit mass, so for a voxel with energy deposition $E$ and mass $m$ you have
$$
D = \frac{E}{m}.
$$
Key definition: Absorbed dose is the energy deposited per unit mass. In SI units,
$$D = \frac{E}{m}, \quad [D] = \text{Gy},\ [E] = \text{J},\ [m] = \text{kg}.$$
One gray (Gy) is defined as $1\ \text{Gy} = 1\ \text{J} / 1\ \text{kg}$.
In a GATE simulation you never compute these values manually. Instead, you configure the dose actor with a voxel grid, attach it to the chosen volume, and let the actor perform the calculation internally. The crucial choices you make concern where you place the grid, how fine the voxel spacing should be, and which units you request for the output.
The dose actor grid is defined in the local coordinates of its parent volume. This means you must ensure that the grid fully covers the region where you expect relevant dose. If the grid is too small, parts of the dose distribution will be truncated. If it is too large or too fine, the simulation may become unnecessarily slow or memory intensive.
Spatial resolution has a strong influence on both accuracy and computational cost. Smaller voxels capture steeper dose gradients more accurately but require more histories and more memory. Larger voxels reduce noise for a given number of histories but blur fine structures. In practice, you choose a voxel size that is comparable to or slightly smaller than the smallest clinically relevant structure or gradient that you want to resolve, and you accept that the statistical uncertainty per voxel will then depend on the number of simulated histories.
The dose actor records not only the mean dose per voxel, but can also record statistical information such as the squared dose values or the number of hits. From these, you can derive the statistical uncertainty for each voxel after the simulation. In Monte Carlo simulations, this uncertainty typically decreases in proportion to $1/\sqrt{N}$, where $N$ is the number of primary particles or histories. If you double the number of histories, the uncertainty improves by roughly a factor of $\sqrt{2}$, not by a factor of 2.
Important rule: To reduce the relative statistical uncertainty of the dose by a factor of $k$, you must increase the number of primary histories by a factor of $k^2$.
The dose actor output is usually written as a voxel image. You can select formats that are common in medical physics workflows, for example MHD or NIfTI, so that the same images can be imported into external visualization and analysis tools. It is important to store not only the dose values but also the voxel spacing, origin, and orientation, so that downstream tools can interpret the dose correctly. When you set up the actor, keep track of these parameters and of the coordinate system conventions that you use.
Although the dose actor is designed to give physical dose in gray, you must ensure that all lengths, energies, and densities in your geometry and materials are defined with correct units. Any unit inconsistency in the geometry or materials will propagate directly into the dose calculation, because both the mass of each voxel and the energy deposition depend on those definitions.
Finally, it is a good practice to attach dose actors not only to the patient or phantom, but also to reference geometries such as a water phantom, especially when you need benchmark curves like depth dose or field profiles. This allows you to validate that the configuration of physics, cuts, and source settings produces the expected dose patterns before you move to complex clinical geometries.
Patient dose
When you move from simple phantoms to realistic patient geometries, dose calculation becomes directly relevant for clinical questions. In a patient-based simulation, the dose actor is typically attached to a voxelized volume created from CT data. Each CT voxel has a size in physical space and an assigned material and density. The dose actor uses this information to compute the voxel mass and then the absorbed dose per voxel.
The basic definition of patient dose is the same as for any phantom, but the interpretation is different. Instead of thinking of anonymous voxels, you consider dose at the level of anatomical structures such as organs or target volumes. From the 3D dose grid, you integrate dose over regions of interest to obtain quantities like mean organ dose or dose to the planning target volume.
Because CT-based patient geometries contain heterogeneous tissues, the mass of each voxel varies with both its volume and its density. Given a voxel with size $v$ and density $\rho$, its mass is $m = \rho v$. For the same deposited energy $E$, a higher density leads to a smaller dose, while lower density leads to a higher dose. The dose actor takes care of this automatically, as long as each voxel has a correct material and density assignment.
Once the simulation has produced a 3D dose map, you can calculate clinically important metrics. A common way to summarize patient dose is the dose volume histogram, or DVH. For a chosen structure, such as an organ at risk or a clinical target volume, the DVH shows the volume fraction of that structure receiving at least a given dose level. To create a DVH, you must know which voxels belong to the structure. This information is usually derived from contours or segmentation masks that are aligned with the CT and, by extension, with the dose grid.
For patient analysis, always link each dose voxel to a specific anatomical region before computing metrics such as mean dose or DVHs. Without consistent structure definitions and coordinate alignment, patient dose metrics can be misleading.
Because Monte Carlo dose computations are stochastic, every voxel in the patient dose map has an associated uncertainty. In regions with high dose, where many particles contribute, the relative uncertainty is usually low. In low dose regions, or in small structures, the uncertainty can be much larger. When interpreting patient dose metrics, especially maximum or minimum dose values in small volumes, you must account for this uncertainty.
In treatment planning contexts, you often compare simulated Monte Carlo dose distributions with reference calculations, for example with a clinical planning system or with measurements in homogeneous phantoms. For patient dose, you might compare derived quantities like mean organ dose or DVH points. Large discrepancies may indicate issues in the geometry, materials, source description, or physics configuration, and should be investigated before using the simulation for scientific conclusions.
Finally, patient dose simulations are sensitive to both physics and geometry details. In the patient, many beams intersect, scatter, and interact with complex anatomy. The dose actor just reports the consequence of your chosen setup. It is therefore important to treat the dose map as the result of a complete chain of assumptions. Before you rely on patient dose results, verify that the CT to material conversion, coordinate systems, beam definitions, and physics settings are all consistent and validated in simpler reference cases.
Views: 14