KAHIBARO
34.6. Dose Validation
Table of Contents
Depth-dose curves
In radiation therapy simulation, depth-dose curves are one of the most important tools for validating that your GATE model represents a real treatment beam correctly. A depth-dose curve describes how the absorbed dose changes as a function of depth in a medium, usually water, along the central axis of the beam. Clinically, the reference is the measured percent depth dose (PDD) curve or the tissue phantom ratio (TPR / TMR) curve acquired with a dosimetry phantom. In GATE, you use dose actors in a water phantom to reproduce these measurements and compare the simulated and measured curves.
To obtain a depth-dose curve in GATE, you typically define a water phantom that extends beyond the maximum range of the beam, align the beam along one axis, and configure a dose actor that records dose in small voxels along the beam axis. The voxel size along depth should be chosen fine enough to resolve rapid gradients, such as the build-up region and dose falloff. Transverse voxel dimensions can be larger if you are only interested in the central axis, but they must still represent a uniform region around the beam center.
Once the simulation is complete, you extract the dose along the central axis line. This can be done by selecting the voxel indices that lie on the beam axis from the dose image output. You then obtain two arrays: depth $z$ and dose $D(z)$. For validation, you usually normalize the dose to a reference point, for example the dose at the depth of maximum dose $d_\text{max}$, and compare with the measured percent depth dose curve.
You can represent a PDD curve as
$$
\text{PDD}(z) = 100 \times \frac{D(z)}{D(d_\text{ref})},
$$
where $d_\text{ref}$ is typically $d_\text{max}$, or a standardized reference depth.
For validating your GATE simulation, important aspects include the location of $d_\text{max}$, the shape of the build-up region, the slope of the dose falloff, and the overall agreement with reference data. Differences may arise from beam energy spectrum, source size and divergence, or the physics configuration. You should adjust only physically meaningful simulation parameters, not artificially tune results.
A useful way to present your validation is to plot both simulated and measured PDD on the same graph and then compute the relative difference as a function of depth.
You can define a relative dose difference $\Delta(z)$ as
$$
\Delta(z) = 100 \times \frac{D_\text{sim}(z) - D_\text{meas}(z)}{D_\text{meas}(d_\text{ref})}.
$$
Another powerful method is gamma analysis, which compares dose distributions in both space and dose. For depth-dose validation along a line you can use a one-dimensional gamma test with chosen distance and dose criteria, for example 2 mm and 2 percent, as used in many clinical tolerances. For a beginner, it is enough to understand that gamma analysis combines spatial and dose differences into a single passing rate.
When you interpret differences between simulated and measured depth-dose curves, you should pay attention to specific regions. In the build-up region, uncertainties in electron transport and surface geometry can cause deviations. Around $d_\text{max}$, collimator configuration and source modeling are important. In the tail region at large depths, the beam energy spectrum and the treatment head materials strongly influence dose. Understanding which part of the model dominates which region helps guide corrections to the geometry or beam model.
You should also consider statistical uncertainty. Each voxel’s simulated dose has a Monte Carlo uncertainty that decreases with the number of simulated histories. For reliable comparison, the relative uncertainty in the dose along the central axis, especially around the reference depth and high-gradient regions, should be small, for example below a few percent.
Repeated simulations with different random seeds can help verify that the observed differences are systematic and not just random fluctuations. When the depth-dose curve from GATE matches the measured reference within clinically acceptable criteria, you gain confidence that your model of the beam, phantom, and physics is suitable for further dose calculations in more complex geometries, such as patient CT datasets.
For depth-dose validation, always normalize simulated and measured curves consistently, use sufficiently small voxels along depth, and ensure that Monte Carlo statistical uncertainty is low enough before judging agreement.
Lateral profiles
Lateral dose profiles describe how dose varies across the beam perpendicular to the central axis at a fixed depth. While depth-dose curves validate beam penetration and energy spectrum, lateral profiles validate beam shape, field size, and penumbra. In GATE, you obtain lateral profiles by analyzing the dose distribution in a transverse plane within the phantom at a chosen depth and comparing with measured profiles from water tank scans.
To compute a lateral profile in GATE, you select a plane at a given depth $z_0$ and extract dose values along one transverse coordinate, for example $x$, while holding $y$ fixed at the beam center. This yields $D(x, z_0)$ along a line. For a square or rectangular field, you usually analyze in both in-plane and cross-plane directions. As for depth-dose, you must choose the voxel size carefully. Transverse voxel dimensions should be small enough to resolve the penumbra, which is the region of rapid dose falloff near the field edge. If the penumbra is around a few millimeters, voxel sizes must be significantly smaller than this width.
You frequently normalize lateral profiles either to the central axis dose at the same depth, $D(0, z_0)$, or to the maximum dose in the profile. A normalized lateral profile at depth $z_0$ can be written as
$$
P(x; z_0) = 100 \times \frac{D(x, z_0)}{D(0, z_0)}.
$$
You then compare $P_\text{sim}(x; z_0)$ from GATE with $P_\text{meas}(x; z_0)$ from measurements.
Several key features are important in validation. The flatness and symmetry of the central region verify that the field is uniform and centered. The field size, often defined by the 50 percent dose points, should match the planned field width. The penumbra width, commonly measured between 80 percent and 20 percent dose levels, tests the collimation, source size, and scattering behavior in the simulation. Any systematic broadening or narrowing of the penumbra suggests issues with the source spot size, jaw or multileaf collimator modeling, or phantom voxelization.
For a more quantitative comparison, you can compute point-by-point differences or perform gamma analysis in the transverse direction. A typical criterion might be 2 percent dose difference and 2 mm distance-to-agreement. The gamma pass rate across the profile provides a single metric of agreement. However, for an introductory validation, visual comparison of normalized profiles combined with examination of key metrics, such as field size at 50 percent and penumbra widths, is often sufficient.
When your lateral profiles show discrepancies, you should analyze their pattern. If the profile is too narrow, the source spot may be too small or the collimator openings may be under-modeled. If the penumbra is too wide, voxel size might be too coarse or the source model too large. If there is an asymmetry, you might have misaligned geometry, incorrect beam angle, or non symmetric collimator settings. Each of these issues can be investigated by carefully checking geometry, materials, and source definition in the GATE script.
It is also important to consider statistical noise. Because lateral profiles involve many voxels at relatively low dose levels near the field edge, you may need a larger number of primary histories to obtain smooth curves with acceptable uncertainty. You can reduce noise further by averaging across multiple adjacent voxels transversely or by exploiting symmetry, for example by mirroring left and right sides, if the beam is symmetric.
Lateral profiles at multiple depths, for example near $d_\text{max}$, mid-depth, and near the field edge in depth, give a more complete picture. Agreement across all these planes increases confidence that your GATE beam model is robust and that it will produce realistic dose distributions in more complex scenarios, such as full treatment plans in patient geometries.
For lateral profile validation, always use voxel sizes fine enough to resolve the penumbra, normalize profiles consistently, and check field size, flatness, symmetry, and penumbra width before accepting the simulation as validated.
Views: 9