KAHIBARO
23.4. CT-to-Material Conversion
Table of Contents
Density mapping
In voxelized geometry, each image voxel must be assigned both a material and a mass density so that GATE and Geant4 can simulate particle interactions correctly. CT images provide Hounsfield units (HU), which are related to the x ray attenuation properties of tissues, not directly to density or material. CT to material conversion is the process that bridges this gap.
The starting point is the HU value in each voxel. For a clinical CT, HU are defined so that water has HU equal to 0 and air has HU equal to −1000. In practice you never use HU directly in physics simulations. Instead, you build or load a conversion curve that maps HU to physical mass density $\rho$ in g/cm$^3$, and then from $\rho$ (or HU) to a discrete material label such as lung, soft tissue, or bone.
GATE does not invent this conversion by itself. You must provide the mapping, usually as a table or a parameterization based on scanner calibration. Often this mapping comes from phantom measurements or from the CT calibration protocol used in your clinic or experiment.
A typical approach is to define a piecewise linear curve between HU and density. For each HU interval you specify a linear relationship between HU and $\rho$. The CT processing step reads the image, converts each voxel HU to density using this curve, and stores the density as part of the voxelized volume definition. Later, physics processes will use this density together with the assigned material to compute cross sections and particle transport.
The general idea of a piecewise linear mapping can be written as
$$
\rho(HU) = a_i \cdot HU + b_i \quad \text{for} \quad HU_i \le HU < HU_{i+1}
$$
where $a_i$ and $b_i$ are constants for each HU segment $[HU_i, HU_{i+1})$.
In practice, you define a finite set of HU breakpoints and the associated densities. The simplest possible mapping uses only a few key points such as air, lung, water, and bone, and interpolates between them. More advanced mappings add several types of bone and soft tissue.
Below is an example of a conceptual HU to density table that might be used as a starting point for a generic CT phantom. Exact values must be derived from proper calibration, not copied blindly.
| HU range | Example density model (g/cm³) | Typical tissue class |
|---|---|---|
| HU ≤ −950 | 0.001 to 0.05 | Air / very low density |
| −950 to −500 | 0.05 to 0.3 | Lung |
| −500 to −100 | 0.3 to 0.9 | Mixture lung / fat |
| −100 to +100 | 0.9 to 1.1 | Water / soft tissue |
| +100 to +1000 | 1.1 to 2.5 | Bone |
| > +1000 | > 2.5 | Very dense bone / metal |
In GATE, this mapping is typically implemented through an image to material converter object, which reads a text file describing HU breakpoints and the corresponding densities and material names. When you configure this converter, you must ensure that the HU range covered by the table includes all HU values present in your CT image. Voxels with HU outside the defined range may default to an incorrect density or produce errors.
For medical simulations, it is essential that the density mapping is consistent with the physics question. For attenuation studies or CT based radiotherapy dose calculations, density errors of a few percent can already affect the results. In proton therapy especially, small density misassignments can shift the Bragg peak position noticeably.
It is also important to remember that CT images contain artifacts such as metal streaks, beam hardening, or noise. These can produce unrealistic HU values. Some workflows include pre processing steps that clip, smooth, or correct HU values before converting them to density. In a GATE based workflow, such corrections are typically performed with external image processing tools, then the corrected image is used for voxelized geometry creation.
Material ranges
Density alone is not enough. The Monte Carlo engine also needs to know which material each voxel represents, because different materials have different elemental compositions and therefore different interaction cross sections. CT to material conversion therefore assigns both a mass density and a discrete material label.
The usual strategy is to define HU (or density) intervals associated with specific materials. Each interval is called a material range. During the conversion, the code checks which interval the voxel value falls into and assigns the corresponding material. Within each interval, you may also apply the density mapping described earlier, so that two voxels in the same material class can have slightly different densities while sharing the same elemental composition.
A simple conceptual example of material ranges based on HU is shown below. The numbers are illustrative only.
| HU range | Material name | Typical use |
|---|---|---|
| HU ≤ −950 | G4_AIR | Air outside and in lung voxels with very low HU |
| −950 to −300 | Lung_ICRP | Normal lung tissue |
| −300 to −50 | Adipose_Tissue | Fat |
| −50 to +50 | Soft_Tissue | Muscle, organs, water like tissues |
| +50 to +300 | Cartilage | Dense soft tissue / light bone |
| +300 to +2000 | Cortical_Bone | Bone |
| > +2000 | High_Z_Implant | Metal implants or very dense material |
In GATE, the image to material conversion configuration usually consists of a text file or Python structure that lists, for each material, the HU (or density) bounds and the name of the Geant4 material to use. For example, you might specify that HU in [−1024, −300) map to a lung material, [−300, 100) to soft tissue, and [100, 3000) to bone. The exact materials must exist either in the Geant4 NIST database or as custom materials you define elsewhere in your simulation.
Sometimes material ranges are defined on density rather than HU. In that case, the workflow is HU to density conversion followed by density to material assignment. This can be useful when your calibration curve from HU to density is more reliable than a direct HU to material classification.
A typical two step algorithm then looks like:
- Convert HU to density using the piecewise linear mapping.
- For each voxel density, select the material whose density interval contains this value.
For dose calculation, and particularly for proton therapy or internal dosimetry, you might want more detailed material classes, such as separate definitions for different types of soft tissues or different bone compositions. These are implemented simply by refining the material ranges and adding more intervals and materials.
In practice, you must verify that the material ranges reflect the clinical or experimental reality you want to model. For instance, if bone HU vary widely in your dataset, a single broad bone range might be too crude, and you could split it into trabecular and cortical bone ranges with different densities and compositions.
Also, pay attention to overlapping or missing ranges. Every possible voxel value after HU to density conversion should fall into exactly one material range. Overlaps can cause ambiguous assignments. Gaps can leave voxels without a defined material. When you configure GATE, double check the table to ensure that the intervals cover the full value range continuously and without overlaps.
Finally, it is good practice to test the conversion by visualizing the voxelized geometry with materials displayed as different colors. This allows you to quickly see whether air, lung, soft tissue, and bone are located where you expect them in the patient or phantom image. Checking a few representative voxels by printing their HU, density, and material assignment can also help you confirm that the CT to material conversion is correctly configured.
Views: 9