KAHIBARO
23.5. Patient Geometry
Table of Contents
Anatomical images
In voxelized simulations you no longer describe the patient with a few simple boxes or cylinders. Instead, you use real anatomical images, most often CT, sometimes MR or synthetic phantoms, to define a 3D grid of voxels that represents the patient. Each voxel has a position in space, a size, and one or more values attached to it, such as Hounsfield units for CT. GATE uses this information, together with material mapping, to build a realistic model of patient anatomy.
An anatomical image is essentially a 3D array. In medical imaging this is commonly stored as a stack of 2D slices, but conceptually it is a single volume. For each voxel you know its indices $(i,j,k)$ in the array, and from the image metadata you know the physical spacing in each direction, such as $\Delta x$, $\Delta y$, $\Delta z$, and the origin of the image. From this, GATE can place every voxel at the correct position in the simulation coordinate system. It is important to distinguish between image coordinates, given by indices in the array, and world coordinates, given in millimeters. The conversion from indices to position is handled when you define the voxelized volume.
CT images are the most common basis for patient geometries. Each voxel contains a CT number in Hounsfield units, which reflects the linear attenuation of the tissue relative to water. This is not yet a material. In earlier chapters you map Hounsfield units to a set of predefined materials and densities. The result of this mapping is that each voxel is assigned a discrete material label, such as lung, soft tissue, or bone, and a corresponding density. The same image can also be used with multiple mapping schemes, for example one mapping optimized for dose calculation and another optimized for imaging studies.
Other anatomical images, such as MR, do not directly provide density or attenuation. In such cases, you either use an additional CT dataset of the same patient, or you rely on pseudo CT generation methods, for example assigning approximate densities to MR-based tissue classes. From the GATE point of view, as long as you provide a 3D image that is translated into a grid of materials and densities, it can be used as a patient geometry. The details of how those images were created or registered are usually handled outside of GATE.
When you create the patient geometry in GATE, you define an image-based volume that loads the voxel data from file. You specify the path to the image, voxel spacing, orientation, and the transformation to place the volume inside the world. You also specify the material look-up table that links voxel values to materials and densities. This is where errors in orientation or misaligned coordinate systems can easily appear. To avoid such problems you typically check the bounding box of the voxelized volume, compare it with known patient dimensions, and verify that the patient is centered or positioned as expected with respect to sources, beams, or detectors.
Visualization is a key step when working with anatomical images. You usually inspect the voxelized patient with GATE’s geometry visualization to confirm that the head, thorax, or other regions look anatomically reasonable and that tissues with different densities are correctly localized. For example, you should clearly see low density lungs inside the rib cage and high density cortical bone surrounding the marrow. If the image appears flipped or rotated, it indicates a mismatch between the DICOM or NIfTI coordinate system and the simulation coordinates, which you need to correct by adjusting translations or rotations.
It is also important to confirm that the material mapping preserves realistic average densities. For a simple check you may calculate the mean density in a region of interest defined on the CT, such as a slice of lung, and compare it to reference values for lung tissue. Large discrepancies usually point to a problem in the CT to material conversion, for instance an incorrect Hounsfield range or density interpolation. Only once the anatomical images have been converted into a geometrically and materially consistent voxelized phantom are you ready to use them for patient-specific simulation.
In patient geometry the image grid defines both anatomy and material composition. It is essential that voxel spacing, orientation, and CT-to-material mapping are correct, or all dose and imaging results will be physically misleading.
Patient-specific simulation
With a voxelized patient geometry in place, GATE can simulate radiation transport in a way that reflects the actual anatomy of a given patient. Patient-specific simulation means that every voxel corresponds to a real location in that patient, with realistic tissue type and density. As a result, interactions, attenuation, and dose deposition depend on the full complexity of the anatomy. This is very different from simple water phantoms or stylized mathematical phantoms, and it is essential for modern medical physics applications.
In imaging, patient-specific simulation allows you to model how a specific patient’s anatomy affects image quality. For example, in SPECT or PET you can simulate a radiotracer distribution inside the patient geometry and follow photons through heterogeneous tissues to the detectors. Lungs, soft tissue, and bone will attenuate photons differently, which influences detection probability and reconstructed activity concentration. By matching the simulated acquisition settings to those used clinically, you can generate synthetic projection data that resemble real patient images. This is useful when you want to evaluate reconstruction algorithms, perform quantitative corrections, or validate new imaging protocols before applying them to patients.
In radiation therapy, patient-specific simulation focuses on dose. Here you typically define a beam configuration that corresponds to a treatment plan, for example from an external beam radiotherapy or proton therapy system, and deliver it to the voxelized patient. Dose actors record the absorbed dose in each voxel, yielding a 3D dose distribution in units of gray. Because the geometry includes all tissues and organs, the simulation can reveal how dose is deposited not only in the target volume, but also in surrounding organs at risk. You can then compute dose volume histograms or extract dose profiles and compare them with clinical treatment planning system predictions.
One powerful application is radionuclide therapy dosimetry. In this case you have both anatomical CT images for geometry and functional images, such as SPECT or PET, that provide the spatial distribution of activity in the patient. The activity map is usually co-registered to the CT and represented as a voxelized source distribution. GATE then propagates particles from these sources through the patient anatomy and records dose in all voxels. This enables personalized absorbed dose estimates for organs and tumors. You can vary the administered activity or radionuclide and study how this affects organ dose and therapeutic index.
Patient-specific simulation often requires careful alignment of multiple datasets. The patient geometry from CT, the activity map from SPECT or PET, and any defined treatment beams must all share a consistent coordinate system. Misregistration will cause sources or beams to miss the intended anatomical structures. In practice, you ensure consistent origin and orientation when you convert images to voxelized volumes and verify alignment through visualization, for instance by showing source distributions on top of the CT-based geometry.
Because patient-specific simulations can be computationally demanding, especially for large CT volumes and fine voxel grids, you usually pay attention to performance. You may limit the region of interest to the clinically relevant volume, define coarser dose grids where high resolution is not required, or use region based production cuts to speed up calculations in areas far from the beam or activity. Multithreading and high performance computing resources are often used for full patient plans or radionuclide therapies, where many particles are needed to reach acceptable statistical uncertainty.
The interpretation of results from patient-specific simulations must always consider statistical noise. Each voxel’s dose or detector count has an associated Monte Carlo uncertainty that decreases with the number of simulated histories. For clinical decision making or comparison with measured data you generally aim for uncertainties below a few percent in important regions, which may require several simulation runs or many primary particles. Modern analysis tools allow you to compute these uncertainties per voxel and incorporate them in subsequent quantitative evaluations.
In summary, patient-specific simulation connects anatomical images, material mapping, and detailed Monte Carlo transport into a single workflow. Once the geometry is correctly built, you can use the same patient model to test different imaging protocols, beam arrangements, radionuclides, or shielding concepts. This is a central strength of GATE in medical physics, since it allows you to explore and optimize procedures in silico under realistic anatomical conditions, while maintaining full control and transparency over the underlying physical assumptions and numerical settings.
Patient-specific results are only meaningful if all spatial registrations are correct and statistical uncertainties are quantified. Always verify geometry alignment and report uncertainties when interpreting dose or imaging outputs for an individual patient.
Views: 10