X-ray Computed Tomography Glossary

Absorption contrast imaging

Absorption contrast imaging uses the difference in the absorption rate of the sample. The absorption rate primarily depends on the density and thickness of an object. It is the most commonly used X-ray imaging mechanism.

Aliasing artifact

When the X-ray CT (computed tomography) data do not include enough projections, the under-sampling or aliasing artifacts occur. The aliasing artifacts appear as radial lines in X-ray CT slices that are perpendicular to the sample rotation axis. The ideal number of projections is the FOV (field of view) size divided by the voxel size, but half to a third of this number usually suffices.

Watch a webinar clip about artifacts from X-ray Computed Tomography for Materials and Life Science - Introduction.

 Read more about aliasing artifact

Artifact

Artifacts of X-ray CT (computed tomography) are some features, patterns, or any change in gray level observed in X-ray CT images that do not represent the true features of the scanned object. Artifacts arise from the imperfection of the measurement procedure. Typical artifacts include beam hardening artifacts, ring artifacts, and aliasing artifacts.

 Watch a webinar clip about artifacts from X-ray Computed Tomography for Materials and Life Science - Introduction.

 Read "How to Recognize CT Artifacts and Minimize Them During Data Collection"

Beam hardening artifact

X-rays used in X-ray CT (computed tomography) measurements are often polychromatic and have an energy distribution. Because the X-ray absorption rate depends on the X-ray energy as well as the material density, polychromatic X-rays' energy distribution shifts as they go through an object. Higher-energy X-rays (harder X-rays) survive better than lower-energy X-rays (softer X-rays). As a result, X-ray energy shifts towards the higher side, or X-rays "harden" after going through an object. This phenomenon is called beam hardening. Meanwhile, the reconstruction calculation to generate a 3D image from 2D projections assumes monochromatic radiation with only one value of X-ray energy. This gap between the experiment and the calculation causes the beam hardening artifacts. They cause shading of thick and uniform material or streaks radiating from high-density or highly-absorbing material. When there are extremely highly absorbing areas, they can completely stop the X-rays and create dark spots and streaks in the X-ray CT images. This is called the photon starvation artifact.

 Watch a webinar clip about beam hardening artifact from X-ray Computed Tomography for Materials and Life Science - Introduction.

 Read "What Is Beam Hardening in CT?"

 Read "How to Reduce Beam Hardening Artifacts in CT"

Binarization

Binarization is a conversion of a grayscale image, such as an X-ray CT (computed tomography) image, into a black-and-white binary image. The simplest way to binarize an image is by thresholding. There are a variety of algorithms to determine the threshold, such as Otsu and maximum entropy. Binarization is a rather primitive way of image segmentation, but it is a simple way to make the segmentation analysis operator-independent by applying the same binarization algorithm to multiple data sets.

Cone beam geometry

Cone beam geometry is one of the most commonly used X-ray CT (computed tomography) geometry. The X-ray beam starts from a micron-size X-ray source and diverges like a cone to irradiate the scanned object and reaches the detector. The angular divergence of the X-ray beam is used for the image magnification.

Watch a webinar clip about cone beam geometry from X-ray Computed Tomography for Materials and Life Science - Introduction.

Deep learning segmentation

Deep learning is a branch of artificial intelligence and imitates a way humans learn a specific task and improve accuracy through training. We can use deep learning to segment X-ray CT (computed tomography) images as humans would. The process involves defining classes (phases) and showing examples of how humans would classify each voxel to the "machine" to train it. After the training session(s), the machine learning program can carry out the segmentation work for the rest of the image. While machine learning algorithms such as random forest still use the local gray level for segmentation criteria, deep learning can consider the shape, size, complex morphology, and location of each class (phase) and is significantly more sophisticated and capable than machine learning.

 Watch a webinar clip about deep learning segmentation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

 Watch a webinar, "Ask the Expert - Deep Learning Image Segmentation - Practical Knowhow," by Dr. Mike Marsh.

 Watch Dragonfly Daily Webinar series episode 17 - Image segmentation with deep learning.

To learn how deep learning works, watch Deep Learning Chapter 1 by Ian Goodfellow and Neural Network by 3blue1brown.

Diffraction contrast imaging

Crystalline grains in a polycrystalline bulk sample can satisfy a diffraction condition as the sample rotates during an X-ray CT (computed tomography) measurement. This creates distinct diffraction contrasts. You can reconstruct a three-dimensional orientation map, in a similar manner as in the EBSD (electron backscatter diffraction), by extracting and sorting these diffraction contrasts into groups related to individual grains. This technique is called diffraction contrast tomography (DCT). Most work has been done using synchrotron radiation, but laboratory X-ray sources can be used, too.

Learn how it works from the paper by Ludwig et al. (2008) J. Appl. Cryst., 41, 302-309.

See an example of diffraction contrast imaging by Syha et al. (2012) Scr. Mater., 66(1), 1-4.

Field of view (FOV)

A field of view (FOV) is the volume covered in an X-ray CT (computed tomography) scan image. An FOV can be either larger or smaller than the sample size. When the FOV is larger than the sample, the entire sample is imaged. When the FOV is smaller than the sample, only part of the sample is imaged. In general, the FOV is limited to the area of the sample that can be projected on the X-ray area detector. (Except for the offset and helical scans.) Because the number of pixels on the detector is usually about 1K x 1K  ~ 3K x 3K pixels, the voxel resolution of a CT image is limited to FOV/1000 ~ FOV/3000.

Watch a webinar clip about the field of view from X-ray Computed Tomography for Materials and Life Science - Introduction.

Focus correction

To obtain a correct reconstructed X-ray CT (computed tomography) image, the X-ray focus, the center of the sample rotation, and the center of the detector need to be on a straight line. X-ray CT scanners are mechanically aligned to achieve this condition. However, there often is a micron-level misalignment, and it needs to be corrected during the reconstruction process. This correction is called a focus correction or center correction or focus correction. There are other parameters that can be corrected during the reconstruction process, such as X-ray focus drift, detector tilt, and sample movement, but a center correction is the most important one and often needs to be done for high-resolution measurements.

 Watch a webinar clip about the focus correction (center correction) from X-ray Computed Tomography for Materials and Life Science - Foams and Composites Applications.

Gantry geometry

Most industrial X-ray CT (computed tomography) scanners rotate an object (a sample) while the X-ray source and detector are fixed to collect multiple 2D projections and reconstruct a 3D image. This geometry provides flexibility in the sample size and resolution. In contrast, most medical CT scanners keep an object (a patient) stationary and rotate the X-ray source and detector around it. This helps to keep the patient stable and comfortable during the scan. The latter is called a gantry system or a gantry geometry. The gantry geometry is also used for 4D, in-situ, and general high-speed measurements.

Geometric dimensioning and tolerancing (GD&T) analysis

Geometric Dimensioning and Tolerancing (GD&T) is a system for defining part dimensions and tolerances in an explicit way. It allows us to define the dimensions and the functions of each part by specifying the tolerances. GD&T analysis is used to evaluate parts and determine if the sizes, shapes, and locations of each component of a part are within the defined tolerances. The dimensions and tolerances are defined in the design process. The evaluation is done by comparing the measured values to the required dimensions and tolerances. X-ray CT (computed tomography) is a commonly used measurement technique for GD&T analysis, along with coordinate measuring machines (CMMs) and optical scanners.

 Watch a webinar clip about GD&T from X-ray Computed Tomography for Materials and Life Science - Metrology Applications.

Watch a webinar, "Ask the Expert - Voxels to Mesh Data to Simulation – How to Use CT for Metrology and Simulations by Roger Wende.

To learn the basics of GD&T, watch GD&T - Part 1: Basic Set-up Procedure and Part 2: Gauges, Dimensioning and Errors by Infinity MFG.

Histographic segmentation

Histographic segmentation is an advanced gray-level thresholding segmentation technique. You can combine two images and set threshold values using both images. This technique is useful for compensating for shading or cupping artifacts.

 Watch a webinar clip about histographic segmentation from X-ray Computed Tomography for Materials and Life Science - Geology Applications.

 Watch Dragonfly Daily Webinar series episode 26 - Histographic segmentation.

ISO-50 surface determination

ISO-50 surface determination defines an object surface by examining a local gray level profile at the interface and identifying the point where the gray level reaches 50% when the object gray level and air gray level are defined as 100% and 0%, respectively. This technique considers the partial volume artifacts and can achieve 1/10 of a voxel resolution in object surface determination with sophisticated adaptive corrections.

 Watch a webinar clip about ISO-50 surface determination from X-ray Computed Tomography for Materials and Life Science - Metrology Applications.

Watch a webinar, "Ask the Expert - Voxels to Mesh Data to Simulation – How to Use CT for Metrology and Simulations by Roger Wende.

Machine learning segmentation

Machine learning is a branch of artificial intelligence and imitates a way humans learn a specific task and improve accuracy through training. Machine learning includes deep learning, but the term "machine learning" is often used to refer to regression algorithms such as random forest regression. We can use machine learning to segment X-ray CT (computed tomography) images as humans would. The process involves defining classes (phases) and showing examples of how humans would classify each voxel to the "machine" to train it. After the training session(s), the machine learning program can carry out the segmentation work for the rest of the image. Although it is still a gray-level-based segmentation, it is remarkably robust against noise.

 Watch a webinar clip about machine learning segmentation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

 Watch a mini tutorial about machine learning segmentation.

Magnification factor

The magnification factor of X-ray CT (computed tomography) geometry is defined as a scale on a detector divided by the actual size. For example, if a 100-micron pixel on the detector corresponds to 10 microns in a scanned object, the magnification factor is 100/10 = 10. When a cone beam geometry is used, the magnification factor is calculated as SDD (source-to-detector distance) divided by SOD (source-to-object distance). This is called geometric magnification. When an optical lens is used for optical magnification, as in the case of parallel beam geometry, the lens's magnification factor is added. 

Nominal versus actual comparison

You can scan an object using X-ray CT (computed tomography) and create a surface mesh to describe its shape. This surface mesh can be compared to a CAD (computer-aided design) file to measure the deviation of the actual part from the original design. This is called nominal versus actual comparison.

 Watch a webinar clip about nominal versus actual comparison from X-ray Computed Tomography for Materials and Life Science - Metrology Analysis.

Object separation

Image analysis is often used to measure the size distribution of particles, pores, etc. The first step is to separate the measured objects from the background by image segmentation. If the objects are not touching, each cluster of segmented voxels that are connected can be treated as one object, and its size is measured. If all or some of the objects are touching, they need to be separated as individual entities. This process is called object separation. When the expected shape and size of the objects can be assumed, they can be used to separate touching objects. When the shape and size cannot be defined, a watershed transformation can be applied to separate the touching objects.

Watch a webinar clip about object separation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

Watch the workshop "Refining Segmentation Using ImageJ"

Parallel beam geometry

Parallel beam geometry uses an optical lens for magnification. A scintillator placed right after the scanned object converts X-rays into visible light so that an optical lens can magnify the image before the light reaches the detector. This geometry is immune to blurring due to X-ray focus size and drifting and is suitable for high-resolution (submicron) X-ray CT (computed tomography) measurements.

 Watch a workshop clip about parallel beam geometry from Virtual Workshop - High-resolution CT Data Collection Techniques.

Partial volume effect

X-ray CT (computed tomography) images have a finite resolution like any digital image. Because of the limited resolution, the surface of an object or an interface between air and solid, for example, often appears blurred. The voxels at the interface can include both solid and air and exhibit a gray level, which is a combination of the two phases. Depending on the ratio of the two phases, the gray level of these interface voxels varies. It is an artifact, and this artificial intermediate gray level is called the partial volume effect.

Watch a webinar clip about partial volume effect from X-ray Computed Tomography for Materials and Life Science - Geology Applications.

Phase contrast imaging

As X-rays pass through materials, their intensities decrease by absorption, and their phase shifts by refraction simultaneously. Phase contrast imaging uses this phase change instead of the absorption rate. This technique is significantly more sensitive to density variations in light-absorbing materials, such as tissues, than absorption contrast imaging. Because the phase shift cannot be directly measured by an X-ray detector, it is transformed into variations in intensity by using devices, such as the Fresnel zone plates. This phase information to intensity transformation reduces the X-ray intensity, so phase contrast imaging requires a high-intensity X-ray source or a long exposure time. Phase contrast imaging should not be confused with the phase retrieval method, although the latter also uses X-ray refraction.

 Watch a webinar clip about phase contrast imaging from X-ray Computed Tomography for Materials and Life Science - Life Science Applications.

Phase retrieval

As X-rays pass through materials, their intensities decrease by absorption, and their directions slightly change due to refraction. X-rays change their directions the most when they run parallel to the density interface, and more X-rays land on one side of the interface while leaving a "gap" of X-ray intensity on the other side. This change in direction generates a pair of dark and light gray levels on a projection image highlighting interfaces of two materials with different densities. You can use this dark and light line pair to guess what the associated phase change was and improve the contrast of X-ray CT (computed tomography) images. This technique is called phase retrieval.

Watch a webinar clip about phase retrieval imaging from X-ray Computed Tomography for Materials and Life Science - Life Science Applications.

 Read a Rigaku Journal article about the phase retrieval method.

Pore network analysis

Pore network analysis models porous space to describe the flow properties of porous materials such as rocks. The porous space can be segmented using X-ray CT (computed tomography) and modeled as a combination of pores (large porous space) and throats (narrow porous space) to generate a digital pore network. The pore network model can be used to calculate connectivity, tortuosity, permeability, etc., and simulate flow properties.

 Watch a webinar clip about pore network analysis from X-ray Computed Tomography for Materials and Life Science - Geology Applications.

Check out Pore-network Modeling Framework in Python - OpenPNM.

Reconstruction

X-ray CT (computed tomography) data is collected in the form of two-dimensional projections. So these 2D projections need to be reconstructed to obtain a three-dimensional image of an object. This process is called reconstruction. The most widely used algorithm is the filtered back-projection algorithm. Although this is a relatively simple and robust algorithm, it suffers from various artifacts. More advanced techniques, such as iterative reconstruction and deep learning, to overcome artifacts or improve overall image quality have been studied in recent years.

 Watch a Mini Tutorial series - X-ray CT explained with ImageJ - Reconstruction.

Watch the workshop "Demystifying Reconstruction Using ImageJ".

Read "How Does CT Reconstruction Work?".

Region of interest (ROI)

A region of interest (ROI) is a collection of voxels within a data set defined for a particular purpose. This can be the head region of a mouse or void space in a porous material, for example. An ROI can be defined by a geometric shape such as a box or sphere, or a collection of voxels that have a certain range of gray levels.

Resolution

The resolution of an X-ray CT (computed tomography) image usually represents how small of an object can be distinguished. It also occasionally means how sharp or how well-defined an interface of two materials is. There are several ways to define resolution. The commonly used ones are "voxel resolution" and "spatial resolution." The voxel resolution is the size of the voxels. The spatial resolution is the size of the smallest feature that can be observed in an image. The spatial resolution is generally the voxel resolution times two or greater because you need at least two voxels to see that there is a feature (Shannon Nyquist Sampling theorem). The resolution is affected by the X-ray focus size, detector pixel size, detector point spread function, lens and beam divergence magnification factor, and overall system stability.

Watch a webinar clip about resolution from X-ray Computed Tomography for Materials and Life Science - Foams and Composites Applications.

 Watch a Mini Tutorial series - X-ray CT explained with ImageJ - Resolution.

Read a blog article, "How to improve the resolution of X-ray CT images."

Ring artifact

When there is non-uniform sensitivity in the X-ray CT (computed tomography) measurement system, it causes ring artifacts. The ring artifacts appear as concentric rings around the sample rotation axis. It occurs when detector sensitivity and background level are not corrected properly, or there is an X-ray absorbing dust on the X-ray window, filter, or detector surface, changing the observed X-ray intensity level.

Watch a webinar clip about artifacts from X-ray Computed Tomography for Materials and Life Science - Introduction.

Read more about ring artifact

Segmentation

Segmentation is the classification or "labeling" of voxels in X-ray CT (computed tomography) images. For example, voxels are labeled as air, polymer, carbon fiber, etc., and classified as separate regions of interest (ROIs). This is the first step of quantitative analysis of X-ray CT data. Simple thresholding based on the gray levels of each voxel is the most basic and conventional segmentation technique. Although it is simple and fast, thresholding is limited and often does not work well when the noise level is high, the contrast is low, there are artifacts, or some phases have complex morphology. Machine learning or deep learning is recommended for challenging segmentation tasks.

 Watch a webinar clip about thresholding segmentation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

Signal-to-noise ratio (SNR)

SNR stands for a signal-to-noise ratio and is defined as [mean signal value / standard deviation]. The signal value is the gray value in X-ray CT (computed tomography) images, which is related to the number of X-ray photons converted into a signal by the detector. The standard deviation is the fluctuation of this signal. Therefore, you can roughly interpret SNR as [meaningful X-ray signal count/noise count].

Read "How to Improve the Signal-to-noise Ratio of X-ray CT Images"

Source-to-detector distance (SDD)

Source-to-detector distance (SDD) is the distance between the X-ray source and the detector sensor.

Source-to-object distance (SOD)

Source-to-object distance (SOD) is the distance between the X-ray source and the scanned object.

Surface mesh

A surface mesh is a representation of a surface defined by a collection of vertices, edges, and faces. It is also called a polygon mesh. An ROI (region of interest) defined as a collection of voxels or surface of an object defined by the ISO-50 method can be converted into a surface mesh for further analyses or 3D printing.

Thresholding segmentation

You can set a range of gray levels by defining thresholds and give voxels that have a gray level within that range a certain label. For example, you may define that all voxels with a lower than gray value 10/256 are air, and the rest are solid. It is the most basic and conventional segmentation technique. Although it is simple and fast, thresholding is limited and often does not work well when the noise level is high, the contrast is low, there are artifacts, or some phases have complex morphology. Machine learning or deep learning is recommended for challenging segmentation tasks.

 Watch a webinar clip about thresholding segmentation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

Volume mesh

A volume mesh or volumetric mesh is a representation of a volume defined as a collection of tetrahedrons or other elements. Volume meshes are widely used in finite element analysis for mechanical, fluid, thermal, electrical, and other simulation studies. An ROI (region of interest) defined as a collection of voxels or surface of an object defined by the ISO-50 method can be converted into a volume mesh.

Watch a webinar, "Ask the Expert - Voxels to Mesh Data to Simulation – How to Use CT for Metrology and Simulations by Roger Wende.

Voxel

A voxel is a unit of graphic information. It is a 3D version of a pixel, and the word comes from "volume" and "pixel." In X-ray CT (computed tomography) images, a voxel is defined by its X, Y, and Z coordinates and one gray value.

Watershed transformation

A watershed is a transformation of a grayscale image. It treats a grayscale image as a topographic map, with the brightness representing the height. The transformation works as a geological watershed or drainage divide and separates adjacent drainage basins or ROIs. In X-ray CT (computed tomography) image analysis, a watershed transformation is often used to separate objects. The segmented objects defined as an ROI are first converted into a distance map which assigns each voxel a distance between the voxel and the closest object-background border. The inverse of this distance map serves as the topographic map, and a watershed operation can divide the adjacent drainage basins or individual objects.

Watch a webinar clip about watershed transformation from X-ray Computed Tomography for Materials and Life Science - Data Analysis.

Watch the workshop "Refining Segmentation Using ImageJ"

X-ray absorption

When X-rays interact with an object, they mainly interact with the electrons. X-rays are elastically or inelastically scattered, or absorbed through the photoelectric effect or pair production. What matters in absorption contrast X-ray CT (computed tomography) imaging is the level of total absorption for given X-ray energy and the absorption rate and thickness of the imaged object. The general rule is the lower the X-ray energy is, and the higher the object's electron density is, the higher the absorption rate is. The thicker the object is, the more X-rays are absorbed. The absorption rate of a material per unit thickness is expressed as mass attenuation coefficients. These values have been thoroughly examined and are available through the database created by the National Institute of Technology and Standard. A more practical tool for X-ray CT measurements is available from CSRRI (Center for synchrotron radiation research and instrumentation) at the Illinois Institute of Technology. Their Mucal Periodic Table is a convenient calculator to find a good combination of X-ray energy, material, and thickness.

 Watch a webinar clip about X-ray absorption from X-ray Computed Tomography for Materials and Life Science - Introduction.

X-ray computed tomography (CT, X-ray CT, XCT)

X-ray computed tomography (CT) is an X-ray imaging technique that can non-destructively scan the density distribution of an object in 3D. The most common form of X-ray CT is absorption contrast imaging. Multiple 2D projections of an object are collected using a micrometer-sized X-ray source and a 2D detector from different angles. A 3D image is generated from those 2D projections using reconstruction calculation. The first commercial X-ray CT scanner was developed in the early '70s by Allan M. Cormack and Godfrey N. Hounsfield. They shared The Nobel Prize in Physiology in 1979 for their work. The technique has been widely used in the medical field since then, but it has become a popular tool for materials and life science and metrology in recent years.

Watch a webinar clip about X-ray computed tomography from X-ray Computed Tomography for Materials and Life Science - Introduction.

 Read a brief introduction to X-ray computed tomography.

Read a comprehensive article, "What Is Micro-CT?"-CT