X-ray computed tomography, often shortened to CT or XCT, uses X-ray measurements taken from different directions to reconstruct cross-sectional information about an object. It can reveal internal features that a single radiograph cannot separate cleanly along the X-ray path.
The gold standard is not merely an attractive three-dimensional rendering. It is a traceable chain from source and detector measurements to a reconstructed quantity, with image limitations and decision uncertainty stated before a diagnosis or dimension is asserted.
eduKateSG · X-ray Computed Tomography
Find your next learning step
Calculate attenuation from transmitted photons, understand why many views matter and distinguish reconstructed appearance from validated measurement.
Full chapter index · Practice and transfer · Education root · How X Works
Full chapter index
Define the measurement · Chapters 1–2
Calculate attenuation · Chapters 3–4
Check the image · Chapters 5–7
Interpret and transfer · Chapters 8–10
Read across the ecosystem · Chapters 11–12
CHAPTER 1 OF 12 · Define the measurement
1. CT changes the question from a shadow to an interior
A radiograph records a two-dimensional pattern of transmitted radiation. Many different arrangements of material along one ray can produce similar overall attenuation, so overlapping structures are hard to separate from one view.
CT acquires many projection measurements from different angles and uses a reconstruction model to estimate spatial differences inside the object. Medical CT often displays slices of a patient’s body; industrial micro-CT may examine pores, internal features or dimensional properties of manufactured components.
Define the intended output before scanning: a visible flaw, a material contrast, a quantitative dimension, a density-related estimate or an image for clinical interpretation. These objectives have different sensitivity, reconstruction and validation requirements.
A voxel is a reconstructed volume element, not an independently observed cube inside the specimen. The measurement has been inferred from many X-ray paths, detector responses and model choices.
Where people are scanned, acquisition and exposure decisions belong to qualified clinical professionals. This article concerns the physical reasoning and measurement evidence, not a patient protocol.
CHAPTER 2 OF 12 · Define the measurement
2. From X-ray source to projections to reconstruction
A source emits X-rays across a beam, the object attenuates and scatters them, and detectors measure outgoing signals as source, detector or object rotates. Geometry and exposure conditions determine which paths were sampled.
In simplified monoenergetic transmission physics, an unscattered ray through a material follows I = I₀ exp(−∫μ(s)ds), where I₀ is the incident intensity, μ is a linear attenuation coefficient and the integral follows the ray path.
Computing −ln(I/I₀) turns this idealized exponential expression into a path integral. Many such integrals at different angles form the projection data used in tomographic reconstruction.
Actual instruments may have polychromatic sources, scatter, detector nonlinearities, dark current, saturation, motion and imperfect geometry. A raw pixel value is not automatically an ideal line integral; calibration and corrections determine how closely the mathematical model applies.
The FDA and NIBIB explanations of medical CT describe rotating source–detector arrangements and cross-sectional image reconstruction. Industrial systems share the central tomographic principle while differing in objectives, geometry and operational constraints.
CHAPTER 3 OF 12 · Calculate attenuation
3. Worked example: infer an ideal line attenuation
Assume a monochromatic, scatter-free ray and an accurately measured flat-field incident count I₀ = 10,000. A detector measures I = 3,679 counts after transmission through a uniform sample. Counts here stand in for comparable ideal intensity measurements.
The measured transmission is T = 3,679/10,000 = 0.3679. The log attenuation is A = −ln(T) ≈ 1.000. A value of A is dimensionless because it is the logarithm of an intensity ratio.
If the physical path length is L = 2.0 cm and attenuation is uniform, μ = A/L ≈ 0.500 cm⁻¹. The units belong to μ, not to the logarithm.
This is a model calculation, not an attenuation coefficient for an identified material. Without a known spectrum, calibration, material and beam path, one observed count ratio cannot be promoted to a tissue composition or certified density.
Repeat with I = 6,065 while retaining I₀ = 10,000 and L = 2 cm. Since −ln(0.6065) ≈ 0.5, the implied μ is approximately 0.25 cm⁻¹ under the same ideal assumptions.
CHAPTER 4 OF 12 · Calculate attenuation
4. Worked example: why one view cannot locate two internal regions
Imagine one ray crossing two 1 cm segments. Segment A has an ideal attenuation coefficient of 0.20 cm⁻¹ and segment B has 0.50 cm⁻¹. Their total log attenuation is 0.20 × 1 + 0.50 × 1 = 0.70.
For I₀ = 10,000, the transmitted intensity is approximately 10,000 exp(−0.70) ≈ 4,966. This ray does not identify which segment contributed the larger attenuation.
Two different segments each at μ = 0.35 cm⁻¹ give the same total 0.70 along that path. A single projection therefore cannot distinguish these internal distributions even with perfect noise-free data.
Other directions create new equations for the unknown spatial distribution. Enough informative views, appropriate geometry and a suitable reconstruction assumption can constrain an interior estimate, although limited angles and measurement noise still leave uncertainty.
That is the conceptual difference between tomography and simply enhancing a radiograph. Reconstruction depends on collecting enough independent information, not on making a shadow visually sharper.
CHAPTER 5 OF 12 · Check the image
5. Beam spectrum, scatter and detector physics change the data
Clinical and many laboratory X-ray sources are polychromatic. Lower-energy photons are often attenuated more strongly than higher-energy ones, so the surviving beam spectrum changes as it passes through an object. A constant monoenergetic μ model can then misrepresent thick or heterogeneous paths.
Beam hardening can produce shading or streak artefacts when corrections are inadequate. Scatter can add signal that did not follow the assumed primary ray path; photon starvation can make a highly attenuating direction noisy.
Detector response includes resolution, gain, offset, efficiency and possible lag. Flat-field and dark-field correction help, but calibration drift or a damaged channel can create reconstruction artefacts that resemble specimen features.
Describe source settings, filters, object material range, detector conditions and correction assumptions when reporting a quantitative result. Two scanners can form visually similar images with different error structures.
Artefact suppression algorithms can improve a display while altering measured edges. Always distinguish an image-processing improvement from independent validation of the underlying object’s geometry.
CHAPTER 6 OF 12 · Check the image
6. Reconstruction algorithms are part of the measurement
Filtered backprojection and iterative reconstruction are families of approaches for estimating an interior from many projections. Algorithms differ in assumptions, regularisation, noise behaviour, computational cost and their handling of incomplete data.
A reconstruction with fewer streaks can still be biased. Regularisation tends to favour some image properties, such as smoothness or sparse variation, and may blur small features that are important to an inspection.
Report the reconstruction method and important settings. In research, preserve raw projections as well as the reconstructed volume when practical, because changing the method can change the inferred boundary or defect count.
Voxel pitch is the spacing of reconstructed samples, not a guarantee of spatial resolution. Blur, focal-spot size, detector sampling, motion, noise and reconstruction filtering determine how faithfully fine details are resolved.
Use an appropriate resolution or task-performance test rather than treating the smallest displayed voxel as the smallest reliably measurable feature.
A dimensional measurement requires a scale that has been checked and an uncertainty statement that reflects the method. Industrial CT errors can arise from rotation-stage motion, source–detector distances, detector alignment, thermal changes and threshold-dependent surface location.
NIST reports on ASME B89.4.23-2020 discuss performance evaluations for dimensional industrial XCT systems. Well-chosen calibrated artefacts are designed to expose relevant sources of error rather than merely producing a pleasing image.
NIST’s 2026 review of CT phantoms describes reference features for spatial resolution and defect-detection evaluation. A phantom can test one class of performance without proving every material or geometry will be measured accurately.
Reverification should include relevant times, orientations, feature sizes and object materials. A single central sphere or straight rod cannot necessarily establish accuracy for an angled internal channel with strong attenuation contrast.
Traceability requires an unbroken and documented measurement comparison with uncertainties, not simply a certificate placed beside the scanner.
CHAPTER 8 OF 12 · Interpret and transfer
8. Segmentation converts grey values into consequential decisions
A reconstructed image contains intensity values influenced by attenuation and reconstruction. Segmentation assigns regions or boundaries to categories such as material, air or suspected flaw. That decision is a modelling step, not another independent X-ray reading.
Consider an idealized region of 1,000 equally sized voxels, with 30 classified as void. The reported voxel fraction is 30/1,000 = 3%. It is a segmentation-dependent result, not automatically the true physical porosity.
Changing the threshold can move the boundary, especially where one voxel contains a mixture of materials. Partial-volume effects, noise and uneven beam hardening can alter the apparent size and count of small voids.
Validation may require repeated scans, alternative thresholds, a representative reference specimen or an independent measurement technique. Report the uncertainty of a final geometry or fraction in terms meaningful to the decision.
A high-resolution rendering can hide a large systematic boundary bias. Distinguish detection (does a flaw appear?) from quantification (how large is it, with what uncertainty?).
CHAPTER 9 OF 12 · Interpret and transfer
9. Clinical, materials and archaeological uses have different safeguards
Medical CT balances diagnostic benefit with radiation exposure, clinical indication, patient factors and reconstruction choices. FDA guidance treats justification and optimization as important because CT involves ionizing radiation. A useful physical example must not be mistaken for advice to order, refuse or modify a clinical examination.
Industrial XCT can examine internal channels, additive-manufacturing defects and dimensional features without destructively sectioning the object. Yet material composition, path length, scan time and accepted measurement uncertainty govern suitability.
Research CT of fossils, rocks or soils may reveal interior structures but still require a distinction between naturally different attenuation and genuine composition or fabric. A bright region does not uniquely identify a chemical phase.
Cross-method comparisons help when they answer a specified uncertainty: spectroscopy may identify surface chemistry; diffraction tests crystalline arrangement; radiography and CT describe attenuation geometry at different spatial resolutions.
The evidential boundary remains the same: state what the image directly constrains and what would require additional analysis.
CHAPTER 10 OF 12 · Interpret and transfer
10. Practice, FAQs and transfer between objects
Practice A: Calculate ideal log attenuation for I₀ = 5,000 and I = 2,500. Since T = 0.5, A = −ln(0.5) ≈ 0.693. Name two real instrument effects that could disturb this calculation.
Practice B: Imagine a 10.00 mm reference dimension is estimated as 10.12 mm. The error for this one test is +0.12 mm. Why is one measurement insufficient to certify full-volume accuracy? Consider position, direction, material, repeats and boundary segmentation.
FAQ — Is voxel size the same as resolution? No. Voxel spacing describes image sampling; actual feature resolution also depends on optics, detector, geometry, noise and reconstruction.
FAQ — Is every bright region a denser material? No. Display mapping, energy dependence, contrast materials, artefacts and different attenuation mechanisms all matter. A calibrated material model is needed for composition claims.
FAQ — Does 3D visualization make dimensions accurate? No. The rendered surface inherits segmentation and calibration choices; traceability and uncertainty come from performance evidence.
Transfer the same measurement chain from a medical scan to a manufactured part, then to a rock specimen. Which quantities, reference objects, safety considerations and acceptance criteria must change?
CHAPTER 11 OF 12 · Read across the ecosystem
11. Connected X-ray physics and analysis routes
CT belongs to a family of X-ray methods that interrogate matter in different ways. Absorption provides attenuation contrast, diffraction interrogates crystallographic order, fluorescence reports emitted characteristic photons, and surface electron methods probe different depths.
Reading across these specialist articles prevents a common error: treating a beautiful reconstruction as proof of chemistry, crystal structure or a precise dimension without a supporting method.
Follow the connected eduKateSG articles according to the next question you need to answer. Links lead to distinct measurements or methods, not interchangeable evidence.
- The Gold Standard of X-ray Absorption — attenuation and energy dependence
- The Gold Standard of X-ray Optics — beam formation and optical constraints
- The Gold Standard of X-ray Diffraction — crystal-order information beyond CT contrast
- The Gold Standard of X-ray Fluorescence — element-sensitive emitted photon signals
- The Gold Standard of Physics — radiation, interactions and modelling
- The Gold Standard of Wavelet Analysis — multiscale signals and the limits of image processing
- The Gold Standard of Spectroscopy — how measured spectra differ from reconstructed structures
CHAPTER 12 OF 12 · Read across the ecosystem
12. Primary technical references and scope
Every count, coefficient, segment and voxel number in this guide is a hypothetical teaching case. The equations assume conditions stated alongside them and cannot verify an actual patient, object or instrument.
A next learning task is to make an evidence ledger for a proposed CT claim: source settings, specimen, projection model, detector correction, reconstruction, segmentation, reference artefact, uncertainty and independent confirmation. If any link is missing, constrain the claim rather than filling the gap with visual certainty.
Primary sources and technical references support the instrument physics, uncertainty and interpretation boundaries:
- FDA, Computed Tomography (CT) — imaging geometry, medical context and radiation-related cautions
- NIH NIBIB, Computed Tomography — source–detector rotation and sectional reconstruction
- NIST, ASME B89.4.23 XCT Performance Evaluation — industrial dimensional performance and geometry sensitivity
- NIST, Charting the Course Towards Dimensional Measurement Traceability by XCT — scale calibration, uncertainty and measurement traceability
- NIST, Literature Review of Fabricated Artifacts for XCT (2026) — phantoms and controlled-resolution or defect-reference features
No clinical dose, accuracy, diagnosis, material composition or specimen property is inferred from the imaginary examples. Actual CT claims require system-specific procedures and qualified interpretation.
The instructional standard is to define a measurable question, show a worked method and challenge it with a changed context. The immutable Clementi Secondary 1 Mathematics teaching reference illustrates that educational discipline but is not a specialist technical reference for X-ray Computed Tomography.