CT Image Artifacts: Causes and Correction

A CT artifact is any systematic discrepancy between the reconstructed CT numbers and the true attenuation of the patient — and nearly every one of them can be traced to a specific physical assumption that the scan violated. Recognizing which assumption failed is what separates a confident "that is beam hardening" from a mistaken call of hemorrhage, and it determines whether the fix belongs in the protocol, the reconstruction, or the scanner's quality-control program. ^{1, 2}

Introduction

Computed tomography reconstructs a cross-sectional map of linear attenuation coefficients from thousands of X-ray projection measurements. That reconstruction rests on a short list of assumptions: the beam is effectively monochromatic, the patient holds still, the projections are sampled finely enough, the detector channels are calibrated identically, and each measurement behaves like a clean line integral of attenuation. When one of those assumptions fails, the mathematics does not break gracefully — it redistributes the error across the image as a structured pattern that can mimic or obscure disease. ^{1, 3}

Artifacts matter for three reasons. First, they can be mistaken for pathology, or they can hide it. Second, in quantitative workflows — radiation therapy simulation, PET/CT and SPECT/CT attenuation correction, coronary calcium scoring, and CT number–based tissue characterization — an artifact is not just a cosmetic problem; the corrupted Hounsfield Units feed downstream calculations. Third, some artifacts are a signal that the scanner itself needs attention, making artifact recognition part of the physics and QC program, not only the reading room. ^{1, 2, 5}

This guide walks through the major artifact families a diagnostic team encounters — physics-based (beam hardening, photon starvation, partial volume, undersampling), patient-based (motion, metal), and scanner-based (ring artifacts) — plus the helical and cone-beam artifacts introduced by the reconstruction geometry itself. For each, it covers the physical cause, the visual signature, and the practical correction. DRPS supports imaging facilities across Florida, Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware with CT physics testing and accreditation support that make artifact evaluation a documented part of routine quality control.

Topic Explanation

What is a CT artifact?

A CT artifact is a structured error in the reconstructed image that does not correspond to the true attenuation distribution of the patient. Unlike random quantum noise, which is statistically distributed, an artifact has a deterministic geometry tied to its cause: streaks radiate from a source, rings center on the isocenter, cupping follows the beam path, and blurring follows motion. That geometry is the diagnostic clue. ¹

It helps to group artifacts by where the assumption failed:

Physics-based artifacts arise from the physical processes of data acquisition — the polychromatic spectrum, finite photon statistics, finite sampling, and finite voxel size.
Patient-based artifacts arise from the patient — voluntary or physiologic motion, and materials such as metal or contrast that lie far outside the attenuation range CT was optimized for.
Scanner-based artifacts arise from imperfections in the hardware — detector channel miscalibration, gain drift, or mechanical instability.
Reconstruction/technique artifacts arise from the helical and cone-beam geometry and the interpolation used to build images from a moving source-detector system. ¹

A single clinical picture often mixes families. A hip prosthesis produces streaks that are simultaneously beam hardening (spectral), photon starvation (statistical), and partial volume (geometric). Naming the dominant mechanism is what points to the right correction. For the specific case of implants, our detailed guide to metal artifact reduction in CT explains the algorithms in depth.

Why artifacts are a physics and QC issue, not only a reading-room issue

Some artifacts are unavoidable consequences of a particular patient and can only be managed with protocol and reconstruction choices. Others — rings, persistent nonuniformity, and certain streaks — indicate that the scanner's calibration or hardware has drifted. Distinguishing the two is a core function of the qualified medical physicist's evaluation, and it is why artifact assessment is an explicit part of the ACR CT accreditation image-quality review. ^{5, 6}

Key Technical Principles

The reconstruction assumptions, stated precisely

The measured intensity along a ray for a monochromatic beam of energy $E$ obeys the Beer–Lambert law:

I = I_{0} e^{- \int μ (x, E) d l}

so that the log-transformed projection is a clean line integral of attenuation, $p = ln (I_{0} / I) = \int μ d l$ . Filtered back-projection and iterative reconstruction both assume this linear relationship holds. The CT number of a reconstructed voxel is then defined relative to water:

HU = 1000 \times \frac{μ - μ _{water}}{μ _{water}}

Every major artifact can be understood as a violation of the line-integral assumption or of the sampling and calibration conditions that back-projection requires. ^{1, 3, 7}

Beam hardening: a spectral violation

Clinical X-ray beams are polychromatic. The measured intensity is really a spectral integral:

I = \int I_{0} (E) e^{- \int μ (x, E) d l} d E

Because $μ$ falls with energy, low-energy photons are absorbed preferentially and the surviving beam's mean energy rises — it "hardens" — as it traverses tissue. The log transform of a spectral integral is not a linear line integral, so the reconstruction underestimates attenuation along long, dense paths. The signatures are cupping (the center of a uniform object reads lower than its edges) and dark bands or streaks between dense structures such as the petrous bones or contrast-filled vessels. Scanners correct much of this with beam pre-filtration (bowtie filters), a water-based beam-hardening correction applied to the projections, and dual-energy or spectral methods that estimate the monochromatic-equivalent image. ^{1, 7} Our overview of dual-energy and spectral CT covers how virtual monoenergetic images reduce these effects.

Photon starvation: a statistical violation

Where the path is most attenuating — through the shoulders, the pelvis, or metal — so few photons reach the detector that the projection measurement is dominated by quantum noise. Reconstruction cannot distinguish "true low signal" from "noise," so it back-projects the noisy rays as bright and dark streaks aligned with the direction of greatest attenuation. The correction is more signal along those paths: adequate mAs, tube-current modulation that boosts output for the highest-attenuation projection angles, and adaptive projection filtering. Iterative and deep-learning reconstruction further suppress the streaks by weighting noisy measurements less heavily. ¹ See our discussion of iterative and deep-learning reconstruction for how these methods handle noisy projections.

Undersampling and aliasing: a sampling violation

Back-projection assumes the object is sampled finely enough in both the radial (detector) and angular (view) directions. When detector spacing or the number of projections is too coarse to capture the highest spatial frequencies present, high-frequency information aliases into fine, striated lines, typically radiating from sharp, high-contrast edges. The governing constraint is the sampling theorem: to represent a spatial frequency $f_{m a x}$ without aliasing, the sampling interval $Δ$ must satisfy

Δ \leq \frac{1}{2 f _{m a x}}

For example, resolving $7 lp/cm$ requires a sampling pitch no coarser than $Δ \leq 1/ (2 \times 7) \approx 0.071 cm = 0.71 mm$ . Scanners mitigate view aliasing with a quarter-detector offset, a flying focal spot, and by acquiring enough projections per rotation. ^{1, 8}

Partial-volume averaging: a geometric violation

Each voxel reports a single CT number, but a voxel that straddles two tissues reports their volume-weighted average attenuation:

HU_{voxel} = i \sum f_{i} HU_{i}, i \sum f_{i} = 1

where $f_{i}$ is the volume fraction of tissue $i$ . Consider a voxel that is 60% soft tissue ( $HU = 40$ ) and 40% cortical bone ( $HU = 1000$ ):

HU_{voxel} = 0.60 \times 40 + 0.40 \times 1000 = 24 + 400 = 424 HU

The voxel reads as neither tissue — a value that can masquerade as calcification, hemorrhage, or a lesion at a boundary. Thin sections reduce the effect by shrinking the voxel relative to the anatomy; the classic culprits are the skull base, the diaphragm, and small structures oriented obliquely to the scan plane. ¹

Comparison of the major CT artifacts

Artifact	Family	Physical cause	Visual signature	Primary correction
Beam hardening	Physics (spectral)	Polychromatic beam hardens through dense tissue	Cupping; dark bands between dense structures	Bowtie filter, software beam-hardening correction, dual-energy/monoenergetic
Photon starvation	Physics (statistical)	Too few photons along high-attenuation paths	Bright/dark streaks along the most-attenuating direction	Higher mAs, tube-current modulation, iterative/DL reconstruction
Aliasing (undersampling)	Physics (sampling)	Radial or angular sampling below Nyquist	Fine striations from sharp high-contrast edges	Quarter-detector offset, flying focal spot, more projections
Partial volume	Physics (geometric)	Voxel spans multiple tissues	False intermediate HU at boundaries	Thinner sections, smaller voxels
Motion	Patient	Patient or physiologic motion during acquisition	Blurring, doubling, ghosting, streaks	Faster rotation, gating, breath-hold, motion-correction algorithms
Metal	Patient	Very high density: hardening + starvation + partial volume	Dense bright/dark streaks radiating from implant	High kVp, MAR algorithms, monoenergetic reconstruction
Ring	Scanner	Miscalibrated or drifting detector channel	Concentric ring(s) centered on the isocenter	Detector air calibration; service
Cone-beam / windmill	Reconstruction	Wide-cone geometry and helical interpolation	Windmill/streaks near high-contrast axial edges	Cone-beam reconstruction, narrower collimation, thinner slices, pitch selection

The table is a starting point for recognition; the definitive assessment of a scanner's artifact behavior belongs in the physicist's phantom-based evaluation. ^{1, 2, 6}

Clinical Impact

Artifacts change diagnoses and downstream calculations, not just image aesthetics. A posterior-fossa dark band from beam hardening (the Hounsfield artifact) can simulate or obscure an infarct or hemorrhage. Photon-starvation streaks across the shoulders can bury a lung apex or a small pneumothorax. Partial-volume averaging at the skull base can imitate a fracture or a subtle mass. Motion doubling at the lung bases can mimic bronchiectasis or a nodule. ^{1, 3}

The stakes rise in quantitative imaging. In radiation therapy simulation, corrupted CT numbers near hardware distort electron-density mapping and dose calculation. In PET/CT and SPECT/CT, distorted CT numbers propagate into the attenuation map and can create false hot or cold regions on the emission images. In coronary calcium scoring and CT number–based tissue characterization, beam hardening and partial volume shift the very numbers the task depends on. Metal-artifact studies show that reduction algorithms can restore diagnostic confidence — for example, improving vessel conspicuity in CT venography around hip and knee prostheses — but the same algorithms alter local HU, so the benefit must be weighed against quantitative fidelity. ^{2, 4} Postoperative imaging near spinal hardware is a recurring example where parameter optimization and metal-artifact reduction materially change what the radiologist can report. ³

Practical Optimization Tips

Anticipate the artifact before scanning

Expect beam hardening and streaks whenever dense structures dominate the field: the posterior fossa, the shoulders, the pelvis, contrast-filled vessels, and any implant. Plan kVp and mAs accordingly.
Raise kVp for dense anatomy and metal. Higher kVp improves penetration and reduces both beam hardening and photon starvation.
Keep the anatomy of interest near the isocenter and use the correct bowtie/scan field of view, because beam-hardening and scatter corrections are optimized for the calibrated geometry.

Choose sections and reconstruction deliberately

Use thinner sections where partial volume threatens interpretation — the skull base, small nodules, and oblique boundaries — then reformat.
Match the kernel to the task. Sharp kernels amplify streaks and rings; smoother kernels and iterative/deep-learning reconstruction suppress them but change texture. Review HU-sensitive and low-contrast tasks on a defined, validated reconstruction. Our guide to reconstruction kernels explains how kernel choice trades noise against resolution.
Read both MAR and non-MAR series for implants, and both reconstructions when a monoenergetic option exists.

Manage motion

Immobilize, coach breath-hold, and shorten acquisition. Faster gantry rotation and higher pitch reduce the time the anatomy has to move.
Gate cardiac and respiratory studies when temporal resolution is the limiting factor; motion at the coronary or diaphragmatic level is a temporal-resolution problem before it is a reconstruction problem.

Treat rings and persistent nonuniformity as QC events

A recurring ring is a detector-calibration finding, not a protocol setting. Run the scanner's air/detector calibration and, if it persists, place a service call.
Log and trend artifacts so a slowly drifting channel is caught before it reaches the reading room.

Common pitfalls to avoid

Calling an artifact pathology — or dismissing real pathology as an artifact — without checking a second series, phase, or reconstruction.
Leaving MAR on by default for HU-sensitive workflows without commissioning it.
Assuming iterative or deep-learning reconstruction removes the underlying error. It suppresses the appearance; the corrupted projection data are still corrupted, and out-of-distribution cases deserve extra scrutiny.
Ignoring a recurring ring or nonuniformity because "the images still look fine" — it is an early sign of detector drift.

Regulatory Considerations

CT scanners are FDA-regulated radiation-producing devices, and their image-quality performance — including artifact evaluation — is governed by accreditation and state radiation-control requirements rather than by a single "artifact standard." The relevant frameworks are:

FDA 21 CFR 1020.33 establishes federal performance standards for CT equipment, and manufacturers build in the beam-hardening, scatter, and detector-calibration corrections that suppress artifacts at the source. ⁹
ACR CT Accreditation Program requires a qualified medical physicist to evaluate each unit at least annually (with up to 14 months permitted between surveys), and the image-quality assessment explicitly includes artifact evaluation alongside spatial resolution, low-contrast resolution, uniformity, and noise. ⁵
ACR–AAPM Technical Standard for Diagnostic Medical Physics Performance Monitoring of CT describes the physicist's role in acquiring phantom data, inspecting for artifacts, interpreting the results, and issuing a signed report. ⁶
State radiation-control programs register and inspect CT scanners as radiation machines. In Florida, CT is regulated by the Florida Department of Health, Bureau of Radiation Control under Florida Administrative Code Chapter 64E-5; DRPS also serves Maryland, Virginia, Washington DC, California, Nevada, Pennsylvania, New York, New Jersey, and Delaware, where state radiation-control authorities impose parallel machine-registration and survey expectations. Always confirm requirements with the authority having jurisdiction.

Because X-ray machines are FDA plus state-regulated (distinct from radioactive material, which the NRC or an Agreement State licenses), a CT program's artifact and image-quality documentation lives inside the annual physics survey and the facility's accreditation record. For the broader compliance picture, see our guide to ACR accreditation physics requirements. ^{5, 6}

Frequently Asked Questions (FAQs)

What causes most CT artifacts?

Most CT artifacts occur when the physical situation violates an assumption built into image reconstruction. Reconstruction assumes monochromatic X-rays, a stationary patient, adequately sampled projections, and attenuation values that behave like well-behaved line integrals. Beam hardening breaks the monochromatic assumption, motion breaks the stationary assumption, undersampling breaks the sampling assumption, and dense metal breaks several at once.

How can you tell a CT artifact from real pathology?

Artifacts usually have a physics-based geometry that does not respect anatomy: streaks that radiate from a dense object, dark bands between two high-density structures, concentric rings centered on the isocenter, or blurring that follows a moving structure. When a finding does not match any anatomic boundary and lines up with a known artifact pattern, it should be confirmed on a different reconstruction, phase, or series before it is called disease.

What is the difference between beam hardening and photon starvation?

Beam hardening is a spectral effect: as a polychromatic beam passes through dense tissue, lower-energy photons are preferentially absorbed, raising the beam's mean energy and producing cupping and dark bands. Photon starvation is a statistical effect: so few photons survive along the most attenuating paths that the projection data become extremely noisy, reconstructing as bright and dark streaks. High kVp helps both; adequate mAs and tube-current modulation specifically address photon starvation.

Does metal artifact reduction (MAR) make the image more accurate?

Not always. MAR reliably reduces the visible streaks around implants, but it can also alter Hounsfield Units in nearby tissue and occasionally introduce new secondary artifacts. Because those altered numbers propagate into radiation therapy dose calculation and PET/CT and SPECT/CT attenuation correction, MAR should be validated during commissioning and reviewed alongside the non-MAR series for HU-sensitive tasks.

Are ring artifacts a scanner problem or a protocol problem?

Ring artifacts are almost always a scanner problem. They come from one or more detector channels that are miscalibrated or drifting, so the same channel error appears at every projection angle and reconstructs as a ring or partial ring centered on the isocenter. The fix is detector air calibration or service, not a protocol change, which is why a recurring ring is a QC and physics finding.

How does the annual medical physicist survey address artifacts?

The ACR CT Accreditation Program and the ACR–AAPM technical standard require a qualified medical physicist to evaluate each CT scanner at least annually, and artifact evaluation is an explicit component of that image-quality assessment alongside spatial resolution, low-contrast resolution, uniformity, and noise. The physicist scans a uniform phantom, inspects for rings, streaks, and nonuniformity, and confirms the scanner's own correction software is performing as expected.

Can reconstruction settings create or hide artifacts?

Yes. Sharper kernels amplify noise and can make streaks and rings more conspicuous, while smoother kernels and iterative or deep-learning reconstruction can suppress them, sometimes at the cost of altering texture or masking a subtle finding. Because reconstruction changes both real anatomy and artifacts, HU-sensitive and low-contrast tasks should be reviewed on a defined, validated reconstruction rather than whatever series looks cleanest.

Key Takeaways

An artifact is a violated assumption. Name the assumption that failed — monochromatic beam, stationary patient, adequate sampling, calibrated detectors, clean line integrals — and the correction follows.
Geometry is the clue. Cupping and dark bands mean beam hardening; directional streaks mean photon starvation or metal; concentric rings mean a detector channel; blurring means motion; false boundary HU means partial volume.
Quantitative workflows raise the stakes. Corrupted CT numbers propagate into RT dose calculation and PET/CT and SPECT/CT attenuation correction, so artifact control is a commissioning and QC issue, not a cosmetic one.
Some artifacts are protocol problems and some are scanner problems. Rings and persistent nonuniformity point to calibration or hardware; beam hardening, motion, and partial volume are managed with technique and reconstruction.
Reconstruction hides appearances, not causes. Iterative and deep-learning methods suppress streaks but do not un-corrupt the projection data; validate them, especially on unfamiliar anatomy and implants.
Artifact evaluation is a required part of the annual physics survey under ACR accreditation and the ACR–AAPM technical standard.

Conclusion

CT artifacts are not random blemishes to be tolerated; they are structured, explainable consequences of physics and hardware. A team that can look at a streak, a ring, or a cup and immediately ask "which assumption did the scan violate?" is a team that will make fewer diagnostic errors and run a tighter quality program. The same reasoning tells you where the fix lives — in the kVp and mAs, in the section thickness, in the reconstruction, in a gating decision, or in a service call for a drifting detector. Building that recognition into daily practice, and documenting the scanner's artifact behavior in the annual physics survey, is how facilities keep CT both diagnostically reliable and quantitatively defensible.

How DRPS Can Help

Diagnostic Radiation Physics Services helps imaging facilities turn artifact recognition into a documented quality process. Our board-certified medical physicists provide CT physics testing, phantom-based artifact and image-quality evaluation, protocol and reconstruction review, metal-artifact and spectral-reconstruction commissioning, and accreditation support aligned with ACR and state requirements. When a ring, a streak, or a nonuniformity shows up, we help distinguish a protocol issue from a scanner issue and document the finding defensibly.

DRPS supports facilities across our service locations, including Florida, Maryland, Virginia, Washington DC, California, Nevada, New York, Pennsylvania, New Jersey, and Delaware. To connect artifact evaluation to your QC program, contact our team.

Related Resources

References

Barrett JF, Keat N. Artifacts in CT: recognition and avoidance. RadioGraphics. 2004;24(6):1679-1691. doi:10.1148/rg.246045065. PubMed
Triche BL, Nelson JT, McGill NS, et al. Recognizing and minimizing artifacts at CT, MRI, US, and molecular imaging. RadioGraphics. 2019;39(4):1017-1018. doi:10.1148/rg.2019180022. PubMed
Ghodasara N, Yi PH, Clark K, Fishman EK, Farshad M, Fritz J. Postoperative spinal CT: what the radiologist needs to know. RadioGraphics. 2019;39(6):1840-1861. doi:10.1148/rg.2019190050. PubMed
Woo H, Chai JW, Choi YH, Jin KN. Metal artifact reduction for orthopedic prosthesis in lower extremity CT venography: evaluation of image quality and vessel conspicuity. Cardiovascular and Interventional Radiology. 2019;42(11):1619-1626. doi:10.1007/s00270-019-02326-2. PubMed
American College of Radiology. CT Accreditation Program Requirements. acraccreditation.org
American College of Radiology; American Association of Physicists in Medicine. ACR–AAPM Technical Standard for Diagnostic Medical Physics Performance Monitoring of Computed Tomography (CT) Equipment. acr.org
National Institute of Standards and Technology. XCOM: Photon Cross Sections Database. nist.gov
International Atomic Energy Agency. Diagnostic Radiology Physics: A Handbook for Teachers and Students. Vienna: IAEA; 2014. iaea.org
U.S. Food and Drug Administration. Computed Tomography (CT) — Performance Standards for Diagnostic X-Ray Systems (21 CFR 1020.33). fda.gov
American College of Radiology. Computed Tomography Quality Control Manual. Reston, VA: ACR. acr.org