Metal Artifact Reduction in CT: How MAR Works

Introduction

Metal artifact reduction (MAR) is a core imaging physics capability that suppresses the bright and dark streaks metal implants create in CT images, restoring anatomical visibility and Hounsfield Unit (HU) accuracy. Metal implants are increasingly common in today's imaging population, and with them comes a familiar challenge: streaks that obscure anatomy, distort HU, and compromise quantitative workflows. In modern CT practice, MAR is no longer just a convenience feature. It is a clinically relevant imaging physics issue with direct implications for diagnostic CT, radiation therapy (RT) planning, and attenuation correction in PET/CT and SPECT/CT.^1,2

Orthopedic prostheses, dental restorations, spinal hardware, pacemakers, vascular clips, and other metallic devices can all produce structured artifacts that reduce confidence in image interpretation. As CT becomes more quantitative and more closely tied to downstream clinical decisions, understanding MAR becomes increasingly important for both technologists and physicists.^1,2

This Physics Pulse article explains why metal causes artifacts, how the main MAR strategies work, what the major vendors offer, and how a qualified medical physicist should validate these tools before clinical use.

Why Metal Causes Artifacts in CT

Metal causes CT artifacts because it violates the assumptions built into conventional reconstruction models. CT reconstruction works best when attenuation measurements behave like reasonably well-sampled, linear line integrals through tissue. Dense, high-atomic-number materials do not behave that way.^2,3

Beam Hardening

Beam hardening is the preferential absorption of lower-energy photons as a polychromatic X-ray beam passes through dense material, raising the transmitted beam's average energy. CT uses a polychromatic beam, meaning photons span a range of energies. When this beam passes through dense metal, lower-energy photons are preferentially absorbed, leaving a transmitted beam with a higher average energy. Since the reconstruction algorithm does not fully account for this nonlinear spectral shift, the result can be cupping, dark bands, and streak artifacts near or between metallic objects.^2,4

Photon Starvation

Photon starvation occurs when metal is sufficiently dense or thick that very few photons reach the detector along certain projection paths. This produces extremely noisy or unstable projection data, which reconstruct as alternating bright and dark streaks radiating from the implant. Photon starvation is a major cause of severe metal streak artifacts.^2,3,15

Scatter, Partial Volume, and Detector Limitations

Scatter contributes unwanted signal at the detector, while partial volume effects occur when a voxel contains a mixture of metal and adjacent tissue. In addition, detector saturation, lag, nonlinear response, and limited dynamic range can worsen the instability of heavily attenuated projection data. These departures from ideal conditions further increase structured artifacts in the reconstructed image.^3,5

Clinical Impact

Metal artifacts are not just a cosmetic image quality issue. They can directly affect diagnosis, quantitation, and treatment planning.^1,2

• Diagnostic CT. Artifacts can obscure adjacent anatomy and make it more difficult to evaluate fractures, hardware loosening, fluid collections, or other pathology near implanted devices.^2,3
• Radiation therapy planning. Corrupted HU values can affect electron density mapping and dose calculation. Even when the implant itself is not in the target, artifact in the surrounding soft tissue can introduce uncertainty into treatment planning workflows.^1,8
• PET/CT and SPECT/CT. The CT dataset is often used to generate the attenuation map. If metal artifact distorts the CT numbers, those errors may propagate into attenuation correction and create false hot or cold regions on the nuclear medicine images.^1,2

For medical physicists, this is why MAR should be viewed not as a simple post-processing enhancement, but as a clinically relevant imaging physics tool that requires understanding and validation. The same dose-versus-image-quality tradeoffs that govern routine scanning also apply here; see our companion article on CT protocol optimization for the broader framework.

Key Technical Principles: Main Strategies for MAR

Broadly, MAR approaches fall into three categories: acquisition/protocol optimization, projection-space MAR, and image-space or hybrid MAR.^2,6,7

1. Acquisition and Protocol Optimization

The first opportunity to reduce artifact is during image acquisition, before any correction algorithm runs.

• Higher tube potential (kVp). Increasing kVp improves beam penetration and reduces the severity of beam hardening and photon starvation, although it may come with some loss of soft tissue contrast.^5,8
• Adequate mAs. Higher tube current can improve signal reliability in highly attenuated projections, particularly when paired with iterative reconstruction methods that help control image noise.⁵
• Dual-energy or spectral CT. A particularly important development, spectral CT can generate virtual monoenergetic images at higher keV. These high-keV images often reduce streaking around metal and improve visualization of adjacent structures. In many cases, virtual monoenergetic imaging is one of the most effective scanner-based tools for reducing metal artifact while preserving local anatomy.^2,8

2. Projection-Space MAR

Projection-space MAR operates in the sinogram domain, where corrupted raw data are identified and corrected before final image reconstruction.^2,7

A typical workflow includes:

1. reconstructing an initial image,
2. segmenting the metal,
3. forward-projecting the metal mask into the sinogram,
4. identifying metal-affected measurements,
5. estimating replacement values using interpolation, model-based fitting, or prior-image guidance, and
6. reconstructing a corrected image with the true metal reinserted afterward.^7,9

These methods are especially effective for reducing streaks caused by photon starvation and are the foundation of many commercial vendor MAR solutions.^1,2,8

3. Image-Space, Hybrid, and Deep Learning MAR

Some MAR methods operate directly on the reconstructed image rather than the raw projection data. These image-space approaches may use filtering, iterative regularization, or hybrid strategies that combine image- and projection-domain corrections.^2,10

More recently, deep learning–based MAR has emerged as a major area of active development. Neural networks can be trained to map artifact-corrupted images or sinograms to cleaner outputs, often reducing complex streak patterns more effectively than conventional interpolation-only approaches.^9,11

Some of these methods are physics-informed, meaning they incorporate prior knowledge of beam hardening or CT system behavior into the learning process. These approaches are intended to reduce oversmoothing and improve robustness.^9,12

Still, caution is warranted. A cleaner image is not always a more accurate one. Deep learning MAR may behave unpredictably with unfamiliar implants or scanner conditions, so proper local validation remains essential before using these tools in quantitative or high-consequence workflows.^11–13

Vendor-Specific MAR Solutions

Several major CT vendors offer proprietary metal artifact reduction tools. Most fall into either the projection-space/hybrid MAR category or the spectral/acquisition-based MAR category. Because vendor implementations are often proprietary, the exact internal algorithm may not always be fully disclosed. However, their general operating principles can still be described in clinically useful terms.

The important practical point is that not all MAR is a single button. Some artifact reduction comes from a dedicated reconstruction algorithm, some comes from dual-energy or spectral reconstruction choices, and some workflows combine both. Because reconstruction kernel choice also interacts with artifact appearance and quantitative accuracy, it pairs closely with MAR; see our deep dive on Siemens CT reconstruction kernels.

Practical Tips for Technologists

For technologists, patients with metallic implants often require a more deliberate scanning strategy.

• Expect artifact when dense hardware is present, particularly with large orthopedic implants, spinal rods, dental hardware, or multiple metallic objects in the scan field.^2,3
• Use the appropriate protocol, including higher kVp or optimized technique factors when clinically indicated.^5,8
• Know what MAR tools are available on your scanner and how they affect image appearance. MAR may improve visualization, but it can also change the appearance of tissues adjacent to hardware.^1,2
• In workflows such as PET/CT, SPECT/CT, or RT simulation, review both MAR and non-MAR images, since one may look cleaner while the other preserves more of the original data characteristics.^1,8
• Communicate with the interpreting physician or physicist when artifact is severe enough to compromise anatomy, attenuation correction, or planning confidence.

Practical Tips for Physicists

For medical physicists, MAR should be approached as a commissioning, QA, and workflow issue rather than a purely cosmetic reconstruction choice.

Local validation should include both qualitative artifact assessment and quantitative testing, especially when MAR-corrected images may be used in HU-sensitive applications.^1,2

Representative phantom testing with hardware similar to what is encountered clinically can help evaluate:

• degree of artifact suppression,
• restoration of local HU accuracy,
• preservation of adjacent anatomy, and
• impact on downstream tasks such as attenuation correction or dose calculation.^1,8

This is especially important for RT simulation, PET/CT, and SPECT/CT, where CT inaccuracies may directly affect treatment planning or quantitation.¹

Deep learning–based MAR tools deserve special scrutiny. Although they may substantially reduce visible artifact, physicists should be cautious about potential subtle image alterations, especially for unfamiliar implant types or out-of-distribution cases.^11–13

Regulatory and Accreditation Considerations

MAR itself is not governed by a single dedicated standard, but the CT systems that use it operate within established accreditation and QA frameworks. Under the ACR CT Accreditation Program, facilities must demonstrate ongoing image quality and a qualified medical physicist's annual survey, which provides the natural venue to confirm that MAR and spectral reconstruction settings are commissioned and behaving as expected. The AAPM CT Metal Artifact Reduction (CT-MAR) Grand Challenge has further driven standardized evaluation of MAR algorithm performance across vendors.^13,14

DRPS provides ACR-aligned CT physics services to imaging centers and hospitals across Florida, Maryland, Virginia, Washington DC, California, and Nevada, including states where state-level regulations layer on top of federal requirements. For a broader view of how physics requirements map to accreditation, see our overview of ACR accreditation physics requirements.

Frequently Asked Questions (FAQs)

What is metal artifact reduction (MAR) in CT?

Metal artifact reduction (MAR) is a set of acquisition, projection-space, and image-space techniques that suppress the bright and dark streaks metal implants create in CT images. The goal is to restore anatomical visibility and Hounsfield Unit accuracy near hardware.

Why does metal cause streak artifacts in CT?

Metal violates the linear, well-sampled line-integral assumptions of CT reconstruction. Beam hardening, photon starvation, scatter, and partial volume effects all corrupt the projection data, which reconstructs as cupping, dark bands, and radiating streaks.

Does MAR change Hounsfield Unit (HU) values?

Yes. MAR can alter HU values in tissue adjacent to hardware, which matters for radiation therapy planning and PET/CT attenuation correction. Reviewing both MAR and non-MAR images and validating HU accuracy with phantoms is recommended for quantitative workflows.

What are the vendor names for MAR algorithms?

Common vendor MAR tools include Siemens iMAR, Philips O-MAR, Canon SEMAR, and GE Smart MAR. Dual-energy virtual monoenergetic imaging at higher keV is a complementary, acquisition-based approach to reducing metal artifact.

Should a medical physicist validate MAR before clinical use?

Yes. MAR is a commissioning and QA issue, not a cosmetic post-processing toggle. A qualified medical physicist should perform phantom-based qualitative and quantitative testing, especially for HU-sensitive RT simulation, PET/CT, and SPECT/CT applications.

Key Takeaways

• MAR addresses a physics problem, not a cosmetic one. Metal artifacts arise because dense, high-atomic-number implants break CT's linear line-integral assumptions through beam hardening, photon starvation, scatter, and partial volume effects.^2,3
• Three strategy families exist: acquisition/protocol optimization (higher kVp, adequate mAs, dual-energy virtual monoenergetic imaging), projection-space (sinogram) MAR, and image-space, hybrid, or deep learning MAR.^2,6,7
• Vendor tools differ in mechanism. Siemens iMAR, Philips O-MAR, Canon SEMAR, and GE Smart MAR are not interchangeable, and some workflows combine a dedicated algorithm with spectral monoenergetic reconstructions.^1,2,7,8
• MAR can change HU values. Because corrupted or altered CT numbers propagate into RT dose calculation and PET/CT and SPECT/CT attenuation correction, MAR is clinically consequential beyond image appearance.^1,8
• A cleaner image is not always a more accurate one. Deep learning MAR can behave unpredictably on unfamiliar implants, so local validation by a qualified medical physicist is essential.^11–13

How DRPS Can Help

Diagnostic Radiation Physics Services (DRPS) helps imaging centers, hospitals, and radiation oncology departments commission and validate MAR and spectral CT reconstruction as part of routine CT physics support. Our board-certified medical physicists perform phantom-based qualitative and quantitative MAR testing, verify HU accuracy for RT simulation and PET/CT and SPECT/CT attenuation correction, and align CT performance with ACR CT Accreditation requirements. We serve clients across Florida, Maryland, Virginia, Washington DC, California, and Nevada. Contact DRPS to discuss MAR validation or your CT physics program.

Conclusion

Metal artifact reduction in CT sits at the intersection of X-ray physics, image reconstruction, and clinical workflow. As implanted hardware becomes more common and as quantitative imaging expands, MAR is becoming an increasingly important competency for technologists and medical physicists alike.

The goal is not simply to make the image look cleaner. The real goal is to improve anatomical visibility, preserve quantitative accuracy, and support better downstream clinical decisions in patients with metal hardware.^1,2,14

Related Resources

• CT protocol optimization
• Siemens CT reconstruction kernels
• ACR accreditation physics requirements

References

1. Boas FE, Fleischmann D. The application of metal artifact reduction methods on computed tomography scans for radiotherapy applications: A literature review. Front Oncol. 2021. Available at: pmc.ncbi.nlm.nih.gov
2. Gjesteby L, De Man B, Jin Y, et al. Current and novel techniques for metal artifact reduction at CT. RadioGraphics. 2016;36(6):1770-1791. Available at: pubs.rsna.org
3. Yu L, Li H, Mueller J, et al. Reduction of metal artifacts: beam hardening and photon starvation. Proc SPIE. 2014. Available at: spiedigitallibrary.org
4. Rigaku. What Is Beam Hardening in CT? Available at: rigaku.com
5. Barrett JF, Keat N. CT artifacts: causes and reduction techniques. Available at: openaccessjournals.com
6. Meyers J, et al. Advances in metal artifact reduction in CT images. Diagn Interv Imaging. 2023. Available at: sciencedirect.com
7. Meyer E, Raupach R, Lell M, Schmidt B, Kachelrieß M. CT metal artifact reduction algorithms: Toward a framework for comparing performance. Available at: pmc.ncbi.nlm.nih.gov
8. Wellenberg RHH, Boomsma MF, van Osch JAC, et al. Metal artifact reduction techniques for single-energy CT and dual-energy CT: a review. Br J Radiol Open. 2018. Available at: academic.oup.com
9. Ghazi P, et al. Physics-informed sinogram completion for metal artifact reduction in CT. Med Phys. 2023. Available at: PubMed
10. Siemens Healthineers educational overview. Understanding CT Artifacts: A Comprehensive Guide. Available at: medical-professionals.com
11. Jeon HG, et al. Metal artifacts reduction in CT scans using convolutional neural network. Available at: PubMed
12. Residual Metal Artifact Reduction in CT Images: An Unsupervised Deep Learning Approach. Med Phys. Available at: aapm.onlinelibrary.wiley.com
13. Haneda S, et al. AAPM CT metal artifact reduction grand challenge. Med Phys. 2025. Available at: aapm.onlinelibrary.wiley.com
14. AAPM. CT Metal Artifact Reduction (CT-MAR) Grand Challenge. Available at: aapm.org
15. Yu L, et al. Reduction of metal artifacts: beam hardening and photon starvation. Proc SPIE. Available at: spiedigitallibrary.org