Physics-Based Artifacts: Beam Hardening, Cupping & Partial Volume Averaging

Polychromatic attenuation and beam hardening

A diagnostic CT tube produces a spectrum of x-ray energies. Lower-energy photons are generally attenuated more strongly than higher-energy photons, so the transmitted spectrum becomes harder after passing through the patient. This is beam hardening. The ideal monoenergetic relationship between measured transmission and the line integral of attenuation is not exact for a changing polychromatic spectrum. Calibration and reconstruction must address that departure.

For a monoenergetic ideal beam through a uniform material, transmission is $I=I_0e^{-\mu x}$. Taking the negative logarithm of the normalized signal gives $-\ln(I/I_0)=\mu x$. In real CT, different energies have different attenuation coefficients; a single constant cannot describe every beam path exactly. This model remains useful for understanding why transmission measurements must be transformed before reconstruction, but it must not be treated as a perfect description of a clinical polyenergetic beam.

Cupping, bands and overcorrection

A uniform object can appear less attenuating at its center than at its edges when uncorrected beam hardening affects longer central paths differently. This is cupping. Dense structures can also produce dark bands between them, such as between petrous bones or metal components. An inappropriate correction can create a converse elevation or “capping” pattern. Inspect spatial distribution and phantom behavior instead of assuming that every dark band is a real low-attenuation lesion.

Prepatient filtration removes some low-energy photons. A bowtie filter also shapes fluence across the field according to the expected object. Correct centering and filter selection are necessary because a patient far from isocenter does not match the intended geometry. Calibration may apply water and material-related corrections, while modern iterative and spectral methods can reduce selected artifacts. None universally eliminates all beam-hardening behavior.

Acquisition and reconstruction choices

When dense bone creates posterior-fossa bands, an appropriate acquisition plane, exposure and reconstruction may help. Physical positioning must remain safe, especially in trauma. Reformatting an already acquired image changes the viewing plane but cannot remove data corruption or radiation already delivered. A higher-voltage protocol can improve penetration in selected patients, but adds or changes exposure and must be justified through the approved protocol.

Virtual monoenergetic images synthesized from spectral data can reduce selected beam-hardening effects, especially at higher energies near metal. Lower energies can enhance iodine conspicuity but may increase artifacts or noise. Review whether the selected energy improves the actual task instead of accepting one universal energy value for every patient and material.

A beam-hardening troubleshooting example

A water phantom develops central shading. First verify that the correct phantom, filter, field of view and centering were used. Review the approved calibration and repeat the test as directed. If the problem persists, document and escalate it rather than increase mAs until the image looks smoother. More photons reduce statistical noise; they do not inherently correct an inappropriate spectral calibration.

In a patient, a dark band between dense petrous bones may be an artifact even when it resembles a focal low-attenuation region. Compare adjacent slices, reformations and alternative available reconstructions. Do not dismiss a real lesion solely because an artifact is possible; communicate the limitation and obtain the responsible interpretation.

Partial-volume averaging

A reconstructed voxel represents a finite tissue volume. If it contains more than one material, its value can reflect a mixture rather than either material alone. In a simple ideal example with half water at 0 HU and half bone at 1000 HU, the volume-weighted value is 500 HU. This calculation illustrates averaging; real edge reconstruction and beam-hardening behavior can complicate the result.

A small dense lesion can therefore appear less dense when averaged with surrounding soft tissue, and a fluid-filled structure can seem denser near adjacent tissue. Thicker slices increase the opportunity for longitudinal mixing. Oblique anatomy and large pixels can also contribute. Thin appropriately sampled data and targeted reconstruction reduce the effect, but no finite-resolution technique eliminates all partial volume.

Distinguish acquired sampling, reconstructed thickness and reconstruction interval. Overlapping thick images improve continuity but do not reproduce the resolving capability of genuinely thinner acquired and reconstructed data. Display zoom enlarges the existing pixels; it does not undo partial-volume mixing.

Photon statistics and photon starvation

Detected photons fluctuate statistically. In a quantum-limited approximation, relative noise varies inversely with the square root of the number detected. Paths through a thick torso, shoulders or metal can have very low counts, making the logarithmic attenuation estimate unstable. Reconstruction can spread these inconsistent measurements into streaks. Electronic noise and correction behavior become especially relevant at very low signals.

A suitable exposure, voltage and modulation strategy can improve penetration and statistics. Positioning arms away from the torso when safe may reduce difficult paths. Smoothing or iterative reconstruction can reduce noise but changes resolution or texture and cannot guarantee recovery of information never adequately measured. Dose should be adjusted for the task and patient rather than increased indiscriminately.

Undersampling and aliasing

Undersampling occurs when measurements are too sparse to represent spatial variation adequately. Too few angular views or insufficient ray sampling can produce streaks, moiré-like patterns or false detail. The Nyquist principle requires sampling sufficiently finely for the signal frequencies to be represented, but actual scanner performance also includes finite focal spot, detector aperture and reconstruction bandwidth.

Flying focal spots, detector geometry and appropriate acquisition/reconstruction can improve supported sampling. Increasing the displayed matrix after acquisition does not create missing projection views. Similarly, a sharply enhanced image can look detailed while containing aliasing or ringing. Evaluate source data, scanner capabilities and task-specific quality rather than judging only pixel count.

Two applied distinctions

A tiny renal calculus reconstructed in a thick slice may blend with urine and become inconspicuous. A thin reconstruction from adequate existing acquisition can help without a repeat scan. By contrast, broad shoulder streaks in a low-count chest acquisition may require a review of positioning and exposure rather than a smaller display field alone.

When a high-frequency phantom pattern forms false broad bands, consider undersampling. More mAs improves noise but does not automatically improve angular sampling. When a uniform region instead has random fluctuations that improve with increased counts under otherwise identical conditions, statistical noise is the stronger explanation. Artifact identification is therefore a causal exercise: select the correction that addresses the mechanism, and verify that it improves the required image information.

References: AAPM CT image detail and noise tutorial, AAPM CT protocols.