A CT artifact is a systematic distortion — streak, band, ring, grain — that is not anatomy. It can hide a finding or impersonate one. This page is the catalog. It is not a slice-thickness protocol and it is not the HU scale.
If you meant slice thickness as the cause of partial volume → CT slice thickness. If you meant HU drift from beam hardening → what Hounsfield units are.
Categories
| Category | Usual cause | Looks like |
|---|---|---|
| Physics | X-ray vs tissue (hardening, averaging, starvation) | Cupping, dark bands, grain, lost small objects |
| Patient | Motion; metal in or on the body | Blur, ghosting, starburst streaks |
| Scanner | A bad detector or a drifted cal | Rings, a line that repeats |
| Helical / cone-beam | MDCT geometry | Windmill at the edge; helical smear on a steep structure |
Physics
Beam hardening. The tube puts out a spectrum, not one energy. Soft X-rays die first in skin and bone; the beam that remains is harder. Reconstruction that assumes a single energy then paints a cup (center of a uniform object darker than the rim) and dark bands between two dense objects — the classic petrous-bone streaks. Nearby soft tissue can also read a lower HU than it should. That number caveat is on the Hounsfield page; the streaks and the cup live here. Vendor hardening correction is the usual software fix.
Partial volume averaging. One voxel, two tissues: the HU is the blend. A 4 mm lesion on a 5 mm slice can vanish into lung. Edges go fuzzy. The mm choice that causes it is slice thickness. The look — lost object, fake grey — is this catalog entry.
Photon starvation. Shoulders, hips, a large abdomen: so few photons reach the detectors that the picture is streaky grain. More mA helps and costs dose. Adaptive filtering and iterative reconstruction are the modern cleanup.
Patient
Motion. Voluntary: breath, swallow, a shift. Involuntary: heart, bowel, vessel pulse. The reconstruction then stacks inconsistent angles — blur, double outlines, streaks. Coaching a breath-hold is the first fix. Cardiac gating times the acquisition to the ECG; that freeze is also why temporal resolution exists as a pillar on resolution of a CT scan.
Metal and MAR. Fillings, hips, screws, clips stop the beam. Detectors behind the metal see a hole. Reconstruction fills it with bright/dark starbursts and a shadow that can hide the tissue you need. MAR (metal artifact reduction) finds the metal, estimates the missing rays, and rebuilds. It is the difference between “unreadable around the implant” and “we can look for infection next to the stem.”
Scanner and geometry
Ring. Third-generation geometry (detectors rotate with the tube): one dead or drifted element draws a circle as the gantry spins. That is a service call — air-cal, replace the element — not a patient-coaching problem.
Cone-beam / windmill. Wide cone, outer detector rows, big density steps: a star or windmill at the edge of the volume.
Helical distortion. A spiral acquisition can smear an object that sits steep to the scan plane. You would not see that smear on a classic axial step-and-shoot.
Daily air-cal and phantom scans (known densities, check uniformity and rings) catch scanner ghosts before a patient does.
What you do about it
- Recognize it so you do not call a petrous streak a bleed.
- Change the acquisition when you can: thinner slices for partial volume, breath-hold or gating for motion, more mA or iterative recon for starvation.
- Turn on MAR around metal. Dual-energy CT is another lever on hardening and metal; it is a technique name here, not a second product page.
- Recalibrate when the pattern is a ring.
PYCAD builds viewers that have to show a streak as a streak, not as anatomy. Case studies.