How do you tell an artifact from real pathology on a radiograph?

Telling a radiographic artifact from real pathology is a reading skill, and it is less settled than the tests for it suggest. Some of the radiolucencies you will be asked to call are not lesions at all — they are the beam, the geometry of the tooth, or your own visual system.

The tests get used as though they close the question: take another view, check whether it makes anatomical sense, mask the enamel and see what survives. Across the studies that have measured it, optical effects were read as caries or fracture in roughly 5% to 26% of observations, and the studies disagree enough that a 2025 systematic review graded the certainty of its own pooled estimate as low to very low.

What the question actually is

Not every radiographic artifact is a diagnostic problem. A cone cut, a herringbone artifact, a ghost image and elongation and foreshortening each have a name, a cause and an entry of their own.

The hard group is smaller and more specific: the effects that appear where a lesion would appear and look like one. On intraoral radiographs the 2025 review found that group is dominated by two optical effects — the Mach band effect and non-carious triangular-shaped radiolucencies, or TSR — alongside cervical burnout. They are not rare. Pooled across two prevalence studies, non-carious TSR were present on 26.44% of the maxillary molars examined (270/1021).

Each has a mechanism, and the mechanism is the reason the appearance is convincing. Cervical burnout is beam geometry: the beam passes tangentially across the curved proximal surface of the tooth and meets less dental structure there, so that region reads as a localised radiolucency. Nothing is missing from the tooth. The review notes the effect is more pronounced in multirooted teeth whose buccal and lingual roots do not overlap cleanly in the projection.

The Mach band effect is not in the image at all. It is the visual system exaggerating contrast where two densities meet — at the enamel-dentine junction, that reads as a band of apparent radiolucency. TSR are attributed to the Carabelli cusp superimposing on maxillary molar anatomy, and in the studies the review included they were most frequently observed on the mesial surfaces of maxillary molars. None were reported on mandibular molars or on distal surfaces.

Digital processing adds a fourth. When 16 radiographs of restored teeth were shown to 42 participants in both processed and unprocessed form, 52% of the confident "definitely no pathology" responses on the unprocessed images (81/157) moved to "pathology" once the images were processed.

The tests you are taught, and what each one is good for

Mask the enamel

For a suspected Mach band, the described manoeuvre is to visually cover the more radiopaque enamel next to the band. If the band goes with it, it is likely a false-positive optical effect; if it persists, it may be a real lesion in the outer dentine. The review describes this manoeuvre but includes no study that measured how well it works, so treat it as a way of interrogating what you are seeing rather than as a decision rule.

Take a second projection

This one deserves care, because the evidence behind it is not the evidence people think it is. The parallax principle — shifting the tube and seeing what moves — is well validated, for localisation. In a series of 522 patient records, checked against the surgical records of the cases that went to surgery, vertical parallax placed an impacted maxillary canine correctly in 97.7% of cases and horizontal parallax in 92.6%. That is a different task from deciding whether a radiolucency is real.

Two views can also disagree with each other. Comparing bitewing and oblique lateral radiographs in 53 patients aged 6 to 18, agreement on dentine caries reached Cohen's kappa of 0.64 and 0.53 for approximal surfaces and 0.72 and 0.56 for occlusal — fair to good, and lower than the reproducibility reported for bitewings alone. That study used film, in an era before digital sensors, so read the figures as a demonstration that a second view is another observation rather than as current performance data.

Check the anatomy

TSR are mostly a mesial-surface, maxillary-molar finding in the published samples. Cervical burnout sits where the beam thins the projected tissue. When a suspicious radiolucency appears somewhere an artifact is described and a lesion is not, that is information — but it is a prior, not a verdict.

Interrogate the processing

The authors of the image-processing study recommend that when a suspected finding sits on a high-contrast boundary, you look at unrelated high-contrast features elsewhere in the same image, and go back to the unprocessed data if there is doubt.

What the evidence shows

The 2025 review is the closest thing to a synthesis. Its meta-analysis found that optical effects were read as caries or fractures in approximately 13% of observations (60/464). The individual studies behind that spread ranged from 4.6% (5/109 teeth, in vivo, paediatric patients) to 12.8% (36/282 observations, in vitro) to 26.0% (19/73 participants, a simulated case in which a Mach band was presented as a fracture).

Who is reading matters. In an in-vitro study of 80 extracted human teeth read by eight observers across eight digital modalities and film, observers with no experience of digital radiography carried about six times the risk of a false positive (OR 6.09, CI 4.9-8.1). Observers in that study also spent longer on the surfaces they scored as false positives than on any other outcome (OR 1.05, CI 1.03-1.05). Because the teeth were extracted and validated against microscopy, treat those odds as a signal about observer experience, not as a chairside rate.

Error is also not one thing. A 2025 mixed-methods study of fifth- and sixth-year dental students separated the failures by stage: overlooking, inattentional blindness and satisfaction of search belong to visual scanning; recognition errors happen when something is seen but not correctly named; prevalence effects and decision errors happen in the reasoning that follows. In that cohort false negatives were the most frequent error type, not false positives — fifth-year students made 206 errors (7.92 ± 2.86) and sixth-year students 172 (6.62 ± 3.50), with a significant difference only in incorrect diagnoses (p = 0.041). Those are students working through an online test, so the counts belong to that setting.

Where it is unsettled

Start with how thin the base is. The 2025 review screened 640 records and included five studies. Heterogeneity was above 90%, and the review graded its own evidence low to very low, citing the observational design of the included studies, the inconsistency between them, and wide confidence intervals from a small number of studies. The authors attribute the spread in false-positive rates to three things at once: which optical effect was studied, the experience level of the observer, and the experimental conditions. That is why a single number does not exist here.

Two more gaps are worth naming.

The first is direction. The review reports that a Mach band can also raise the detection threshold for small features beside it, which would produce false negatives as well as false positives. That is reported inside the review as another group's finding rather than measured by it, and it complicates the framing of artifacts as purely a source of overcalls.

The second is what the manoeuvres are worth. The review's own recommendations for future research name observer training in distinguishing optical effects from true lesions as a need still to be met. Meanwhile one of the interventions people reach for did not help in the study that tested it: image enhancement had no significant effect on false-positive risk, and neither did the observer's type of employment.

So the split is not between two camps arguing. It is between what the tests promise and what anyone has actually measured — a gap the review names itself when it asks for studies on observer training.

What this means at the chairside

The ADA's patient-selection guidance puts the clinical examination first. Images are prescribed to answer a question the examination and the history have already raised for that patient, and the dentist who knows that history is the one positioned to make the judgement. Nothing in the artifact literature displaces that — the review's own conclusion is that clinicians should exercise caution and supplement radiographic findings with a thorough clinical assessment.

Which means the conservative call is usually the right one. A radiolucency in a place where an optical effect is described, on a tooth with no clinical corroboration, is a finding to treat with caution and check against a thorough clinical assessment. A finding on a high-contrast boundary in a processed image is worth checking against the unprocessed data before anything else happens. And a systematic search pattern does work the student error data is pointing at, because the errors that study found in scanning happen before any test of artifact-versus-lesion can be applied.

The one thing the evidence will not give you is a rule that ends the question on a single radiograph. If you want the specific case where this goes wrong most often, why cervical burnout is so easily mistaken for caries takes one mimic apart in detail.

The only way to get better at this call is to make it, repeatedly, on real radiographs — with the answer available afterwards. Start practising on annotated cases.