Radiographic caries depth classification

Radiographic caries depth classification grades how far a proximal radiolucency has penetrated the tooth on a bitewing radiograph. The common ladder runs E1 and E2 through the outer and inner halves of enamel, then D1, D2 and D3 through the outer, middle and inner thirds of dentine. The band sets what happens next.

Synonyms: E1 E2 D1 D2 D3; caries depth scoring; radiographic caries grading; proximal caries depth; caries depth classification

What is radiographic caries depth classification?

It is a way of turning a shadow into a decision. You look at a proximal surface on a bitewing, you judge how deep the proximal radiolucency reaches, and you put it in a band. The band then feeds the choice between watching the lesion, sealing it, or cutting.

The bands are taught as a ladder with crisp rungs. They are not crisp. The boundary you are being asked to judge is the enamel–dentine junction, and almost everything difficult about depth classification sits close to it.

What are the depth bands, and where does each boundary sit?

Two things get divided: enamel into halves, dentine into thirds. [6][16]

The E1/E2 boundary is the midpoint of the enamel thickness. The E2/D1 boundary is the enamel–dentine junction itself. The D1/D2 and D2/D3 boundaries are geometric — you're dividing the remaining dentine thickness by eye, between the junction and the pulp.

UK guidance frames the same judgement in the same terms: whether caries extends beyond the enamel–dentine junction into the outer, middle or inner third of dentine, and for advanced lesions, whether a clear band of normal-looking dentine still separates the lesion from the pulp. [16]

Why does a lesion look shallower than it is?

Because radiographic depth systematically understates true depth — against intra-operative assessment in patients, and separately, against histology on extracted teeth.

Bin-Shuwaish and colleagues measured 51 Class II lesions in patients against intra-operative photographs of the cross-sectioned lesion at its deepest point. Both the digital sensor and D-speed film significantly underestimated the clinical depth (p < 0.0001), with the digital images significantly closer to true depth than the D-speed film (p = 0.0031). [8]

The same direction shows up against histology. Measuring lesion depth in radiographs from four digital systems against histological sections, two untrained observers underestimated depth in every system tested, and each of two trained observers underestimated it in one. [7]

But hold the conclusion loosely for any single lesion. The authors of that study also concluded that a high probability exists that an individual measurement is either overestimated or underestimated relative to true depth. [7] The bias runs one way across a population of lesions. It does not guarantee that the one in front of you is deeper than it looks.

There's a second, blunter version of the same problem, and it is the same one behind the question of how early proximal caries can be detected at all. Pooling 104 datasets from 77 studies covering 15,518 tooth sites, the Cochrane review of imaging for early caries put summary sensitivity at 0.47 (95% CI 0.40 to 0.53) and specificity at 0.88 (95% CI 0.84 to 0.92), on low-certainty evidence. Applied to 1,000 surfaces with 63% enamel caries prevalence, that is 337 surfaces called disease-free when enamel caries was present, against 43 called diseased when it wasn't. [1] That review covers non-cavitated enamel-only lesions specifically: cavitated and obviously-into-dentine lesions were excluded, and only 17 of the 77 studies were done in a clinical setting. [1] It is not a figure for how well you see a D2.

For dentine lesions the picture improves. A separate meta-analysis of 117 studies found pooled sensitivity for proximal dentine lesions of 0.36 (0.24 to 0.49), rising above 0.60 for cavitated proximal lesions. [2] That paper carries a 2021 corrigendum, so treat its figures as indicative rather than settled. [3]

Which band boundary is hardest to call?

No study has published agreement rates boundary by boundary. What the evidence does support is that depth calls are unreliable in general, and that the reasons converge on the enamel–dentine junction.

When 27 licensed dentists scored 120 approximal surfaces twice on a four-point depth scale, against histological serial sectioning, inter-rater reliability was low: mean ICC 0.27 for intraoral bitewings and 0.19 for extraoral. Test–retest agreement was 0.48 and 0.44. For the smallest enamel lesions, the authors describe accuracy as just above random guessing. [9]

Trained examiners on an easier sample do much better. With three trained examiners scoring directly visible proximal lesions on a five-point radiographic scale, radiographic intra-examiner weighted kappa was 0.6 to 0.8 and inter-examiner above 0.6 — against above 0.9 and above 0.85 for visual ICDAS scoring of the same surfaces. Accuracy against histology was ≥0.45 for the radiograph and ≥0.85 for the visual score. [10] Training and a clean sample raise the ceiling. They don't make the radiograph competitive with looking at an exposed surface.

Three things point at the E2/D1 call specifically.

The Mach band effect, an optical illusion produced by lateral inhibition in the retina, appears in dentine along the proximal enamel–dentine junction. That is exactly where the E2/D1 decision is made. Perception varies between observers and shifts with projection, contour and density. [13]

Mejàre's longitudinal scoring system needed a state of its own for a radiolucency with a broken enamel–dentine border but no obvious progression in the dentine, sitting between an enamel lesion and one spreading into dentine. [5] A scheme built to track lesions over a decade found that territory needed naming.

And the shallowest band is independently the least reliably identified. A detection model trained on the ICCMS scoring system reached precision 0.35 and recall 0.23 for the outer-half-of-enamel class, against 0.77 and 0.61 for the deepest class. [6] That is a model's performance, not a dentist's — but it says something about how little signal the outermost band carries.

What does each band imply for the decision that follows?

The band is a proxy for two questions: is the surface cavitated, and how fast is it moving?

On cavitation, the reference figures come from Pitts and Rimmer, who compared bitewing appearance against direct visual inspection after a week of elastomeric separation. In permanent teeth, 0% of radiolucencies in the outer half of enamel, 10.5% in the inner half of enamel, 40.9% reaching the outer half of dentine and 100% reaching the inner half of dentine were found to be cavitated. In primary teeth the same four bands gave 2.0%, 2.9%, 28.3% and 95.5%. [4] Read those with the study in view: 211 children aged 5 to 15 in Scottish general practice, one observer, 1992, and bands drawn as halves of dentine rather than thirds. [4] The authors flagged the permanent-tooth numbers as needing larger samples themselves. [4]

A separate in vivo study found that all lesions judged radiographically to extend more than 1 mm into dentine were cavitated. [8]

On speed, Mejàre followed 536 Swedish children from 11 to 22 with annual bitewings under a remineralising rather than restorative policy. Once a lesion had broken the enamel–dentine border without obvious spread in dentine, 75% of surfaces survived only 1.3 years before reaching the outer half of dentine — against 4.8 years for a lesion still confined to enamel. Median survival across that transition was 3.1 years, and the caries rate over it was 20.3 new events per 100 tooth-surface-years, against 5.4 from an enamel lesion. [5]

That is the practical content of the E2/D1 boundary. Crossing it changes how quickly the lesion moves on.

What clinicians do with that varies. In a 1999 survey of 651 Swedish dentists, more than 90% said they would not automatically restore a proximal lesion in a low-risk adolescent unless it showed obvious progression into the outer third to half of dentine. [11] But thresholds differed by region, by the age of the dentist, and between private and public practice [11] — and a French survey found the same kind of variability persisting a decade apart, though on occlusal rather than proximal lesions. [12] The band informs the decision. It doesn't make it.

Which classification systems are in use, and do they agree?

Four schemes appear regularly, and they do not map onto each other cleanly.

The ICCMS radiographic scoring system uses R0 for a healthy tooth, then RA1, RA2 and RA3 for the outer half of enamel, the inner half of enamel and the outer third of dentine; RB4 for the middle third of dentine; and RC5 and RC6 for the inner third of dentine and involvement of the pulp chamber. [6] That maps onto the E1–D3 ladder one to one.

Pitts and Rimmer [4] and Mejàre [5] both divide dentine into halves rather than thirds. So a "D1" and an "outer half of dentine" are not the same territory, and a cavitation percentage reported against halves cannot be read straight onto a thirds-based band.

Mejàre's scheme carries a further complication: its state 3, a broken enamel–dentine border with no obvious progression in dentine, has no single counterpart in either thirds-based system. [5] It straddles the E2/D1 line rather than sitting on one side of it.

The practical consequence is small but real. When you read a figure, a cavitation rate or a progression time, check which scheme produced it before you attach it to the band you're looking at.

What it's confused with

Common errors

• Treating the band as a measurement. It's an estimate, and one that runs shallow. [7][8]
• Reading a number against the wrong scheme. Halves and thirds are not interchangeable. [4][5][6]
• Scoring through an overlap. Depth is the last call in a bitewing reading sequence, not the first. In one study of dental students, 38% of surfaces overlapped, and both positive and negative agreement were significantly lower on those surfaces (p < .001). [14] In a clinical comparison, 4.1% of proximal surfaces on intraoral bitewings were non-readable, defined as more than half the enamel layer overlapped or the surface not visible. [15] A surface you can't see is not a surface you scored E1.
• Trusting your own repeatability. Test–retest agreement among 27 licensed dentists was 0.48 on intraoral bitewings. [9] You will not always agree with yourself.