Digital model accuracy: surface versus CBCT

Simplified Orthodontics Academy · Draft prepared for review 30 October 2026

Accuracy is a property of a measurement

A digital model is not simply accurate or inaccurate. A number is produced by a chain: image capture, reconstruction, calibration, landmark placement, measurement path, and recording. Repeating the same chain may show good repeatability, while comparison with a reference model addresses agreement or accuracy. These are related but different questions.

For research reading, ask four questions: What was measured? Which model supplied it? What reference or comparator was used? Were the same landmarks and paths used by each examiner? A small mean difference does not by itself show that every individual measurement agrees, and a statistically significant difference does not automatically describe its practical importance.

Measurements that are often easier to reproduce

Clear linear measurements are usually easier to repeat than measurements that accumulate many local choices. Examples include a defined mesiodistal tooth width, a transverse distance between specified cusps, or a straight distance between two clearly visible landmarks. Even these measures depend on the exact endpoint, surface, tooth orientation, mesh resolution, and display scale.

Rossini and colleagues’ systematic review reported that many digital-model measurements agreed well with plaster models in the included studies. The review repeatedly identified landmark identification and representation of interproximal surfaces as important sources of variation. Its conclusion should be read as a qualified synthesis of heterogeneous studies, not as a guarantee for every scanner, software package, file, examiner, or sample. [1]

What Zhang 2016 actually showed

Zhang et al. compared full-arch intraoral scans with plaster models in 20 selected subjects. In that sample and workflow, anteroposterior differences were reported as less than 0.2 mm, while tooth-width and tooth-height differences were less than 0.1 mm. One lower intermolar-width comparison showed a difference of 0.28 mm, with P=0.005. [3]

Those figures describe that study’s subjects, capture process, software, reference, and measurement definitions. They are not scanner tolerances, acceptance limits, or universal evidence that all surface scans and plaster models are equivalent. The lower intermolar result is especially useful as a reminder that one measurement category can behave differently from another inside the same investigation.

Why arch perimeter and serial comparisons need more detail

Arch perimeter is a cumulative or curved measurement. Its value can change with the chosen path, how contact areas are crossed, tooth rotation, missing or worn surfaces, and small mesh defects. Two examiners can therefore obtain different values while both following a plausible method. A smooth-looking surface does not reveal whether the same geometric path was used.

Serial comparisons add another layer. Scan quality, scale, orientation, segmentation, occlusal registration, and landmark definitions must be comparable before a difference can be interpreted as a difference between models. A change near a contact point or an unclear gingival margin may reflect capture or marking choices rather than a change in the underlying anatomy. Reporting extra decimal places does not remove this uncertainty.

Interproximal measurements deserve particular attention. Saliva, restoration margins, incomplete capture, holes, smoothing, and mesh processing can alter the apparent contact. Record whether the distance is straight, projected, or curved, and preserve the original values when a measurement is repeated.

Surface models and CBCT-derived models answer different questions

A surface model represents visible external geometry from an intraoral scan, an impression scan, or a cast scan. It can support many crown-level linear measurements and arch dimensions when the capture and measurement protocol are consistent. It does not expose hidden root anatomy or reliably define structures that are not visible on the recorded surface.

A CBCT-derived dental model is reconstructed from volumetric radiographic data. Its result depends on voxel data, segmentation, thresholding, artifacts, reconstruction choices, and the original scan. Ferreira and colleagues’ systematic review and meta-analysis included five studies and judged their methodological quality poor to moderate. Agreement varied by measurement: intermolar width, crowding, and some tooth-width groups were reported as acceptable in the included evidence, while maxillary arch perimeter was an important exception. Reproducibility was better within one examiner than between examiners. [2]

This does not make CBCT a more accurate version of every digital cast. It is a different data source with different error pathways. Evidence from an intraoral surface scan should not be transferred automatically to a CBCT reconstruction, and a CBCT result should not be used to generalize about all surface models.

A transparent repeatability workflow

Before collecting a value, write down the landmark, scale, orientation, and path. State the tooth numbering system and whether the measurement is straight-line, projected, or curved. Define how worn, restored, rotated, partly missing, or poorly captured landmarks will be handled. Confirm that the file is calibrated and displayed at true dimensions rather than measured from a resized image.

Repeat a sample of measurements using the same protocol, and ask a second examiner to repeat selected measurements when inter-examiner variation is part of the research question. Keep the original and repeated values, report the differences, and identify whether the comparison concerns repeatability, reproducibility, or agreement with a reference. If an ambiguous landmark changes the result, report that sensitivity instead of hiding it behind a rounded number.

What the evidence does—and does not—support

The evidence supports a cautious statement: many routine linear measurements on controlled digital surface models can be reproducible, but performance varies by measure and method. Systematic reviews are summaries of their included studies, and their conclusions inherit differences in samples, devices, references, examiner training, and definitions. The CBCT evidence is particularly limited by the small number and quality of included studies.

There is no universal threshold in this article for calling a model accurate, accepting a device, or deciding whether two models are interchangeable. The study-specific values above should be reported with their source and uncertainty. This is educational and research content only; it does not diagnose, prescribe, recommend a device, or provide patient-specific guidance.

Draft for review by owner Dr Salama. Educational content for clinicians and students; no patient data or images are used.

References

1. Rossini G, et al. Diagnostic accuracy and measurement sensitivity of digital models for orthodontic purposes: a systematic review. Am J Orthod Dentofacial Orthop. 2016. doi: 10.1016/j.ajodo.2015.06.029. PMID: 26827972
2. Ferreira JB, et al. Accuracy and reproducibility of dental measurements on tomographic digital models: a systematic review and meta-analysis. Dentomaxillofac Radiol. 2017. doi: 10.1259/dmfr.20160455. PMID: 28362170
3. Zhang F, Suh K-J, Lee K-M. Validity of Intraoral Scans Compared with Plaster Models: An In-Vivo Comparison of Dental Measurements and 3D Surface Analysis. PLoS One. 2016;11(6):e0157713. doi: 10.1371/journal.pone.0157713. PMID: 27304976

More to read

• Study model analysis in Simplified CastContext for digital study-model measurements
• Reading a CBCT scanSeparate educational discussion of CBCT image interpretation
• Mixed dentition analysis explainedA different topic: estimated tooth widths in developing dentitions
• دقة النموذج الرقميThe same article in Arabic
• All articlesOrthodontic education and research notes

This article is for education. Clinical decisions remain the responsibility of the treating clinician.