Digital workflows have become part of everyday dental practice, and one of the newest tools to join the digital toolbox is the Facial Scanner (FS). By generating a fully digitalized three-dimensional model of a patient’s face, facial scanners allow clinicians to capture perioral soft tissue — the lips, cheeks and surrounding structures that CBCT and intraoral scanners simply cannot reproduce. When combined with intraoral scans and CBCT data, facial scans make it possible to build a complete three-dimensional “virtual patient,” supporting more informed treatment planning across prosthodontics, implantology, orthodontics and restorative dentistry.
But how reliable are these devices? And do the low-cost smartphone apps that promise the same result as professional systems actually hold up?
A new in-vitro study by Francesco Mintrone, Omar Shanableh, Giacomo Fabbri, Renaud Noharet and Luca Giannetti, published in the Digital Dentistry Journal (2026), set out to answer exactly that question, comparing five professional facial scanners with five popular scanning apps.
Why Trueness and Precision Matter
Before looking at the results, it’s worth clarifying three terms that are often used loosely but have distinct meanings in metrology:
- Accuracy – how close a single measurement is to an accepted reference value.
- Precision – how consistent repeated measurements are with each other, under the same conditions.
- Trueness – how close the average of multiple measurements is to the reference value, regardless of how much they vary.
A scanner can be precise without being true (consistently wrong) or true without being precise (correct on average, but noisy). A clinically reliable facial scanner needs to demonstrate both qualities: it must reproduce results close to reality, and it must do so consistently, scan after scan. Inaccurate scans are not just a technical curiosity — they can translate directly into treatment failures, such as dental restorations that don’t fit properly.
How the Study Was Designed
To keep the comparison controlled and repeatable, the researchers used a three-dimensional facial model made of polyvinyl chloride, built with dimensions, shape and proportions comparable to a real human face. Six local markers made of Cobalt-Chrome (coated in pink enamel to resemble skin tone) were permanently bonded onto the model at six anatomical reference points: the tragus, the ala of the nose, and the chin, on both the right and left sides. A convex spherical marker shape was chosen specifically because it is easier to scan accurately.
A reference file was captured using an industrial-grade laser scanner — the FARO® Edge 14000, an ISO 9001-certified device calibrated to ISO/IEC 17025:2017 standards — establishing the “ground truth” against which every other scanner would be judged.
Ten devices were then tested, six scans each
Professional facial scanners:
- Face Hunter (Zirkonzahn)
- Space Spider (Artec 3D)
- RayFace (Micerium)
- Easy Face Scan (Revopoint)
- MetiSmile (Shining 3D)
Consumer scanning apps:
- Metascan
- TRNIO
- Scaniverse
- MagiScan
- Polycam
The professional devices all use structured light technology, while the apps rely on photogrammetry — reconstructing a 3D object from multiple 2D photographs through feature detection and triangulation.
Each of the six scans per device was imported into Exocad DentalCAD, where a custom volumetric conversion library — developed in collaboration with New Ancorvis S.r.l. — converted each spherical marker into a corresponding conical geometry. This allowed precise, repeatable measurement of eight standardized linear distances (e.g., right ala–right tragus, right chin–left chin, right ala–right chin), giving a full three-dimensional picture of scanner performance along the X, Y and Z axes.
Statistical analysis was carried out in R, using linear mixed regression models, the intraclass correlation coefficient (ICC), the concordance correlation coefficient (CCC), and Bland-Altman analysis to estimate bias and limits of agreement.
The Apps Fell at the First Hurdle
Before any detailed measurement could even begin, a fundamental problem emerged with the consumer apps. Although the live scan preview looked convincing on-screen, once exported as .stl files, the app-generated scans showed dramatic differences in image resolution, and — more critically — were exported at the wrong dimensional scale, either oversized or undersized relative to the reference model.
This scaling error made it impossible to align the hemispherical markers from the app scans with the conical reference geometry built in Exocad. In practical terms: the measurements simply could not be taken, because the files weren’t dimensionally comparable to reality in the first place.
The scale of the discrepancy is striking. For example, comparing the bounding dimensions of the scans against the reference model (173.89 × 253.85 × 179.09 mm):
- Metascan: 690.75 × 547.58 × 1034.76 mm
- Polycam: 0.72 × 1.06 × 0.54 mm
- TRNIO: 0.23 × 0.21 × 0.17 mm
- Scaniverse: 0.175 × 0.21 × 0.13 mm
- MagiScan: 0.57 × 0.99 × 0.64 mm
A preliminary check using 3Shape Viewer confirmed that only the professional devices produced scan files that were correctly scaled to the reference — the apps, without exception, were not. As a result, the five apps were excluded from the subsequent measurement phase entirely, and the researchers concluded that meaningful clinical use of these apps for facial scanning currently remains unsuitable given the tested limitations.
The Professional Scanners: Excellent Global Performance
With the five professional scanners, the picture was very different.
Global trueness (across all eight linear distances)
- The intraclass correlation coefficient (ICC) was 1.000 for every scanner tested — a perfect score, indicating excellent agreement with the reference and excellent repeatability.
- The concordance correlation coefficient (CCC) was also equal to 1 for all scanners, confirming both strong correlation and minimal systematic bias relative to the reference.
- When comparing average linear distances across the six scans to the reference values, most scanners showed no statistically significant difference. The single exception was Easy Face Scan, which showed a statistically significant difference (p = 0.003).
- Even for Easy Face Scan, the actual discrepancy was small: an average underestimation of only 0.304 mm (95% CI: −0.497 mm to −0.110 mm) — meaning the generated 3D model was very slightly smaller than the real object. Since deviations below 0.5 mm are generally regarded as clinically insignificant, the authors note that this statistically significant result has limited practical relevance.
- All five scanners also demonstrated high global precision, with no statistically significant differences across the six repeated scans for any device.
Local trueness (measured at each individual marker point)
Here, the picture becomes more nuanced. When trueness was assessed locally — at the level of each specific anatomical reference point rather than the distances between them — all five scanners showed statistically significant discrepancies compared to the reference (p < 0.001). However, for most devices these differences were very close to zero in absolute terms.
Ranking the professional scanners by local trueness (best to worst):
- Space Spider (Artec 3D) — highest local trueness (mean difference: 0.024 mm)
- MetiSmile (Shining 3D) — mean difference: 0.027 mm
- Easy Face Scan (Revopoint) — mean difference: 0.057 mm
- Face Hunter (Zirkonzahn) — mean difference: 0.101 mm
- RayFace (Micerium) — mean difference: 1.203 mm — notably higher than the other four
RayFace was the clear outlier, showing local discrepancies more than ten times larger than the next closest scanner, suggesting potential limitations for treatments that demand very high spatial resolution, such as orthodontics or prosthodontics.
Local precision (how consistent each scanner was across its six repeated scans, independent of the reference value) was high across the board, with no statistically significant differences detected between measurements (p > 0.05 in all cases).
What This Means in the Clinic
The clinical takeaway is direct: professional-grade structured-light facial scanners can be trusted to build an accurate, reproducible digital representation of a patient’s face, supporting facially driven treatment planning in prosthodontics, orthodontics and implantology.
This matters for several everyday clinical tasks:
- Evaluating facial aesthetics — lip position, facial midline, nasolabial angle, profile — as part of restorative or prosthetic planning.
- Accurately assessing the vertical dimension of occlusion (VDO) directly from a face scan, supporting more individualized prosthetic decisions.
- Contextualizing tooth position on the facial planes and applying the “rule of thirds” in extensive rehabilitations.
- Locating the Camper’s plane through cutaneous landmarks.
- In implantology, integrating soft-tissue information with a validated 3D facial model to support prosthetically driven implant positioning.
These findings are consistent with a broader systematic review and network meta-analysis (Canullo et al.), which found facial scanning technologies generally comparable to traditional anthropometric measurement, with deviations often under 1 mm, and suggested that an accuracy threshold below 0.60 mm can be considered clinically acceptable.
At the same time, the results are a clear caution against relying on consumer smartphone scanning apps for clinical facial scanning — at least in their current state. The dimensional scaling errors observed in this study weren’t minor rounding differences; they made the files fundamentally incompatible with clinical measurement software. The authors note that some more recent studies have reported more positive outcomes for these applications, so this remains an evolving area that warrants continued, independent testing rather than a permanent verdict.
Limitations and What Comes Next
The study itself acknowledges its boundaries: it was conducted on a rigid, artificial facial model, which removes the natural variability of a real patient (breathing, micro-movement, skin elasticity) but also means it doesn’t fully reflect real clinical scanning conditions. The authors call for future in-vivo studies to validate these findings on real patients, along with standardized clinical guidelines addressing trueness thresholds, compatibility standards, and best practices for integrating facial scanning technology into routine digital workflows.
Bottom Line
- Professional facial scanners (Face Hunter, Space Spider, RayFace, Easy Face Scan, MetiSmile) demonstrated perfect global trueness and precision (ICC = 1.000, CCC = 1) relative to an industrial-grade reference scanner, with only minor, clinically negligible local discrepancies — Space Spider by Artec 3D performed best overall, while RayFace by Micerium showed the largest local deviations.
- Consumer scanning apps (Metascan, TRNIO, Scaniverse, MagiScan, Polycam) produced files with such significant dimensional scaling errors that reliable measurement wasn’t possible at all in this study — a limitation the authors say should continue to be monitored as the technology develops.
- For clinical workflows where facial data feeds directly into treatment planning — from smile design to implant positioning — the evidence currently favors professional structured-light devices over low-cost photogrammetry-based apps.
Based on: Mintrone F, Shanableh O, Fabbri G, Noharet R, Giannetti L. “Evaluation of 3D facial scanners: trueness and precision.” Digital Dentistry Journal 3 (2026) 100054. Open access under CC BY 4.0.
The full study, including complete statistical tables and figures, is available to download below.