An in-vitro study on factors influencing the accuracy of current intraoral scanners

Abstract

Intraoral scanning devices are an essential part of many digital pathways in dentistry. The aim of this first factorial comparative in-vitro study was to evaluate the influence of different jaw model configurations (model types: upper jaw [upper], upper jaw with palate [upper-pal], lower jaw [lower]) and examiners on trueness and precision of digital full-arch impressions across seven intraoral scanners. Reference models were scanned with Emerald Green [EME]; iTero 5D [ITE]; Medit i900 [MED]; Primescan [PRI]; Primescan2 [PS2]; Aoralscan 3 [SHI]; Trios 5 [TR5]. Each model type was scanned ten times (n = 10) per device, divided between two examiners with different experience levels. Trueness and precision were calculated using the (90_10)/2-percentile. Data were analyzed using a linear mixed-effects model with model type and examiner as fixed effects and scanning device as a random effect. The trueness values ranged from PRI (29.0 ± 3.7 μm, [upper-pal]) to EME (94.9 ± 18.5 μm, [lower]). Examiner and model type had no significant effect on trueness. For precision, the model type [upper-pal] showed a significant positive effect (p < 0.001), factor examiner had a significant effect (p = 0.04). The model type and examiner experience can influence the accuracy of digital impressions; however, the values observed were within clinically acceptable thresholds (< 100 μm). Including the palate in upper jaw scans can improve accuracy for certain intraoral scanners. Optimizing scanning strategies, exercising and considering jaw anatomy—especially the inclusion of the palate—may enhance scanners’ accuracy.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-24904-w.

Introduction

The transition from conventional to digital impression methods in dentistry has led to considerable research on their accuracy, particularly for full-arch impressions. Research indicates that while digital impressions demonstrate equal or even superior accuracy in short-span scans^1,2, there is still a debate whether conventional impressions still exhibit higher trueness in full-arch cases^3–6. A recently published in vitro study with current intraoral scanning systems still indicates the superior accuracy of conventional impression technique in such indications⁷. Therefore, the question is, which factors may influence the accuracy of intraoral scanning devices (IOS). In literature, several factors are discussed for this finding, such as error propagation from the anterior region, which provides only limited topological information, resulting in increased deformation towards the distal end of the dental arch⁸ and other influencing factors that can be categorized as operator-related and patient-related factors⁹. Operator-related factors include the examiners’ experience^10–15 and the scanning strategy^16,17, while patient-related factors include the material scanned (e.g., restorative materials)^18–20, moisture conditions²⁰, and the design of the tooth preparation^21,22.

To maximize accuracy, it is important to reduce negative influences while optimizing the positive ones during the digital impression. A key benefit of controlled in-vitro studies is their ability to standardize, eliminate, isolate, or vary these variables, allowing conclusions about their impact on accuracy. Up to date, only a few studies systematically investigated the impact of influencing factors onto the accuracy of IOS. To the authors knowledge previous investigations have either focused on upper jaw with palatal versus without palatal scans or on comparisons between mandibular and maxillary scans.

Using new generations of IOS, the aim of the study was to investigate whether the model type and the examiner influence trueness and precision, and whether there is an interaction between the two. The respective hypotheses were tested:

• H1t (Model type—Trueness): Model type has no effect on trueness.

• H1p (Model type—Precision): Model type has no effect on precision.

• H2t (Examiner—Trueness): Examiner has no effect on trueness.

• H2p (Examiner—Precision): Examiner has no effect on precision.

• H3t (Interaction—Trueness): There is no interaction effect between model type and examiner for trueness.

• H3p (Interaction—Precision): There is no interaction effect between model type and examiner for precision.

Because measurements were obtained using different devices, the factor measurement device was treated as a random effect; thus, all inferences concern effects averaged across devices.

Materials and methods

Custom-made complete-arch models of a typical lower and upper jaw were utilized as reference models^2,23 (Fig. 1).

Fig. 1.

Reference models upper and lower jaw. Customized complete-arch maxillary and mandibular models are used as reference models. For scanning the upper jaw with and without palate, the same reference model was used.

To approximate the optical properties of natural teeth, which have refractive indices (RI) ranging from dentin (1.49) to enamel (1.64)^24,25, the maxillary teeth were made from feldspar ceramic (Cerec Blocs; Dentsply Sirona) with an RI of 1.52²⁶, and the mandibular teeth from zirconia-reinforced glass ceramic (Celtra Duo; Dentsply Sirona) with an RI of 1.57²⁶. Furthermore, upper jaw teeth consist of unprepared teeth reaching to the second molars, whereas the lower jaw reference model included various types of single-tooth preparations. Tooth 41 and 46 were prepared for full contour crowns, tooth 44 for a mesio-occlusal-distal inlay and tooth 36 for a mesio-occlusal inlay. Tooth 37 had a class I preparation. The preparations, as well as the model configuration were performed as described by Zimmermann et al²³. The reference models were scanned using a high-resolution reference scanner (ATOS III, Triple Scan MV60; GOM)—which has been shown to provide trueness of 3 μm and precision of 2 μm for jaw-sized scans^27,28—resulting in the generation of the reference file (exported as standard tessellation language (STL) file).

Digital scans of the reference models were obtained using seven different intraoral scanning devices, representing a cross-section of systems across different price ranges and technological approaches (parallel measuring techniques versus triangulation/stereo vision techniques) currently available on the market: Emerald Green (PLANMECA [EME]); iTero 5D (Align Technology [ITE]); Medit i900 (Medit [MED]); Primescan (Dentsply Sirona [PRI]); Primescan2 (Dentsply Sirona [PS2]); Aoralscan 3 (Shining 3D [SHI]); Trios 5 (3Shape [TR5]). The scanning devices were utilized in accordance with the manufacturer’s recommendations, including calibration and scanning strategy. Preliminary testing was performed for all IOS to verify the optimal scanning approach. For PRI, however, an optimized scanning strategy was applied, based on preliminary testing, clinical experience, and, most important, according to prior studies proofing that segmenting the scan can improve accuracy^29,30. All scanning strategies, along with links to the manufacturer-recommended scanning paths, are provided in Table 1.

Table 1.

List of all test groups, including the scanners’ soft- and hardware, impression technique and the respective post-processing used to generate STL files.

Test group	System	Software	Impression technique	Postprocessing
EME	Emerald Green	Romexis v.6.4.7.129	As recommended by manufacturer https://www.planmeca.com/globalassets/planmeca/1-planmeca-usa-modernization/cadcam/dental-scanning/emerald-s/manuals/6.4-emerald-s-user-manual.pdf// (User manual) First, the occlusal surfaces were scanned from distal to mesial (slightly tilted forward), followed by the buccal surfaces, and finally the oral/lingual surfaces	Direct export to STL
ITE	iTero 5D	iTero Element Software v.2.8.40.229	As recommended by manufacturer https://storagy-itero-production-eu.s3.amazonaws.com/download/en/iTero-Element-5D-User-Manual.pdf (User manual) The complete occlusal surface was scanned, starting from either terminal molar. The scan continued by rolling to the lingual side to capture the entire lingual surfaces. The buccal side was then scanned from the molar to the midline. This procedure was repeated on the opposite side to complete the buccal surfaces. The arch was finished by rolling over the anterior teeth, beginning in the lingual area behind the cuspids and lateral teeth, and then rolling over to the buccal side. The same procedure was repeated for the opposite side (cuspids and lateral teeth)	Direct export to STL
MED	Medit i900	Medit Link v.3.0.4	As recommended by manufacturer https://www.medit.com/de/a-proper-scan-strategy-for-scanning-a-full-arch/ (User manual) The first step was to select the occlusal surface with the clearest shape to build the main scan data. When scanning the anterior teeth area, the scanner was moved toward the labial and lingual surfaces, centered around the incisal edge. When scanning the opposite posterior area, the occlusal surface with the clearest shape was also selected to begin scanning. During the scanning of the buccal and lingual surfaces, the tip of the scanner was tilted at a 45-degree angle	Direct export to STL
PRI	Primescan	Cerec Software v.5.2.9	Scanning strategy: Scan started at the oral surface of the canine, up to the contralateral distal tooth rotating the camera to the occlusal aspect and scanning the occlusal/incisal surface back to the starting point. The scanner was repositioned at the distal occlusal surface and rotated to the buccal aspect, scanning the buccal surface towards the starting point of the canine tooth. The final scanning path started disto-occlusally, turned orally. The camera was guided in a 90-degree angle to the tooth-surfaces towards the incisors. Same procedure was repeated for the contralateral half of the dental arch to complete the full arch scan	Direct export to STL
PS2	Primescan 2	Firmenware v.1.5.1 (update March 2025)	As recommended by manufacturer https://www.dentsplysirona.com/content/dam/master/product-procedure-brand-categories/digital-impression/product-categories/intraoral-scanners/ifu/DIM-IFU-Primescan%E2%84%A2-2-w-o-CD-(REF%206831015)-EN-6748037-2024-09-05.pdf (User manual) The scan was started from the occlusal surface in the incisal area, with the scanner tilted approximately 60° in the oral direction, and moved orally along the dental arch to the distal area. This procedure was repeated on the opposite side. Care was taken to ensure that the segments overlap at the starting point. The scanner was then guided in the occlusal direction from the distal area across the entire dental arch to the other side. To complete the scan, the scanner was tilted approximately 60° in the buccal direction and moved along the entire dental arch in the buccal direction	Direct export to STL
SHI	Aoralscan 3	IntraoralScan v.3.5.6.1	As recommended by manufacturer https://www.shining3d.top/download/TIajufeE062hp0oZ6CpOVg%3D%3D%3AZ36TB1zJSmzFb34uKqA2W36%2FL5QdtstmlHkHQqTQnu42eLi3N7qGSfP0fjZxxW6KDtM0Nx9XqImn9a5A8CjeTA%3D%3D (User manual) Scanning was performed by moving the scanner slowly while centering on the dental arch, capturing occlusal surfaces at a 90° angle and oral, then buccal surfaces at approximately 45°, and ensuring at least 30% overlap between frames to allow proper data stitching	Direct export to STL
TR5	Trios 5	Trios Software v.24.3.1	As recommended by manufacturer https://3shape.widen.net/view/pdf/bitoot5rkw/TRIOS-Module-User-Manuals---R1.0-B-DE.pdf?t.download=true&u=0vknty (User manual) For the mandible, scanning was performed first on the occlusal surfaces, tilting slightly in the anterior region. This was followed by scanning the lingual surfaces at an angle of approximately 60–90°, and finally the buccal surfaces For the maxilla, scanning also started with the occlusal surfaces, tilting in the anterior region. Next, the buccal and then the lingual surfaces were scanned. The palatal surfaces were scanned last, starting ventrally and proceeding in a zigzag pattern toward the posterior region	Direct export to STL

For each scanner, 10 scans (n = 10) were performed for the lower jaw model [lower], for the upper jaw model [upper] and for the upper jaw model including the palate [upper-pal], resulting in 30 scans per scanner. Power Analysis (G*Power v3.1.9.7; Germany; https://www.psychologie.hhu.de/arbeitsgruppen/allgemeine-psychologie-und-arbeitspsychologie/gpower) was conducted for group comparison based on the findings of Schmalzl et al.³¹. The given results resulted in a sample size 4 for alpha error 0.05 a power 0.9 for every group comparison. Consecutively two examiners did each five scans per group. Examiner 1 was highly experienced, with more than 20 years of intraoral scanning experience, whereas Examiner 2 was classified as experienced, with two years of experience in IOS scanning. The models were scanned in an unpowdered state. To prevent overheating of the devices and examiner fatigue, the scanners were used alternately in the same order, and regular breaks were taken. All scans were exported as STL files for further processing. To avoid confusion, the term model type is used throughout instead of jaw, as the upper jaw exists in two configurations (with and without palate). The variable model type therefore includes [lower], [upper], and [upper-pal]. The evaluation protocol was conducted in accordance with the methodology established by Ender et al. (2019)². To assess accuracy, the impressions from each test group were superimposed using the best-fit alignment method (GOM Inspect 2018 v2.0.1; Rev 111729; GOM; Germany; https://www.zeiss.com/metrology/en/software/metrology-software.html). Following this, signed 3D distances were calculated for each surface point and analyzed using 3D analysis software (GOM Inspect 2018 v2.0.1). Trueness (n = 5) and precision (n = 10) values for each group and model were determined using (90_10)/2-percentile values³², which in total corresponds to n = 210 trueness values and n = 420 precision values. Reasons for using (90_10)/2-percentile values are explained in the supplementary materiał.

Statistics was performed with R v4.4.3 and RStudio v2024.12.1 (Posit Software PBC, USA; https://www.r-project.org). Following tests were applied separately for trueness and precision: First, a multifactorial ANOVA was performed to check for variance homogeneity (Levene’s test of equality of variances). The results showed that this assumption had to be rejected. Therefore, a linear mixed-effects model was applied and the model was fitted to the data with model type, examiner and their interaction as fixed effects and device as a random effect (full LME model with formula (R – Wilkinson notation): Outcome ~ Model Type * Examiner + (1 | Device)). Additionally, to test the significance of the interaction term Model type x Examiner, the full LME model was compared with a reduced LME model that excluded the interaction term (Outcome ~ Model Type + Examiner + (1 | Device)). For that, a likelihood ratio test (LRT) using ANOVA was performed to compare the two LME models (Testing Hypotheses H3). Depending on the significance (p < 0.05), the final model was fit by restricted maximum likelihood (REML) for effect estimation. Residual diagnostics (Q–Q plots) were made to indicate violations of model assumptions. In the case, when the reduced model was used, pairwise contrasts between the main effects were analyzed via estimated marginal means (EMMs) and with Tukey adjustment for multiple comparisons (Testing Hypotheses H1 and H2). Similarly, when the interaction was significant, pairwise contrasts between the single effects were analyzed separately for Model type differences within each Examiner, and Examiner differences within each Model type, with Tukey adjustment for multiple comparisons. For each significant contrast, the raw effect in µm with 95% confidence intervals and the standardized effect size (Cohen’s d) (computed as the contrast divided by the model residual SD), with 95% CIs for d, were reported. Additionally, supplementary fixed-effects tables with estimates, CIs, and p-values have been provided. A post-hoc sensitivity analysis was performed to estimate the minimum detectable effect size (in µm) for both main and interaction effects that could be detected with 80% power at α = 0.05, based on observed residual standard error and group sizes.

Results

Trueness

Trueness values dependent on factor model type are shown in Fig. 2a and Table 2. The values ranged from 29.0 ± 3.7 μm ([PRI], upper-pal) up to 94.9 ± 18.5 μm ([EME], lower). Similarly, the trueness values grouped by examiner are shown in Fig. 3a and Table 3 with values ranging from 27.5 ± 6.3 μm ([PRI], Examiner 1) up to 93.9 ± 19.7 μm ([EME], Examiner 2).

Fig. 2.

Mean of (90_10)/2-percentile values. (a) Trueness, (b) Precision per intraoral scanning device, for model types: upper jaw [upper], upper jaw with palate [upper-pal], and lower jaw [lower]. Error bars represent ± standard deviation (SD).

Table 2.

Trueness and precision values by variable model type (upper jaw- [upper], upper jaw with palate- [upper-pal], lower jaw- [lower]) using the (90_10)/2-percentile.

Device	Model type
	Trueness			Precision
	Mean ± SD of (90_10)/2-percentile in μm			Mean ± SD of (90_10)/2-percentile in μm
	[upper]	[upper-pal]	[lower]	[upper]	[upper-pal]	[lower]
EME	92.9 ± 20.0	90.1 ± 13.2	94.9 ± 18.5	53.8 ± 19.6	54.3 ± 19.7	61.2 ± 23.3
ITE	69.7 ± 25.7	56.1 ± 10.3	87.2 ± 24.5	80.7 ± 31.2	35.9 ± 9.6	79.2 ± 34.1
MED	61.1 ± 7.0	61.5 ± 7.0	58.2 ± 11.8	36.2 ± 15.3	36.4 ± 15.3	32.1 ± 10.6
PRI	33.9 ± 13.4	29.0 ± 3.7	32.0 ± 6.3	28.0 ± 14.9	17.2 ± 4.9	18.1 ± 5.5
PS2	40.0 ± 18.0	44.8 ± 3.0	41.4 ± 8.4	42.1 ± 22.0	18.1 ± 4.4	24.7 ± 6.3
SHI	46.0 ± 6.6	47.4 ± 5.9	68.6 ± 22.6	31.8 ± 12.7	29.9 ± 6.9	27.6 ± 9.3
TR5	57.7 ± 15.8	57.7 ± 15.8	58.5 ± 13.1	36.8 ± 10.1	36.8 ± 10.1	47.8 ± 30.3

Values are given as mean ± SD, in μm.

Fig. 3.

Mean of (90_10)/2-percentile values. (a) Trueness, (b) Precision for each intraoral scanning device, grouped by examiner: Examiner 1 and Examiner 2. Error bars represent ± standard deviation (SD).

Table 3.

Trueness and precision values depending on variable examiner (Examiner 1, Examiner 2) using the (90_10)/2-percentile.

Device	Examiner
	Trueness		Precision
	Mean ± SD of (90_10)/2-percentile in μm		Mean ± SD of (90_10)/2-percentile in μm
	Examiner 1	Examiner 2	Examiner 1	Examiner 2
EME	91.6 ± 14.9	93.9 ± 19.7	52.5 ± 22.3	60.4 ± 19.0
ITE	69.1 ± 21.0	73.1 ± 28.0	67.0 ± 34.1	61.2 ± 34.0
MED	59.5 ± 8.9	61.1 ± 8.7	33.5 ± 13.5	36.3 ± 14.2
PRI	27.5 ± 6.3	35.7 ± 9.2	18.8 ± 11.0	23.4 ± 9.9
PS2	42.2 ± 7.7	41.9 ± 14.5	25.1 ± 13.4	31.5 ± 19.2
SHI	46.2 ± 5.6	61.8 ± 21.1	27.1 ± 8.1	32.5 ± 10.9
TR5	63.1 ± 14.4	52.8 ± 13.0	37.8 ± 11.6	43.2 ± 25.4

Values are given as mean ± SD, in μm.

The likelihood ratio test comparing the full LME model with a reduced LME model (excluding the interaction term) indicated a significant improvement in model fit when including the interaction term (LRT χ²(2) = 10.05, p = 0.0066; Type III F(2.193) = 5.02, p = 0.0075). Therefore, the full LME model was used and was specified as: Outcome(Trueness) ~ Model type * Examiner + (1 | Device). Residual diagnostics (Q–Q plots) did not indicate violations of model assumptions. The results of the fixed effects are shown in a table in supplementary material. For Examiner 1, trueness did not differ among models (all |Δ|≤ 0.7 µm; all Tukey adjusted p ≥ 0.98). In contrast, for Examiner 2, [lower] produced higher trueness values than both [upper] and [upper-pal]: [lower] vs [upper] + 10.4 µm (95% CI + 2.5 to + 18.3; p = 0.006) and [lower] vs [upper-pal] + 13.6 µm (95% CI + 5.7 to + 21.4; p < 0.001) (Table 4). [upper] and [upper-pal] did not differ for Examiner 2 (+ 3.2 µm; 95% CI − 4.7 to + 11.1; p = 0.61). Examining examiner differences within each model type, examiners differed only for [lower], with Examiner 2 exceeding Examiner 1 by + 10.6 µm (95% CI + 4.1 to + 17.2; p = 0.0016) (Table 4), whereas examiner effects were not significant for [upper] (+ 0.18 µm; p = 0.96) or [upper-pal] (+ 3.63 µm; p = 0.27).

Table 4.

Simple effect contrasts (Model differences within each Examiner and examiner differences within each Model type (Block) for Trueness: Estimate (µm), Standard Error (SE µm), 95% CI (µm), Cohen’s d with SE and 95% CI, and p (Tukey).

Block	Stratum	Contrast	Estimate µm	SE_µm	95% CI (µm)	d	SE_d	95% CI (d)	p
Examiner within Model type	[lower]	Exam1–Exam2	− 10.61	3.32	[− 17.16, − 4.06]	− 0.78	0.24	[− 1.25, − 0.30]	0.002
Model type within Examiner	Exam2	[upper]–[lower]	− 10.4	3.35	[− 18.30, − 2.49]	− 0.76	0.24	[− 1.24, − 0.28]	0.006
Model type within Examiner	Exam2	[upper-pal]–[lower]	− 13.58	3.32	[− 21.43, − 5.74]	− 0.99	0.24	[− 1.47, − 0.51]	< 0.001

Only the significant effects are shown (p < 0.05).

Precision

Precision values regarding the factor model type are shown in Fig. 2b and Table 2 and ranged from 17.2.0 ± 4.9 μm for group ([PRI], upper-pal) up to 80.7 ± 31.2 μm for group ([ITE], upper). The respective values for precision regarding the factor examiner ranged from 18.8 ± 11.0 μm ([PRI], Examiner 1) up to 67.0 ± 34.1 μm ([ITE], Examiner 1) (Fig. 3a and Table 3).

The likelihood ratio test comparing the full LME model with a reduced LME model (excluding the interaction term) indicated no significant improvement in model fit when including the interaction term (χ² = 1.348, df = 2, p = 0.51). Therefore, the final model included only main effects (formula: Outcome(Precision) ~ Model type + Examiner + (1|Device)). Model diagnostics for precision likewise revealed no assumption violations. The results of the fixed effects are shown in a table in supplementary material. The factor model type showed significant overall differences for [upper] > [upper-pal] by + 11.94 µm (95% CI + 6.50 to + 17.38; p < 0.001) and [lower] > [upper-pal] by + 9.28 µm (95% CI + 3.83 to + 14.72; p < 0.001) (Table 5), while [upper] ≈ [lower] (+ 2.66 µm; 95% CI − 2.79 to + 8.12; p = 0.484). Examiner also showed a statistically significant overall effect, with Examiner 2 higher than Examiner 1 by + 3.95 µm (95% CI + 0.24 to + 7.67; p = 0.037).

Table 5.

Main Effect contrasts for Precision: Estimate (µm), Standard Error (SE µm), 95% CI (µm), Cohen’s d with SE and 95% CI, and p (Tukey).

Contrast	Estimate (µm)	SE (µm)	95% CI (µm)	d	SE_d	95% CI (d)	p
[upper]–[upper-pal]	11.94	2.31	[6.50, 17.38]	0.62	0.12	[0.383, 0.853]	< 0.001
[upper-pal]–[lower]	− 9.28	2.31	[− 14.72, − 3.83]	− 0.48	0.12	[− 0.716, − 0.245]	< 0.001
Exam1–Exam2	− 3.95	1.89	[− 7.67, − 0.24]	− 0.20	0.10	[− 0.397, − 0.012]	0.0371

Only the significant effects are shown (p < 0.05).

Discussion

This in-vitro study aimed to investigate the influence of the factors model type and examiner on both trueness and precision, as well as the interaction between these factors, while accounting for the variability introduced by the measurement device.

All hypotheses were tested using linear mixed-effects models. For precision, the interaction between model type and examiner (H3p: no interaction effect) was not significant, allowing interpretation of main effects. The null hypotheses stating that model type (H1p) and  examiner (H2p) have no influence on precision were both rejected. For trueness, the interaction hypothesis (H3t: no interaction effect) was rejected, indicating that the effect of model type on trueness depends on the examiner and vice versa. Consequently, the null hypotheses for main effects (H1t and H2t) could not be tested directly and were instead evaluated through simple effects. The findings with simple effects support rejection of the null hypotheses H1t and H2t.

The present study is closely related to a previous in-vitro investigation by Buhl et al. (2025)⁷, which used the same IOS scanning devices and models. However, in that study, data from both maxillary and mandibular models, as well as from different examiners, were pooled to account for variability that may occur in clinical settings. This approach aligned with the primary aim of that initial study, which was to address three key comparisons:

1. Differences between conventional impressions and intraoral scanner (IOS) impressions,

2. Systematic differences between trueness and precision values, and

3. Correlations among different 3D evaluation parameters, such as mean absolute value, absolute 80th percentile, and the (90_10)/2-percentile.

In contrast, the present study aimed to determine how, for specific IOS devices, the factors model type and examiner influence accuracy—i.e., trueness and precision. Consequently, the conceptional and statistical approach differed substantially. Here, trueness and precision were calculated for subgroups (n = 10 comparisons per subgroup) rather than for pooled data (n = 45). Additionally, this study measured and introduced an extra model type to assess the effect of scanning the upper jaw with or without the palate, which was not examined in Buhl et al.⁷. Overall, the findings of this study provide a more detailed insight into how different IOS devices perform under varying model types and examiner conditions. Together, both studies form a basis for designing future experiments that can more specifically target relevant influencing factors.

The values of the present study indicated that the variability in (90_10)/2-percentile due to differences between devices in the statistical LME model is minimal. The standard deviation of the random effect (device) is in the same range as the standard errors of the fixed effects. The variance is quite small, therefore underlining the significant influence of above-mentioned effects on the IOS scanning devices.

In this in-vitro study potential influencing factors were isolated and analyzed as fixed effects. The fixed effect model type included the patient-related factors scanned jaw ([upper], [upper-pal], [lower]), preparation design and scanned material.

Although several statistically significant effects were observed, it should be acknowledged that the study may have been underpowered to detect smaller differences, particularly in interaction effects and examiner-specific contrasts. A post-hoc sensitivity analysis revealed that, for interaction-level comparisons (n = 5 per cell), the minimum detectable effect size is approximately 27.5 µm for trueness and 39.0 µm for precision. In contrast, for main effects—where sample sizes were larger due to aggregation across devices (n = 70 for model type, n = 105 for examiner)—the study was sufficiently powered to detect differences as small as 6.5 µm (trueness) and 7.5 µm (precision), respectively. These thresholds support the robustness of the detected main effects in case of precision for [upper]-[upper-pal] (11.94 µm) and [upper-pal]-[lower] (-9.28 µm), while highlighting the limited sensitivity for Examiner1-Examiner2 (-3.95 µm) and the subtle interaction effects in case of trueness.

The fixed effect of the model type [upper-pal] has statistically significant influence on scanners’ precision. Especially in group PS2 and ITE the precision of maxillary scans with palate was higher than those of the mandibular scans and maxillary scans without palate (Fig. 2b). A plausible explanation for this observation could be the stabilizing effect due to an additional transversal stitching of the single images. Similar findings were reported by Schimmel et al.¹² and Schmalzl et al.³¹.

In contrast the cross-arch comparisons with patient-related factors like scanned material and preparations did not show an effect within the linear mixed-effects model; no statistically significant differences in trueness and precision were found between [upper] and [lower] jaws. A limitation in the assessment is that cross-arch comparisons are influenced not only by differences in material but also by the presence of prepared teeth in the lower jaw versus unprepared teeth in the upper jaw. Consequently, the individual effects of material and preparation design cannot be isolated, since both factors are confounded in cross-arch comparisons.

Previous studies have suggested that certain material properties, such as translucency³³ and refraction indexes³⁴, can influence IOS accuracy. Revilla-León et al.¹⁸ analyzed the effect of gold, cobalt-chromium, zirconia, and PMMA on the accuracy using an IOS (Trios 3) and reported the lowest scanning accuracy on the cobalt-chromium specimens, followed by gold, zirconia, and PMMA materials.

However, in the present study, materials with comparable optical properties—intended to mimic natural teeth—were chosen. While zirconia-reinforced lithium disilicate glass ceramic is known to exhibit greater translucency than feldspar ceramic^35,36 the differences in optical behavior appear too small to produce a statistical effect on full-arch accuracy. These results are in line with the findings of Yatmaz et al.¹⁹, who scanned full-arch models made of zirconium oxide and lithium disilicate with both Primescan and Trios 4. Their data showed that alumina oxide-abraded zirconia and glazed lithium disilicate surfaces did not significantly differ in terms of trueness within the same IOS system when scanning full arches. Nevertheless, in this study, the material and/or preparation design seems to have some impact on two out of the seven tested IOS devices showing larger discrepancies in terms of trueness between [upper] and [lower] jaw compared to the other IOS (Fig. 2a), even if not statistically significant.

The other fixed effect examiner is an operator-related influencing factor, which also influences the accuracy^2,37. To objectively assess this potential variability, scans were performed by two operators. Results indicated no significant user influence for trueness, however for precision the examiner had a significant effect. Overall, the more experienced Examiner 1 achieved slightly higher accuracy. This finding aligns with the study by Resende et al.¹⁵, which reported that more experienced examiner (medium and high experience) demonstrated higher precision than less experienced ones. These results contrast with the findings of Schimmel et al.¹² who observed differences in scanning accuracy when using the Primescan, where the less experienced examiner achieved higher accuracy than the more experienced examiner. The study by Gimenez et al.¹⁰ found that the examiner influenced the intraoral scanners’ accuracy. However, the examiner’s performance was not necessarily dependent on their level of experience.

The lack of significant differences in most of the examined IOS of the present study and the controverse literature may be attributed to advancements over time in scanning software, which is becoming less sensitive to external factors. Additionally, the fact that both operators had different levels of experience, but neither was inexperienced may have minimized the impact of examiner-related variability. Careful adherence to the manufacturer-recommended scanning strategy may have further reduced the examiner’s influence, as it’s known that scanning motion has significant impact on trueness and precision of full-arch scans³⁰.

The clinical accepted accuracy is often defined by a threshold of 100 µm^38,39, serving as a key benchmark for evaluating full-arch impression techniques. However, this threshold should be interpreted with caution, as clinically acceptable limits vary depending on the specific indication. Despite the significant differences observed the tested scanners generally fall within a very narrow range.

The present study has certain limitations regarding applicability. As it was an in-vitro study, the transfer to in-vivo is limited. In the controlled laboratory setting, many intraoral variables could not be replicated. In vivo, additional factors such as saliva, soft tissue movement, patient compliance, tongue interference, and dynamic changes in lighting conditions may significantly influence the scanning process and thereby affect the achievable accuracy. Furthermore, clinical practice often involves a heterogeneous mix of materials (e.g., enamel, dentin, composite, ceramics, and metals), which may cause reflection and scattering effects not fully reproduced in this study.

Therefore, further targeted clinical studies are necessary to validate these findings under real-world conditions. Specifically, investigations should aim to isolate and assess the impact of individual intraoral variables—such as moisture control, patient movement, or the presence of reflective restorative materials—on the full-arch accuracy of intraoral scanners. To better highlight possible operator-related influences, future studies could also involve novices to emphasize any influence of experience more clearly.

Conclusion

The investigated parameters model type and examiner have significant effects on precision, while having no significant main effects on trueness of tested IOS; however, some interactions between patient- and operator-related factors can influence trueness as well. Overall, the observed differences remain within a narrow range and are clinically acceptable (100 µm). Notably, including the palate when scanning the upper jaw has demonstrated a positive effect on the accuracy of certain intraoral scanning systems.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

A.M. and A.E. contributed to conceptualization and methodology. Project administration and investigation were performed by J.B. and A.E. The data curation and analyses were conducted by A.E. and A.M., the visualization was made by A.M. J.B. wrote the original draft, and all authors contributed to the review and editing. Supervision and validation were provided by A.M.

Funding

Open access funding was provided by the University of Zurich. The authors declare that no funds, grants, or other support was received during the preparation of this manuscript.

Data availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

Not Applicable. This study was conducted entirely as an in-vitro investigation, involving no human participants or animal subjects, and thus did not require formal ethics approval.