Accuracy and reproducibility of dental measurements on tomographic digital models: a systematic review and meta-analysis

Abstract

Objectives:

The aim of this systematic review with meta-analysis was to assess the accuracy and reproducibility of dental measurements obtained from digital study models generated from CBCT compared with those acquired from plaster models.

Methods:

The electronic databases Cochrane Library, Medline (via PubMed), Scopus, VHL, Web of Science, and System for Information on Grey Literature in Europe were screened to identify articles from 1998 until February 2016. The inclusion criteria were: prospective and retrospective clinical trials in humans; validation and/or comparison articles of dental study models obtained from CBCT and plaster models; and articles that used dental linear measurements as an assessment tool. The methodological quality of the studies was carried out by Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. A meta-analysis was performed to validate all comparative measurements.

Results:

The databases search identified a total of 3160 items and 554 duplicates were excluded. After reading titles and abstracts, 12 articles were selected. Five articles were included after reading in full. The methodological quality obtained through QUADAS-2 was poor to moderate. In the meta-analysis, there were statistical differences between the mesiodistal widths of mandibular incisors, maxillary canines and premolars, and overall Bolton analysis. Therefore, the measurements considered accurate were maxillary and mandibular crowding, intermolar width and mesiodistal width of maxillary incisors, mandibular canines and premolars, in both arches for molars.

Conclusions:

Digital models obtained from CBCT were not accurate for all measures assessed. The differences were clinically acceptable for all dental linear measurements, except for maxillary arch perimeter. Digital models are reproducible for all measurements when intraexaminer assessment is considered and need improvement in interexaminer evaluation.

Introduction

Dental models are essential for performing dental analysis. Characteristics such as tooth size, overjet, overbite, crowding, arch discrepancy and Bolton analysis are fundamental in diagnosis and for successful treatment results.^1–4 Furthermore, they are relevant for educational use, serve as documentation for orthodontic research⁵ and represent the legal document.⁶

Technological advancement has allowed the production of dental models in the digital format, which affects the orthodontic clinics⁷ by the current trend to replace the plaster models by three-dimensional (3D) digital models.^5,8 However, their comparability to plaster models needs to be proved since the measurement procedure with digital caliper in plaster models is considered the gold standard or reference standard in orthodontic research.^{1,3,5,7,9–12} Plaster models also have advantages such as easy and low cost production.¹³

Since the introduction of digital models in 1990,¹⁴ studies have been extensively performed to investigate models obtained by laser scanning,^3,7,12,15 holographic scanning,^16,17 stereophotogrammetry,¹⁸ intraoral scanning¹⁹ and, most recently, CBCT.^{1,8,9,19–30}

The use of such different types of digital models has increased due to their advantages over plaster models. They are not susceptible to physical damage or degradation, the digital archive facilitates communication between professionals and with patients, and eliminates the need and cost of physical space to store them.^14,18,31,32 In addition, it is possible to perform virtual setups to simulate results of orthodontic treatment;^32–35 apply qualitative-measures ordinal tests, as the indexes or orthodontic scales,^13,14,23,32 and superimpose images to follow up progress and monitor treatment outcomes.³⁶ Digital images of CBCT scan also add information, such as bone levels, root positions and the condition of the temporomandibular joint.³⁷ However, it should be considered that these models omit the soft-tissue outline, therefore there is more information of hard tissue in comparison to plaster models. Furthermore, it is still questionable whether the use of software to perform measurements and calculations facilitates and accelerates the process of dental analysis.^9,38

The main disadvantages of digital models are that their acquisition depends on a high-cost technology, although quickly evolving, and has limitations related to security and privacy, as well as to the risk of loss of electronically filed information.³⁸

There is no consensus between systematic reviews on which types of digital models would be applicable in orthodontics. Two reviews evaluated the validity of digital models^31,32 and two assessed the validity and reliability of these models.^14,15 Numerous types of acquisition of digital models have been covered on the existing studies included on systematic reviews, and only one review individualized a type of acquisition of digital models to be evaluated.¹⁵ Studies evaluating measurements on digital models generated by CBCT scan were included in two reviews;^14,15 however, there was no standardization in relation to how to obtain these models through tomography and what types of models were compared. These differences make it difficult to obtain a clear result with regard to accuracy and reproducibility of digital models.

Considering the technological development and increased use of CBCT in orthodontics, the production of digital models from the patient's CBCT scans was made possible,^9,21,37 avoiding the acquisition of impressions and eliminating the disadvantages of this procedure.^21,23,37 As a result of this technological progress, many studies have been conducted using digital models obtained from the patient's CBCT scan.^{1,8,9,20,22,25–30,39,40} However, no systematic review aimed at analyzing the accuracy and reproducibility of linear dental measurements in these models compared with plaster models. Possibly, this occurs because of occlusal distortions in dental models generated from CBCT scan and limitations that are associated with the individuals submitted to this examination. However, the constant technological evolution of CBCT may make it a single-examination diagnostic method.³⁷

The terms accuracy and reproducibility were considered in this systematic review for the evaluation of measurements. Accuracy refers to the agreement between the values obtained on the index test and reference standard, and it can be replaced by the term validity. The term reproducibility refers to the precision of successive measurements on the same test.^41,42

The purpose of this systematic review and meta-analysis was to assess the accuracy and reproducibility of linear dental measurements on digital study models generated from CBCT scan compared with plaster models.

Methods and materials

The report of this systematic review was based on the guidelines and directives of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses: the PRISMA statement.⁴³ The protocol was registered in the PROSPERO database under the number CRD42015029859 (http://www.crd.york.ac.uk/PROSPERO).

The selection of keywords for the electronic search and the eligibility criteria were based on the following question elaborated with the PICO/PECO format: in orthodontic patients' (P), CBCT scans (E) are dental linear measurements performed on digital study models generated from those examinations compared with the same measurements on plaster models (C) accurate and reproducible (O)?

The following were the criteria for inclusion of eligible studies: (1) prospective and retrospective clinical trials in humans, (2) validation and/or comparison articles of dental study models generated from CBCT scans and plaster models and (3) articles that assessed accuracy and/or reproducibility of dental linear measurements between dental study models generated from patient's CBCT scan and plaster models. The following were the criteria for exclusion: (1) animal studies, (2) in vitro studies, (3) case reports, (4) case series, (5) review articles, (6) opinion articles, (7) editorials, (8) books, (9) technical articles and (10) articles with measurements in direct images of CBCT without obtaining the digital dental study models.

The literature search to identify studies was conducted through electronic search in the following databases: Medline database (via PubMed), Scopus, Web of Science, VHL, Cochrane Library and System for Information on Grey Literature in Europe. Search strategies for each base were prepared with the help and guidance of a librarian using keywords and free terms related to the subject of research to reach the largest number of articles related to the review question. There were no restrictions for language. The search strategy for Medline database is provided in Table 1. This filter was applied due to the development of CBCT in 1998.⁴⁴

Table 1.

Search strategy for Medline database (via PubMed)

Medline

(((((((((Orthodontics[MeSH Terms]) OR Odontometry[MeSH Terms]) OR “Dental Arch”[MeSH Terms]) OR Orthodontic[Title/Abstract]) OR Odontometry[Title/Abstract]) OR Skulls[Title/Abstract]) OR Teeth[Title/Abstract])) AND ((((((“Cone-Beam Computed Tomography”[MeSH Terms]) OR “Imaging, Three-Dimensional”[MeSH Terms]) OR “Cone-Beam Computed Tomography”[Title/Abstract]) OR “Three-Dimensional Imaging”[Title/Abstract]) OR CBCT[Title/Abstract]) OR 3D[Title/Abstract])) AND (((((((((“Dental Models”[MeSH Terms]) OR “Image Processing, Computer-assisted”[MeSH Terms]) OR “Dental Models”[Title/Abstract]) OR “Dental Casts”[Title/Abstract]) OR “Digital Models”[Title/Abstract]) OR “Digital Model”[Title/Abstract]) OR “Study Models”[Title/Abstract]) OR “Virtual Models”[Title/Abstract]) OR “Plaster Models”[Title/Abstract])

Duplicate articles were eliminated after the final result of the search. Then, the titles and abstracts were read independently by two reviewers and the articles that met the inclusion criteria were selected for reading in full. When the title and abstract did not provide sufficient information for eligibility, the research was also examined in full. A manual search was conducted in the references of the selected articles to identify possible relevant titles not found by electronic search. Disagreements between the two reviewers were resolved by a third reviewer in consensus meeting.

The methodological quality of the selected studies was evaluated by two reviewers, through the Quality Assessment of Diagnostic Accuracy Studies checklist (QUADAS-2),^45,46 using the Review Manager software v. 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

QUADAS-2 consists of two main topics for analysis: “Risk of bias” and “Concerns regarding applicability”. From these topics, the following domains are filled for each study: patient selection, index test, reference standard, and flow and timing.

The evaluator can formulate new questions or remove existing questions. Then, the question “If a threshold was used, was it prespecified?” was excluded from the second domain (index test), and five others questions were added to the same domain to extract additional data for the assessment of studies: was the field of view described; was the CBCT scanner described; was the voxel size described; was the type of software for models acquisition described; was the type of software for models measurements described? (Appendix A).

Each study was assigned “yes”, “no” or “unclear” for each question. The risk of bias was judged as low when all questions of each domain were answered “yes”. A potential risk of bias was assigned when any question was answered “no”. The “unclear” category was used when insufficient data prevented judgement.

In this systematic review, we set that values >0.5 mm would be clinically significant for mean differences of mesiodistal tooth width and transversal widths, and >2.0 mm for sums of measurements.^4,47

Data were extracted from selected articles and organized in tables, according to the following aspects: author and year of publication, sample size, characteristics of participants, index test, reference standard, numbers of examiners, readings per examiner, measurements and findings.

The meta-analysis was performed gathering studies which validated the measurements of crowding in dental arch, intermolar width, Bolton discrepancy and mesiodistal tooth width in different subgroups. The mean difference was presented subtracting the measurements of plaster models from the measurements of the digital models generated from CBCT scan in all studies. The data were assessed in the Comprehensive Meta-Analysis software v. 3.2.00089 (Biostat Inc., Englewood, New Jersey,USA) with the random effects 4 model, and the heterogeneity among studies was tested through Q-value, I² index and tau.² When the p-value was <0.05, differences were considered statistically significant. When the heterogeneity was high, a sensitivity analysis was performed.

Results

Study selection

The search of all databases provided a total of 3160 studies. The duplicates were removed and 2606 titles and abstracts were read, and 12 included studies met the eligibility criteria. After full text reading, seven articles were excluded for the following reasons: (1) no linear measurements were presented;⁴⁸ (2) the digital models was generated from the axial CT scan;⁴⁹ (3) measurements were performed in CBCT images without obtaining the digital dental study models.^2,50–53 A flow diagram describing the selection process and the results is illustrated in Figure 1. No article was found in the manual search. To clarify doubts about the digital model generation, the authors were contacted^50,53 with no success.

Figure 1.

Flow diagram of selected studies (Preferred Reporting Items for Systematic Reviews and Meta-Analyses). BBO, Brazilian Dentistry Bibliography; LILACS, Latin American and Caribbean Health Sciences Literature; VHL, Virtual Health Library.

Study characteristics

The selected studies were conducted in Republic of Korea,²⁵ China,²⁶ USA^1,9 and Canada.⁸ Three studies were published in English,^1,8,9 one was published in Korean and the last one in Chinese.^25,26

Five comparative articles were found as a result of the systematic literature search and eligibility criteria.^1,8,9,25,26 All of them included dental linear measurements comparing the investigated methods, and samples of the dental plaster models and models generated directly from patient's CBCT scans. The latter were reported as digital imaging and communications in medicine files,^8,26 STL files^1,26 or not specified.^9,25

The sample size ranged from 20 upper models to 30 sets of models. All selected studies presented a sample of orthodontic patients with permanent dentition. One study²⁶ included patients with crowding up to 5 mm, and the other two studies^1,25 excluded dentition with metal restorations that damaged the image quality. All studies presented CBCT scans with a voxel size of 0.3 × 0.3 × 0.3 mm.^1,8,9,25,26

All five selected studies assessed the accuracy and intraexaminer reproducibility,^1,8,9,25,26 and three studies evaluated the interexaminer reproducibility^1,8,9 of linear measurements. Four digital model systems were assessed in the studies: Accurex,²⁵ Mimics,²⁶ Rapidiform¹ and Anatomodels.^8,9 The number of examiners that performed the measurements ranged between one and five, and the number of reading repetitions were not uniform between the studies. The data extracted from individual studies was presented in Table 2.

Table 2.

Data extracted from studies included

Author, year	Sample size	Characteristics of participants	Index test/reference standard	Examiners (readings per examiner)	Measurements	Findings
Lim and Lim (2009)25	20 sets of models	Orthodontic patient. Permanent dentition with no interproximal metal restorations	Accurex/digital caliper	One examiner (repeated twice at 2-week intervals)	Maxillary and mandibular arch: tooth size (first molar to first molar), sum of anterior teeth, sum of all teeth, required space, available space, arch length discrepancy, intercanine width, intermolar width and Bolton discrepancy (anterior and overall)	CBCT images showed significantly smaller overall Bolton discrepancy measurements. The ranges of measurement errors of the digital models generated from CBCT were clinically acceptable. These models can be used for model analysis.
Lv et al (2012)26	20 upper models	Orthodontic patients with fully erupted permanent dentitions; crowding up to 5 mm	Mimics/digital caliper	One examiner (second measurement made 2 weeks later)	Tooth crown widths, dental arch width, dental arch length and crowding	The consistency of the results was high, and some differences were accepted clinically. Therefore, it is possible to substitute PM for digital models obtained from CBCT. The ICC from repeated measurements was <0.75. The measurements obtained from these digital models were smaller than on PM.
Akyalcin et al (2013)1	30 sets of models	Orthodontic patients between the ages of 14 and 30 years, and without metal restorations	Rapidiform/digital caliper	Two examiners working independently (repeated all measurements twice 3–4 weeks later)	Maxillary and mandibular arch length discrepancies (crowding)	The findings demonstrated that CBCT-generated models are not perfect yet: mean bias (−0.45 ± 1.10) and confidence interval (−2.65 to 1.74). The correlation between CBCT-generated models and PM is 0.85.
Grunheid et al (2014)9	30 upper models	Patients with a variety of typical malocclusions, fully erupted permanent dentitions including incisors, canines, premolars and first molars	Anatomodels/digital calipers	Three (second measurement made 3 weeks later in six cases randomly selected)	Mesiodistal tooth width	The bias, limit of agreement interval and mean squared error from repeatability comparisons of measurements performed on PM were 0.048 mm, 0.704 mm and 0.035 mm, respectively, and the values for their comparisons in digital models generated from CBCT were 0.06 mm, 0.987 mm and 0.063 mm. The bias, limit of agreement interval and mean squared error from comparison of measurements obtained from PM and digital models generated from CBCT were 0.191 mm, 1.521 mm and 0.181 mm, respectively.
Luu et al (2014)8	30 sets of models	Patients with fully erupted permanent dentition	Anatomodels/digital caliper	5 (1 examiner repeated the measurements in 10 subjects 5 times at intervals of 10 days, with assessments limited to 5 cases per day)	Intermolar width, intercanine width, mesiodistal widths, arch perimeter, arch crowding, Bolton 6, Bolton 12, overbite and overjet	SMA using digital models obtained from CBCT was reliable but not always valid when compared with PM. The validity of measurements on digital models compared with PM showed low mean differences. The ICC demonstrated excellent intraexaminer reproducibility and good to excellent interexaminer reproducibility.

CBCT, Cone Beam Computed Tomography; ICC, Intraclass Correlation Coefficient; PM, Plaster Model; SMA, Study Model Analysis.

Risk of bias of included studies

According to QUADAS-2, the risk of bias is low to moderate. The domains that may introduce bias were patient selection, index test and reference standard (Figures 2 and 3). No concerns regarding applicability were detected in all included studies. The lack of description of sample randomization^1,25,26 and blinding^1,8,25,26 were the main sources of bias. Likewise, the field of view⁸ and appropriate interval between the methods²⁶ were not described in specific studies (Table 3). No article was eliminated after QUADAS-2 tool application.

Figure 2.

Risk of bias and applicability concerns summary.

Figure 3.

Risk of bias and applicability concerns graph.

Table 3.

Quality assessment of included studies

Signalling question	Lim and Lim (2009)25	Lv et al (2012)26	Akyalcin et al (2013)1	Grunheid et al (2014)9	Luu et al (2014)8
Was a consecutive or random sample of patients enrolled?	Unclear	Unclear	Unclear	Yes	Yes
Was a case–control design avoided?	Yes	Yes	Yes	Yes	Yes
Did the study avoid inappropriate exclusions?	Yes	Yes	Yes	Yes	Yes
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear	Unclear	Unclear	Yes	Yes
Was the field of view described?	Yes	Yes	Yes	Yes	Unclear
Was the CBCT scanner described?	Yes	Yes	Yes	Yes	Yes
Was the voxel size described?	Yes	Yes	Yes	Yes	Yes
Was the type of software models acquisition described?	Yes	Yes	Yes	Yes	Yes
Was the type of software model measurements described?	Yes	Yes	Yes	Yes	Yes
Is the reference standard likely to correctly classify the target condition?	Yes	Yes	Yes	Yes	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear	Unclear	Unclear	Yes	Unclear
Was there an appropriate interval between the index test and reference standard?	Yes	Unclear	Yes	Yes	Yes
Did all patients receive the same reference standard?	Yes	Yes	Yes	Yes	Yes
Were all patients included in the analysis?	Yes	Yes	Yes	Yes	Yes

Synthesis of results

Results of the selected studies included accuracy (Supplementary Table A), intraexaminer (Supplementary Table B) and interexaminer reproducibility (Supplementary Table C) of linear measurements on digital models when compared with plaster models. Included studies selected the following linear measurements: mesiodistal tooth width, intercanine width, interpremolar width, intermolar width, sum of anterior teeth, sum of all teeth, required space, available space/arch perimeter, arch crowding, arch length and arch length discrepancy.

Accuracy

The measurements selected to evaluate accuracy were organized into four groups: arch length and crowding, tooth width, arch width and Bolton discrepancy.

Arch length and crowding

Four studies assessed the accuracy of measurements associated to space analysis and crowding.^1,8,25,26 The differences between measurements of crowding ranged from 1.06^8,26 to 1.75 mm.⁸ Luu et al⁸ showed a high discrepancy (3.38 mm) for the maxillary perimeter measurement. Two authors^1,8 used intraclass correlation coefficient (ICC) when they compared arch length obtained from plaster models and digital models and found a correlation that ranged from poor to excellent between the methods. The poor correlation was identified for maxillary perimeter measurement⁸ (Supplementary Table A).

In the meta-analysis, the included studies trended towards overestimating crowding measurement on digital models compared with plaster models in the maxillary arch. However, in the mandibular arch, Lim and Lim²⁵ underestimated this measurement on digital models. There were no significant differences in the subgroups and when both arches were considered together (Figure 4). In this comparison, the heterogeneity in both subgroups was high (I²: maxillary arch, 78.30%; mandibular arch, 76.73%). A sensitivity analysis was performed and the heterogeneity decreased (I²  = 0% in both subgroups) when the articles by Luu et al⁸ and Lv et al²⁶ were removed from the mandibular and the maxillary arch, respectively (Figure 5). This may have happened because of the high number of examiners⁸ or because the lower arch was not included in the comparisons.²⁶ However, although the effect size changed, the conclusion that no significant difference was found in the subgroups or in the overall analysis remained.

Figure 4.

Comparison of crowding measurement obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Figure 5.

Comparison of crowding measurement obtained on digital model generated from CBCT and plaster model when performing sensitivity analysis. CI, confidence interval.

Tooth width

Differences in tooth width were measured in three studies.^8,25,26 The mean difference ranged from 0.01 mm, measured in the lower lateral incisor,²⁵ to 0.47 mm, in the upper central incisor.⁸ Most tooth width measurements in digital models were lower than in plaster models, according to the results reported by Lim and Lim²⁵ and Lv et al,²⁶ and were higher in the results of Luu et al.⁸ The ICC calculated by Luu et al verified a correlation that ranged from poor to good in the majority of measurements. Tooth width was evaluated by Grunheid et al⁹ through Bland–Altman analysis, and they found a bias of 0.191 mm with limit of agreement interval width of 1.521 and a mean squared error of 0.181 mm (Supplementary Table A).

In the meta-analysis, the data from the study of Lim and Lim²⁵ was duplicated (right and left sides) since the authors presented only one value for each tooth in the maxillary and mandibular arches. The teeth measurements were assessed with the following subgroups: incisors, canines, premolars and molars (Figures 6–9). When the heterogeneity was high, a sensitivity analysis was performed. When the article by Luu et al⁸ was removed from the canine and premolar comparisons in the maxilla subgroup, the heterogeneity decreased (Figures 10 and 11) as measured by the I² index (from 83.10% to 23.02% in the canines comparison; from 84.03% to 60.54% in the premolars comparison). In the molars comparison, no article removal in the performance of the sensitivity analysis made a difference in the high heterogeneity. The width of the incisors was overestimated in the mandibular arch and the width of canines and premolars was underestimated in the maxillary arch, with significant differences after sensitivity analysis.

Figure 6.

Comparison of mesiodistal tooth width (incisors) obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Figure 9.

Comparison of mesiodistal tooth width (molars) obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Figure 10.

Comparison of mesiodistal tooth width (canines) obtained on digital model generated from CBCT and plaster model when performing sensitivity analysis. CI, confidence interval.

Figure 11.

Comparison of mesiodistal tooth width (premolars) obtained on digital model generated from CBCT and plaster model when performing sensitivity analysis. CI, confidence interval.

Figure 7.

Comparison of mesiodistal tooth width (canines) obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Figure 8.

Comparison of mesiodistal tooth width (premolars) obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Arch width

Transverse dimensions were measured in three studies.^8,25,26 The mean differences ranged from 0.01²⁵ to 0.39 mm,⁸ both measured at mandibular first molar.^8,25 Statistically significant differences were found on mandibular intermolar width,⁸ and maxillary intercanine width in individual studies.²⁵ Luu et al identified an excellent correlation by ICC (Supplementary Table A). In the meta-analysis, only the intermolar width comparison was possible, and no significant difference was observed (Figure 12).

Figure 12.

Comparison of intermolar width obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Bolton analysis discrepancy

Regarding Bolton discrepancy, differences were compared in two studies.^8,25 The mean differences ranged from −0.35 to −1.95 mm, measured for anterior and overall Bolton discrepancies, respectively, measured for anterior²⁵ and overall Bolton discrepancies.⁸ A statistically significant difference was observed for anterior⁸ and overall Bolton discrepancies in individual studies.^8,25 A poor correlation was verified by Luu et al when the ICC was calculated. Lim and Lim²⁵ applied Pearson correlation and verified a correlation of 0.764 and 0.718, respectively, for anterior Bolton and overall Bolton discrepancies (Supplementary Table A). The meta-analysis was performed between two studies,^8,25 and there was a significant difference in the Bolton analysis (Figure 13) to overall Bolton discrepancy and no significant difference to anterior Bolton discrepancy.

Figure 13.

Comparison of Bolton analysis obtained on digital model generated from CBCT and plaster model. CI, confidence interval.

Reproducibility

The results of reproducibility were presented subdivided as intra- and interexaminer reproducibilities. The studies assessed this feature by using different tests: ICC,^1,8,26 the Bland–Altman method⁹ and paired t-tests²⁵ and ANOVA⁸ for mean differences. The ICC showed that the reproducibility may be considered good when >0.75 and as excellent when >0.9.⁵⁴ The limits of agreement of the Bland–Altman method and the mean differences reported were compared with the clinically significant values considered in this review. The paired t-test results were considered significant when p < 0.05.

Intraexaminer reproducibility

All included studies assessed intraexaminer reproducibility of linear measurements on digital models.^1,8,9,25,26 The ICC parameters were verified in three studies,^1,8,26 all of them >0.75, demonstrating a good reproducibility. The limits of agreement of tooth width reported by Grunheid et al⁹ were in absolute values as high as 0.5 mm, and 95% of the mean differences were below this value, which was considered clinically insignificant according to the parameters adopted in this review. Lim and Lim²⁵ showed statistically significant mean differences in the repeated measures of mesiodistal width of the lower lateral incisor and lower second premolar when the paired t-test was applied, whereas Luu et al⁸ showed statistically significant mean differences in the repeated measures of the upper first molar, in the maxillary intermolar width, arch perimeter and arch crowding. However, the absolute mean differences for mesiodistal tooth width of the significant findings were <0.28 mm and for the sum of measurements were <1.71 mm, which were considered clinically insignificant (Supplementary Table B).

Interexaminer reproducibility

The interexaminer reproducibility was reported in three studies.^1,8,9 The ICC parameters verified by Akyalcin et al¹ and Luu et al⁸ were, respectively, >0.92 and >0.559 for all measurements and >0.75 for 67% of the measurements. The limits of agreement of tooth width reported by Grunheid et al⁹ were in absolute values as high as 2.3 mm, which was considered clinically significant. Luu et al⁸ showed statistically significant mean differences in the repeated measures of the upper canine and second premolar, lower lateral and central incisors, maxillary intermolar width, mandibular intercanine width and maxillary and mandibular perimeter and crowding. The absolute mean differences for mesiodistal tooth width of the significant findings were <0.34 mm, which were considered clinically insignificant. However, for sum of measurement values were as high as 5.72 mm, which were considered clinically significant (Supplementary Table C).

Discussion

Modifying what is traditional, reliable and valid requires the use of arguments obtained through numerous, extensive and qualified research.

Dental models have been reported to be an essential record of patient's occlusion and extremely used in the decision of treatment planning.¹³ Although plaster models are currently considered the gold or reference standard for orthodontic diagnostics,^{1,8,9,13,25,26,49,55} measurements in plaster models may involve errors,^8,9 and there are a few known reports.^{1,14,18,31,32} Some random and systematic errors in measurements in these models may be a result of the measuring instrument, difficulties in the measurement process, operator skill and/or by alginate expansion because of imbibition of water.⁸

Nowadays, there is a tendency to incorporate digital models in orthodontic practice, but their potential advantages^14,18,31,32 would be rejected if the accuracy and reproducibility of their analysis methods were not comparable with those taken on plaster models with digital calipers.⁹ Digital models generated from the patient's mouth impressions might be associated with some errors such as alginate shrinkage during transportation to companies that generate digital models,⁷ and/or loss of information because of the scanning process.⁹ However, previous systematic reviews have concluded that digital models are accurate and are reproducibly and clinically acceptable for dental measurements.^14,15,31,32

Although almost all digital models are made from alginate impressions, which are directly scanned or poured in plaster to obtain the study models and then scanned,⁹ the digital models could also be obtained from the CBCT scan of the patients' heads when they have already been submitted to the examination for other clinical reasons. The use of dental models generated from the CBCT scan would bring the advantage of sparing the impression procedure, added to the other benefits of digital models. Nevertheless, digital models generated from CBCT scans exhibit limited anatomical reproduction of the occlusal and interproximal tooth parts.^1,9

When analyzing the methodological quality of the selected studies, we found gaps in few signalling questions, according to the QUADAS-2 tool. Most studies have not clearly described blinding of the investigators when the results of the tests were interpreted, and the samples randomized.^1,25,26 No concerns about applicability of the methods used in the studies were identified, and all of them matched the review question. The lack of information about an appropriate interval between the acquisition of plaster and digital models could cause bias as a result of oral diseases or trauma which could change the morphology of the teeth that would be submitted for measurements.²⁶ In the quantitative analysis performed through the meta-analysis, all previous selected studies were included because of the small number of studies identified in the screening, although one²⁶ of them presented an unclear risk of bias.

Only one article⁹ used Bland–Altman analysis to evaluate accuracy and reproducibility, respectively, by the degree of agreement between the measurements on digital models and those on plaster models, and between replicate measurements. Future studies could include this method that is simple and easy to demonstrate agreement and reliability.

Regarding previous systematic reviews, only Rossini et al¹⁴ selected an article¹ that was in common with our systematic review. This review¹⁴ was the latest published in the literature and conducted the search strategy until November 2014. The difference between the results of the included articles is probably due to different review questions, a search strategy that provided a large literature scan and the differences in our selection criteria. We chose to focus on quantitative linear measurements on digital models generated from direct patient CBCT scans and compared solely with plaster models.

The accuracy of dental linear measurements obtained from digital models generated from the CBCT scan compared with those obtained from the plaster models showed clinically insignificant mean differences for mesiodistal tooth width, and transversal widths (−0.036 to 0.204) and for sums of measurements (−0.097 to −1.654), according to the values assigned in our study, although statistically significant differences were found in some subgroups analyzed. This agrees with the findings of previous reviews.^14,15

Only maxillary perimeter was judged to be clinically significant when compared between digital and plaster models according to Luu et al.⁸ However, only one article assessed this parameter and more researches are necessary to confirm these results. The ICC for this measurement was poor when evaluated in the included studies.^1,8

In orthodontic treatment, the assessment of crowding and intermolar width are frequently used in dental models. The heterogeneity was null (I² = 0%, Figure 5) and moderate (I² = 57% Figure 12) in these groups, and the values trended towards overestimating the measures in the maxillary arch while underestimating them in the mandibular arch.

When plaster models and digital models generated from the CBCT scan were compared in the meta-analysis, the values were overestimated in the mandibular arch while underestimated for mesiodistal width measurements of teeth in the maxillary arch (Figures 6 and 9–11). That may have happened because of the difference in the size of the maxillary and mandibular teeth, the low accuracy of interproximal surfaces, random errors on landmark positions and the methodological acquisition of the segmentation process.

Our findings, summarizing the evidence of intraexaminer reproducibility of the linear dental measurements acquired from CBCT digital models of the included articles agree with the results of the reviews of Luu et al⁸ and Rossini et al.¹⁴ However, regarding interexaminer reproducibility, differently from Rossini et al,¹⁴ we identified problems, which shows that the methods may need further improvement, especially operator training should be emphasized and images could also be enhanced.⁴⁹

Most mean differences obtained from the comparison of measurements performed on the plaster models and digital models generated from CBCT scan suggest that one should contraindicate these digital models for study model analysis mainly on borderline cases. Other known reasons to avoid this method were in dental arches with restorations cases,⁵¹ and for patients with primary and mixed dentitions because they are at increased risk of exposure to ionizing radiation.⁵⁶

Precision of 0.01 mm is adopted to take measurements in plaster models and digital models software in general,^{1,7,9,12,13,26} therefore these digital models generated from CBCT are in fact limited by the resolution of CBCT scan.^8,9 These standardizations already affect the quality of the images, especially digital models generated from CBCT scan. All selected studies for this review used 0.3 × 0.3 × 0.3-mm voxel size.^1,8,9,25,26 Added to this, the image quality might be affected by scanning artefacts, patient's movement while performing CBCT scan^9,25 and by the poor reconstruction of occlusal surface in some cases,⁹ and these problems might affect dental measurements. When poor image quality of occlusal surface is obtained, the arch width measurement would probably be questionable.

One limitation of this review is that the findings were extracted from only five articles.^1,8,9,25,26 We suggest further research on this subject so that more information is added to the scientific community and can strengthen the evidence available. Another potential weakness is that the files of digital models obtained from CBCTs are not in the same format or are not specified by authors.

As demonstrated in our systematic review with meta-analysis, the digital models generated from CBCT scan are not perfect yet. We do not intend to suggest the replacement of plaster models or some digital models, that have proven to be accurate and reproducible, by models generated from CBCT scans, but we advise to take the opportunity of using these models when already part of the diagnostic record of the patient. And in the future we may probably use a unique register of the patient with CBCT and take several analysis of this record.

Conclusions

Based on the moderate level of evidence, it may be concluded that:

• – Digital models obtained from CBCT are accurate for the following maxillary and mandibular dental measures: crowding, intermolar width and mesiodistal width of maxillary incisors, mandibular canines and premolars, and molars in both arches;

• – The differences were clinically acceptable for all dental linear measurements, except for arch perimeter

• – Digital models generated from CBCT are reproducible for all measurements when intraexaminer assessment is considered and need improvement in interexaminer evaluation.

Appendix A

Table A1.

QUADAS-2 adapted

Risk of bias	Concerns regarding applicability
Patient selection	Patient selection
Was a consecutive or random sample of patients enrolled?	Are there concerns that the included patients and settings do not match the review question?
Was a case–control design avoided?
Did the study avoid inappropriate exclusions?
Index test	Index test
Were the index test results interpreted without knowledge of the results of the reference standard?	Are there concerns that the index test, its conduct, or interpretation differ from the review question?
Was the field of view described?
Was the CBCT scanner described?
Was the voxel size described?
Was the type of software for models acquisition described?
Was the type of software for models measurements described?
Reference standard	Reference standard
Is the reference standard likely to correctly classify the target condition?	Are there concerns that the target condition as defined by the reference standard does not match the question?
Were the reference standard results interpreted without knowledge of the results of the index tests?
Flow and timing
Was there an appropriate interval between index test and reference standard?
Did all patients receive the same reference standard?
Were all patients included in the analysis?