ORIGINAL ARTICLE
Comparison of 3-dimensional dental models from different sources: Diagnostic accuracy and surface registration analysis Sercan Akyalcin,a David J. Dyer,b Jeryl D. English,c and Cagla Sard Houston, Tex, Okinawa, Japan, and Ankara, Turkey
Introduction: The aim of this study was to assess the diagnostic accuracy and surface matching characteristics of 3-dimensional digital dental models obtained from various sources. Methods: Three sets of maxillary and mandibular digital models of 30 subjects were included in this study. Three-dimensional stereolithography model files were obtained from a 3-dimensional laser desktop scanner (Ortho-Insight 3D; Motionview Software, Hixson, Tenn), the emodel system (GeoDigm, Chanhassen, Minn), and cone-beam computerized tomography. Arch-length discrepancy measurements were made on the 3-dimensional digital models and compared with direct caliper measurements. Additionally, stereolithography files from the 3 digital model systems were paired and superimposed using a best-fit algorithm. Average linear differences between the stereolithography shells were computed together with surface correlation amounts at various tolerance levels. Data were evaluated using intraclass correlation coefficients and the Tukey mean difference test. Results: Although all 3 digital model groups displayed good correlation with caliper measurements, the virtual scan models had the highest correlation with the manual method (ICC . 0.95). The Tukey mean difference test showed no consistent bias of one approach vs the others compared with caliper measurements; random errors were detected in all the comparisons. For the estimation of arch-length discrepancy, the mean bias of the scanned virtual models in comparison with caliper measurements (0.24 6 0.67 mm) was smaller than the mean biases of the emodels and the models generated from cone-beam computed tomography. Additionally, the best surface overlap correlation was observed between the virtual scanned models and the emodels. The mean linear distances between the stereolithography shells of these 2 model systems were 0.14 and 0.13 mm for the maxillary and mandibular arches, respectively. Conclusions: All 3 digital model systems can provide diagnostic information similar to caliper measurements, with varying degrees of agreement limits. The scanned virtual models had the least mean bias. A strong surface match correlation was observed between the virtual scanned models and the emodels, indicating that these could be used interchangeably. (Am J Orthod Dentofacial Orthop 2013;144:831-7)
P
laster models have been the gold standard in dental diagnosis and treatment procedures for many years. However, they require rigorous archiving and massive physical storage space.1,2 a Assistant professor, Department of Orthodontics, School of Dentistry, The University of Texas Health Science Center at Houston, Houston, Tex. b Orthodontist, United States Navy, Okinawa, Japan. c Chair and program director, Department of Orthodontics, School of Dentistry, The University of Texas Health Science Center at Houston, Houston, Tex. d Assistant professor, Department of Orthodontics, School of Dentistry, Baskent University, Ankara, Turkey. All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest, and none were reported. Address correspondence to: Sercan Akyalcin, 7500 Cambridge St, Suite 5130, Houston, TX 77054; e-mail, [email protected]. Submitted, June 2013; revised and accepted, August 2013. 0889-5406/$36.00 Copyright Ó 2013 by the American Association of Orthodontists. http://dx.doi.org/10.1016/j.ajodo.2013.08.014
Moreover, plaster models are not practical in the long term because of breakage and degradation issues associated with them.2 Accordingly, digital models made a rapid introduction into clinical practice in the late 1990s, and their clinical use has escalated since then.3 There have been numerous efforts to investigate the diagnostic accuracy and measurement sensitivity of digital models compared with plaster models.4-11 It was reported that information obtained from digital models is interchangeable with direct caliper measurements made on conventional plaster models. According to these reports, 2-point linear measurements such as intermolar and intercanine width,6,10 tooth size,9 tooth height,10 overjet, and overbite5,6,9 always show near-excellent agreement between conventional models and the analog alternatives. However, some other 831
Akyalcin et al
832
Fig 1. Digital model systems evaluated in this study: A, scanned virtual models; B, emodels; C, CBCTgenerated models.
studies have identified slightly high discrepancies between digital and stone models for comparisons of more complex diagnostic measurements such as the Bolton ratio, space analysis, and the irregularity ratio.6,8,12 Different techniques used in obtaining digital models inevitably cause standardization issues and might account for differences between the conventional plaster and digital models. Not only do the digital models differ from each other regarding their construction phases,3 but they also require the use of various proprietary softwares.2 Ideally, it should be possible to view, measure, and store 3-dimensional (3D) digital models universally regardless of the technique-specific details. The stereolithography (STL) file format is an industry standard for computer-aided design and manufacturing and allows the 3D files to be viewed and manipulated universally by many software packages. The aims of this study were to (1) evaluate the diagnostic accuracy of 3D STL files of the same dental anatomy obtained from different digitizing procedures and to compare them with direct caliper measurements using the same software, and (2) report the surface correlations of the 3D STL files of these model systems. MATERIAL AND METHODS
Approval for the study was obtained from the institutional review board of the University of Texas Health Science Center at Houston. The sample comprised the records of 30 orthodontic patients between the ages of 14 and 30 years. Each subject had maxillary and mandibular dental impressions made with Identic (Dux Dental, Oxnard, Calif) alginate along with bite registrations taken with polyvinyl siloxane material (Blue Moose; Parkell, Edgewood, NY). The impressions were immediately poured with Snow White Plaster (Kerr, Orange, Calif). The patients recruited for this study also
December 2013 Vol 144 Issue 6
had medium field-of-view (15 3 15 cm) cone-beam computed tomography (CBCT) scans in their files for various diagnostic purposes. All CBCT scans were made on the Galileos Comfort x-ray unit (Sirona Dental Systems, Bensheim, Germany) with exposure parameters of 85 kVp, 21 mA, 14 seconds, and 0.3-mm voxel size. CBCT scans with artifacts caused by patient movement and metal restorations were not included in the sample. Three sets of maxillary and mandibular digital models were obtained via different protocols. The first group consisted of virtual dental models obtained with a 3D desktop scanner (Fig 1, A). The Ortho Insight 3D laser scanner (Motionview Software, Hixson, Tenn) was used to directly scan the plaster models. The images were processed with Motionview software, which generated files with the STL extension for each dental model. The second group (Fig 1, B) included digital models made with the emodel system (GeoDigm, Chanhassen, Minn). For this study, the models were scanned, digitized, and converted to STL files by the company. Once finalized, the files were downloaded from the company's servers. The third group consisted of CBCT-generated digital dental models (Fig 1, C). CBCT scans of the study sample were uploaded to the company's servers (Anatomage, San Jose, Calif). The 3D images were segmented, digitally reformatted, and converted to STL files by the company. The digital models were then ready to download. All STL files in the 3 groups were saved in the same personal computer (Core i7-2600 CPU @3.40 Ghz with 8 MB of memory; Intel, Fort Worth, Tex). The plaster models were measured manually with a digital caliper. These measurements set the gold standard in this study. The same software was used for manipulation and measurements of the 3D STL files. The files were opened in random order with the Rapidform software program (Inus Technology, Seoul, South Korea). Mesiodistal tooth widths and dental arch
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
833
Fig 2. Surface registration analysis based on a best-fit algorithm displaying the deviations between the 2 model shells.
Table I. Mean maxillary and mandibular arch-length discrepancy values in the groups and ICCs for the comparisons
between the caliper measurements and the digital model systems Caliper measurements
Maxillary crowding (mm) Mandibular crowding (mm)
Mean 4.9 4.7
SD 3.8 3.1
Emodels Mean 4.4 4.3
SD 3.3 3.0
segments were measured on both the plaster models and the STL files to the nearest 0.01 mm. Maxillary and mandibular arch-length discrepancies were calculated based on these measurements. STL files from all 3 groups of digital models were also paired and digitally registered to create 3 groups that reflected the differences in the surface areas of these models. Surface registration of the model pairs was accomplished using the best-fit algorithm of the Rapidform software (Fig 2). After the registration protocol, average linear differences between the 2 STL files were recorded using the measurement tool of the same software. Additionally, surface overlap correlations between the file pairs were evaluated at 3 tolerance levels: 0.25, 0.5, and 1 mm. Statistical analysis
Two examiners (S.A. and C.S.), working independently, repeated all measurements twice 3 to 4 weeks later. The intraobserver repeatability and reproducibility of these observers were evaluated with intraclass correlation coefficients (ICCs). One-way analysis of variance (ANOVA) was used to compare the mean values of the measurements between the 4 groups. Agreement between the digital platforms compared with the caliper measurements was evaluated with both ICCs and the Tukey mean difference test (Bland-Altman analysis).
Virtual scanned models ICC 0.91 0.89
Mean 4.7 4.5
SD 3.3 3.0
ICC 0.95 0.96
CBCT-generated models Mean 4.5 4.2
SD 3.0 3.2
ICC 0.85 0.85
Statistical analyses were carried out with both SPSS statistics (version 20; IBM, Armonk, NY) and XLSTAT Mac (version 2012; Addinsoft, New York, NY). RESULTS
Both the intraobserver repeatability and the reproducibility of the observers showed near perfect agreement (ICCs . 0.92). According to 1-way ANOVA, no significant differences were found between the groups for the arch-length discrepancy measurements (Table I). The ICCs of the digital platforms compared with the caliper measurements are listed in Table I. Although all 3 digital model groups had good correlations with the caliper measurements, the virtual scan models had the highest correlation with the manual method (ICCs . 0.95). The findings from the Tukey mean difference test supported this finding and are summarized in Table II. Tukey mean difference plots can be seen in Figure 3. This analysis showed no consistent bias of 1 approach vs the others compared with the caliper measurements, and random errors were detected in all comparisons. For the estimations of arch-length discrepancies in both jaws, the mean bias (0.24 6 0.67 mm) of the scanned virtual models compared with the caliper measurements was smaller than the mean biases for the emodels and the CBCT-generated models. All 3
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
Akyalcin et al
834
Table II. Summary of the Tukey mean difference test
for the estimation of arch-length discrepancies in both arches Caliper measurements compared with
Bias SD 95% CI
Emodels 0.49 0.81 2.11 to 1.23
Scanned virtual models 0.24 0.67 1.62 to 1.1
CBCT-generated models 0.45 1.10 2.65 to 1.74
digital platforms provided similar measurements to the caliper measurements. However, from a clinical perspective, the measurements on the scanned virtual models provided better agreement with the manual caliper measurements, as shown by the smaller confidence interval (agreement limit) range (Table II). Table III presents the summary of the surface registration analysis of the matched pairs of maxillary and mandibular digital models. The best surface overlap correlation was observed between the virtual scanned models and the emodels. The mean linear distances between the STL shells of these 2 model systems were 0.14 and 0.13 mm for the maxillary and mandibular arches, respectively. However, when the virtual scanned models or the emodels were paired with CBCT-generated models, the mean surface difference increased by almost 0.2 mm in both cases. A 95% surface overlap correlation was observed between the virtual scanned models and the emodels at a tolerance level of 0.5 mm (Table IV). This means that if a 0.5-mm average linear distance between the 2 STL files could be tolerated, 95% of the surface area would coincide. However, the same amount of surface overlap correlation (95%) did not exist in the paired groups, including the CBCT-generated models, until the tolerance level was doubled. At the 1-mm tolerance level, the virtual scanned models and the emodels displayed 99% correlations. DISCUSSION
Previous studies have investigated the measurement accuracy of digital model systems with their brandspecific softwares.4-12 This could lead to conflicting results because of software differences and potential operator calibration issues with the use of multiple programs. Digital dental models that are basically 3D files should ideally be manipulated and analyzed with high precision, independent of the software interface. Therefore, in our study, it was intended to compare the 3D STL shells of the same dental anatomy from various digitizing sources using the same software program. However, landmark identification in any 3D
December 2013 Vol 144 Issue 6
software will likely differ from actual physical measurements because there is no caliper landing on physical points during the process. Additionally, no measurement of a physical object is exact. Some degree of error is associated with the measured data in all experiments. This might explain some of the statistical variability between the mean values of caliper and digital measurements. However, because both the repeatability and reproducibility of the results were confirmed, the diagnostic accuracy of the digital model systems was established using both correlations and agreement limits. According to our results, all 3 digital systems investigated here showed good correlations with manual caliper measurements when diagnosing the archlength discrepancy using ICCs. Although ICCs are a good retest measure,13 determining method reliability requires a comprehensive set of statistical methods.14,15 This is why the Tukey mean difference test was also applied to the data. In essence, the digital platforms showed a similar pattern of systematic errors that can be observed on the Tukey mean difference plots (Fig 3). However, potential differences between the construction phases of the digital models led to varying degrees of bias and agreement limits as they were compared with the manual measurements. As can be evidenced by smaller mean bias and agreement limits, the scanned virtual models slightly outperformed the emodels and the CBCT-generated models. Studies comparing emodels with conventional plaster models have suggested that digital models offer interchangeable information with plaster models when used for orthodontic diagnostic purposes.3,7,8 In our study, the emodels had less bias than did the CBCTgenerated models. Recent studies have demonstrated that linear measurements from CBCT-generated digital models have similar levels of accuracy when compared with other digital model systems.16-19 However, a high level of accuracy as well as high-quality image detail are required before CBCT-generated models can generally be accepted. As demonstrated by our findings, CBCT-generated models are not perfect yet. In our study, they had the largest mean bias and confidence interval. In some instances, almost 3-mm differences were observed for the estimation of dental crowding between the measurements on the CBCT-generated models and the caliper measurements. This much difference might have clinical implications, especially in severely crowded dentitions.20 The second part of our study, evaluating the correlation between the surface areas of the STL files obtained from the digital sources, yielded a parallel outcome to the measurement accuracy findings. In a
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
835
Fig 3. Tukey mean difference plots for comparison of digital model systems with direct caliper measurements. Table III. Mean linear distances between the surface
areas of the maxillary and mandibular digital model pairs
Maxillary model pairs Scanned virtual models and emodels Scanned virtual models and CBCT- generated models CBCT-generated models and emodels Mandibular model pairs Scanned virtual models and emodels Scanned virtual models and CBCT-generated models CBCT-generated models and emodels
n
Mean (mm)
SD
30 30
0.14 0.32
0.04 0.06
30
0.34
0.07
30 30
0.13 0.31
0.04 0.06
30
0.32
0.07
recent report, Park et al21 demonstrated that surface registration was a simpler, more reproducible and reliable method than superimposition based on anatomic structures. Therefore, we used a regional registration with surface-to-surface matching (best-fit method) in our study. According to our results, a 95% surface overlap correlation did not exist in the comparisons of the CBCT-generated models with the other 2 digital model systems until 1.00 mm of difference between the surface areas of the 2 shells was tolerated. However, the surface overlap correlation between the virtual scanned models and the emodels was more than 95% at a 0.5mm tolerance for both the maxillary and mandibular model sets. This finding suggests that the surface areas of the CBCT-generated models could lack the finer details such as occlusal pits and fissures. The lack of detail in this group can even be observed with the naked eye when examining the 3 digital model systems closely (Fig 1). This is partly due to the segmentation process of maxillary and mandibular teeth during the construction phase of the CBCT-generated models. Digital models synthesized from CBCT images are constructed from scans with the patient's teeth in occlusion. This inevitably leads to overlapping of the images and
makes the segmentation process arduous and timeconsuming.22,23 Second, spatial resolution might play an important role on the accuracy of 3D tooth reconstructions. In a study that compared 3D tooth reconstructions using CBCT and microcomputed tomography data, the most important differences occurred mainly where the tooth was thinner, either near the cervix of the still-developing enamel cap or at the incisal edges of the anterior teeth.24 This finding points out the importance of isotropic voxel size when capturing finer details on physical models. It was also reported that field-of-view selection has a significant influence on the quality of the 3D surface models of the dental arches from 1 CBCT system.22,25 Al-Rawi et al22 evaluated the accuracy of 3D CBCT reconstructions of the tooth crowns and occlusal surfaces with 3 field-of-view sizes and concluded that the choice of a large field of view reduced the visibility of the occlusal surfaces compared with small and medium fields of view. This is due to the decrease in voxel size in scans with larger fields of view. Recently, Lightheart et al18 compared CBCTgenerated digital models with OrthoCAD models to evaluate surface overlap correlation between them. They demonstrated fair agreement between the 2 digital model systems, where only 90% matching occurred at the 1-mm to 1.25-mm tolerance level range. Accordingly, they signified the importance of gingival tissues, since digital models obtained by optical laser scanning differ from CBCT-generated-models in these areas. It is clearly impossible to capture the gingiva with CBCT technology. Therefore, they suggested that tasks related to gingival surface areas—eg, design of a vacuumformed appliance—should not be performed on CBCTgenerated models. Our results are in agreement with these findings. Although the types of files were standardized and the linear measurements were all done on STL files in our study, accuracy of impression-based digital
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
Akyalcin et al
836
Table IV. Surface overlap (%) for the maxillary and mandibular digital model pairs at various tolerance levels Scanned virtual models and emodels Maxillary Tolerance level (mm) 0.25 0.50 1.00
Mean 84 95 99
SD 8 2 0
Mandibular Mean 86 96 99
SD 9 4 1
Scanned virtual models and CBCT-generated models Maxillary Mean 51 78 95
models might have been limited by the impression or the stone material. According to Torassian et al,26 the alginate material we used in our project showed a statistically and clinically significant change in all dimensions within 72 hours. In our study, all impressions were poured immediately. Therefore, distortion during the construction phase of the plaster models, if any, was kept minimal. However, it could have been a better approach to use alginate substitutes that are dimensionally accurate over an extended period of time. Perhaps eliminating the impression step completely and acquiring digital models with intraoral scanners will soon be the gold standard to retrieve 3D models of dental arches. However, intraoral scanners need to be developed to overcome their initial shortcomings such as restricted scanning time and the difficulties in scanning contact points and the vertical dimensions of gingivae in the dental arch.27 Based on the limitations of our study, optical laser scanning of the impression or the plaster model is the best method for digitizing patient records. From a radiation-protection point of view, a CBCT scan of the patient must never be the primary source for constructing digital models.28,29 In other words, use of ionizing radiation to generate digital models as a substitute for other digitizing sources is not acceptable. In our study, we used patients who already had their CBCT scans taken because of conclusive evidence suggesting a diagnostic need: eg, surgical intervention. This approach made it possible for us to evaluate the precision of the CBCT-generated models along with other digitizing sources. However, CBCT-generated models are not recommended for routine clinical use. Under justified circumstances, their generation will require a good-quality scan with no movement artifacts and with adequate spatial resolution. Based on our findings and what we know from the previous studies, a voxel size \0.3 mm should give adequate diagnostic information.15,24 However, CBCT-generated models with smaller voxel sizes should be further evaluated for their surface details and precision.
December 2013 Vol 144 Issue 6
Mandibular SD 8 8 2
Mean 52 80 96
SD 9 8 2
CBCT-generated models and emodels Maxillary Mean 48 77 95
SD 9 8 3
Mandibular Mean 52 80 96
SD 10 9 3
CONCLUSIONS
Arch-length discrepancy measurements made on digital model files obtained from different sources including emodels, a desktop scanner (virtual scanned models), and CBCT-generated models had similar patterns of random error when compared with direct caliper measurements. However, scanned virtual models provided better agreement with plasters casts. The best surface overlap correlation was observed between the virtual scanned models and the emodels, indicating a strong match between the STL shells of these 2 systems. REFERENCES 1. Peluso MJ, Josell SD, Levine SW, Lorei BJ. Digital models: an introduction. Semin Orthod 2004;10:226-38. 2. Akyalcin S. Are digital models replacing plaster casts? Dentistry 2011;1:e102. 3. Fleming PS, Marinho V, Johal A. Orthodontic measurements on digital study models compared with plaster models: a systematic review. Orthod Craniofac Res 2011;14:1-16. 4. Goonewardene RW, Goonewardene MS, Razza JM, Murray K. Accuracy and validity of space analysis and irregularity index measurements using digital models. Aust Orthod J 2008;24:83-90. 5. Leifert MF, Leifert MM, Efstratiadis SS, Cangialosi TJ. Comparison of space analysis evaluations with digital models and plaster dental casts. Am J Orthod Dentofacial Orthop 2009;136: 16.e1-4. 6. Quimby ML, Vig KW, Rashid RG, Firestone AR. The accuracy and reliability of measurements made on computer-based digital models. Angle Orthod 2004;74:298-303. 7. Mullen SR, Martin CA, Ngan P, Gladwin M. Accuracy of space analysis with emodels and plaster models. Am J Orthod Dentofacial Orthop 2007;132:346-52. 8. Stevens DR, Flores-Mir C, Nebbe B, Raboud DW, Heo G, Major PW. Validity, reliability, and reproducibility of plaster vs digital study models: comparison of peer assessment rating and Bolton analysis and their constituent measurements. Am J Orthod Dentofacial Orthop 2006;129:794-803. 9. Horton HM, Miller JR, Gaillard PR, Larson BE. Technique comparison for efficient orthodontic tooth measurements using digital models. Angle Orthod 2010;80:254-61. 10. Keating AP, Knox J, Bibb R, Zhurov AI. A comparison of plaster, digital and reconstructed study model accuracy. J Orthod 2008; 35:191-201. 11. Santoro M, Galkin S, Teredesai M, Nicolay O, Cangialosi TJ. Comparison of measurements made on digital and plaster models. Am J Orthod Dentofacial Orthop 2003;124:101-5.
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
12. Tomassetti JJ, Taloumis LJ, Denny JM, Fischer JR Jr. A comparison of 3 computerized Bolton tooth-size analyses with a commonly used method. Angle Orthod 2001;71:351-7. 13. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86:420-8. 14. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8:135-60. 15. Rankin G, Stokes M. Reliability of assessment tools in rehabilitation: an illustration of appropriate statistical analyses. Clin Rehabil 1998;12:187-99. 16. Kau CH, Littlefield J, Rainy N, Nguyen JT, Creed B. Evaluation of CBCT digital models and traditional models using the Little's index. Angle Orthod 2010;80:435-9. 17. Creed B, Kau CH, English JD, Xia JJ, Lee RP. A comparison of the accuracy of linear measurements obtained from cone beam computerized tomography images and digital models. Semin Orthod 2011;17:49-56. 18. Lightheart KG, English JD, Kau CH, Akyalcin S, Bussa HI, McGrory KR, et al. Surface analysis of study models generated from OrthoCAD and cone-beam computed tomography imaging. Am J Orthod Dentofacial Orthop 2012;141:686-93. 19. Tarazona B, Llamas JM, Cibrian R, Gandia JL, Paredes V. A comparison between dental measurements taken from CBCT models and those taken from a digital method. Eur J Orthod 2013;35:1-6. 20. Redich M, Weinstock T, Abed Y, Schneor S, Holdstein Y, Fischer A. A new system for scanning, measuring and analyzing dental casts based on a 3D holographic sensor. Orthod Craniofac Res 2008; 11:90-5. 21. Park TJ, Lee SH, Lee KS. A method for mandibular dental arch superimposition using 3D cone beam CT and orthodontic 3D digital model. Korean J Orthod 2012;42:169-81.
837
22. Al-Rawi B, Hassan B, Vandenberge B, Jacobs R. Accuracy assessment of three-dimensional surface reconstructions of teeth from cone beam computed tomography scans. J Oral Rehabil 2010; 37:352-8. 23. Fourie Z, Damstra J, Schepers RH, Gerrits PO, Ren Y. Segmentation process significantly influences the accuracy of 3D surface models derived from cone beam computed tomography. Eur J Radiol 2012;81:e524-30. 24. Maret D, Molinier F, Braga J, Peters OA, Telmon N, Treil J, et al. Accuracy of 3D reconstructions based on cone beam computed tomography. J Dent Res 2010;89:1465-9. 25. Hassan B, Couto Souza P, Jacobs R, de Azambuja Berti S, van der Stelt P. Influence of scanning and reconstruction parameters on quality of three-dimensional surface models of the dental arches from cone beam computed tomography. Clin Oral Investig 2010; 14:303-10. 26. Torassian G, Kau CH, English JD, Powers J, Bussa HI, Marie SalasLopez A, et al. Digital models vs plaster models using alginate and alginate substitute materials. Angle Orthod 2010;80:474-81. 27. Cuperus AMR, Harms MC, Rangel FA, Bronkhorst EM, Schols JGJH, Breuning KH. Dental models made with an intraoral scanner: a validation study. Am J Orthod Dentofacial Orthop 2012; 142:308-13. 28. Pauwels R, Beinsberger J, Collaert B, Theodorakou C, Rogers J, Walker A, et al., SEDENTEXCT Project Consortium. Effective dose range for dental cone beam computed tomography scanners. Eur J Radiol 2012;81:267-71. 29. American Dental Association Council on Scientific Affairs. The use of cone-beam computed tomography in dentistry: an advisory statement from the American Dental Association Council on Scientific Affairs. J Am Dent Assoc 2012;143:899-902.
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
Comparison of 3-dimensional dental models from different sources: Diagnostic accuracy and surface registration analysis Sercan Akyalcin,a David J. Dyer,b Jeryl D. English,c and Cagla Sard Houston, Tex, Okinawa, Japan, and Ankara, Turkey
Introduction: The aim of this study was to assess the diagnostic accuracy and surface matching characteristics of 3-dimensional digital dental models obtained from various sources. Methods: Three sets of maxillary and mandibular digital models of 30 subjects were included in this study. Three-dimensional stereolithography model files were obtained from a 3-dimensional laser desktop scanner (Ortho-Insight 3D; Motionview Software, Hixson, Tenn), the emodel system (GeoDigm, Chanhassen, Minn), and cone-beam computerized tomography. Arch-length discrepancy measurements were made on the 3-dimensional digital models and compared with direct caliper measurements. Additionally, stereolithography files from the 3 digital model systems were paired and superimposed using a best-fit algorithm. Average linear differences between the stereolithography shells were computed together with surface correlation amounts at various tolerance levels. Data were evaluated using intraclass correlation coefficients and the Tukey mean difference test. Results: Although all 3 digital model groups displayed good correlation with caliper measurements, the virtual scan models had the highest correlation with the manual method (ICC . 0.95). The Tukey mean difference test showed no consistent bias of one approach vs the others compared with caliper measurements; random errors were detected in all the comparisons. For the estimation of arch-length discrepancy, the mean bias of the scanned virtual models in comparison with caliper measurements (0.24 6 0.67 mm) was smaller than the mean biases of the emodels and the models generated from cone-beam computed tomography. Additionally, the best surface overlap correlation was observed between the virtual scanned models and the emodels. The mean linear distances between the stereolithography shells of these 2 model systems were 0.14 and 0.13 mm for the maxillary and mandibular arches, respectively. Conclusions: All 3 digital model systems can provide diagnostic information similar to caliper measurements, with varying degrees of agreement limits. The scanned virtual models had the least mean bias. A strong surface match correlation was observed between the virtual scanned models and the emodels, indicating that these could be used interchangeably. (Am J Orthod Dentofacial Orthop 2013;144:831-7)
P
laster models have been the gold standard in dental diagnosis and treatment procedures for many years. However, they require rigorous archiving and massive physical storage space.1,2 a Assistant professor, Department of Orthodontics, School of Dentistry, The University of Texas Health Science Center at Houston, Houston, Tex. b Orthodontist, United States Navy, Okinawa, Japan. c Chair and program director, Department of Orthodontics, School of Dentistry, The University of Texas Health Science Center at Houston, Houston, Tex. d Assistant professor, Department of Orthodontics, School of Dentistry, Baskent University, Ankara, Turkey. All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest, and none were reported. Address correspondence to: Sercan Akyalcin, 7500 Cambridge St, Suite 5130, Houston, TX 77054; e-mail, [email protected]. Submitted, June 2013; revised and accepted, August 2013. 0889-5406/$36.00 Copyright Ó 2013 by the American Association of Orthodontists. http://dx.doi.org/10.1016/j.ajodo.2013.08.014
Moreover, plaster models are not practical in the long term because of breakage and degradation issues associated with them.2 Accordingly, digital models made a rapid introduction into clinical practice in the late 1990s, and their clinical use has escalated since then.3 There have been numerous efforts to investigate the diagnostic accuracy and measurement sensitivity of digital models compared with plaster models.4-11 It was reported that information obtained from digital models is interchangeable with direct caliper measurements made on conventional plaster models. According to these reports, 2-point linear measurements such as intermolar and intercanine width,6,10 tooth size,9 tooth height,10 overjet, and overbite5,6,9 always show near-excellent agreement between conventional models and the analog alternatives. However, some other 831
Akyalcin et al
832
Fig 1. Digital model systems evaluated in this study: A, scanned virtual models; B, emodels; C, CBCTgenerated models.
studies have identified slightly high discrepancies between digital and stone models for comparisons of more complex diagnostic measurements such as the Bolton ratio, space analysis, and the irregularity ratio.6,8,12 Different techniques used in obtaining digital models inevitably cause standardization issues and might account for differences between the conventional plaster and digital models. Not only do the digital models differ from each other regarding their construction phases,3 but they also require the use of various proprietary softwares.2 Ideally, it should be possible to view, measure, and store 3-dimensional (3D) digital models universally regardless of the technique-specific details. The stereolithography (STL) file format is an industry standard for computer-aided design and manufacturing and allows the 3D files to be viewed and manipulated universally by many software packages. The aims of this study were to (1) evaluate the diagnostic accuracy of 3D STL files of the same dental anatomy obtained from different digitizing procedures and to compare them with direct caliper measurements using the same software, and (2) report the surface correlations of the 3D STL files of these model systems. MATERIAL AND METHODS
Approval for the study was obtained from the institutional review board of the University of Texas Health Science Center at Houston. The sample comprised the records of 30 orthodontic patients between the ages of 14 and 30 years. Each subject had maxillary and mandibular dental impressions made with Identic (Dux Dental, Oxnard, Calif) alginate along with bite registrations taken with polyvinyl siloxane material (Blue Moose; Parkell, Edgewood, NY). The impressions were immediately poured with Snow White Plaster (Kerr, Orange, Calif). The patients recruited for this study also
December 2013 Vol 144 Issue 6
had medium field-of-view (15 3 15 cm) cone-beam computed tomography (CBCT) scans in their files for various diagnostic purposes. All CBCT scans were made on the Galileos Comfort x-ray unit (Sirona Dental Systems, Bensheim, Germany) with exposure parameters of 85 kVp, 21 mA, 14 seconds, and 0.3-mm voxel size. CBCT scans with artifacts caused by patient movement and metal restorations were not included in the sample. Three sets of maxillary and mandibular digital models were obtained via different protocols. The first group consisted of virtual dental models obtained with a 3D desktop scanner (Fig 1, A). The Ortho Insight 3D laser scanner (Motionview Software, Hixson, Tenn) was used to directly scan the plaster models. The images were processed with Motionview software, which generated files with the STL extension for each dental model. The second group (Fig 1, B) included digital models made with the emodel system (GeoDigm, Chanhassen, Minn). For this study, the models were scanned, digitized, and converted to STL files by the company. Once finalized, the files were downloaded from the company's servers. The third group consisted of CBCT-generated digital dental models (Fig 1, C). CBCT scans of the study sample were uploaded to the company's servers (Anatomage, San Jose, Calif). The 3D images were segmented, digitally reformatted, and converted to STL files by the company. The digital models were then ready to download. All STL files in the 3 groups were saved in the same personal computer (Core i7-2600 CPU @3.40 Ghz with 8 MB of memory; Intel, Fort Worth, Tex). The plaster models were measured manually with a digital caliper. These measurements set the gold standard in this study. The same software was used for manipulation and measurements of the 3D STL files. The files were opened in random order with the Rapidform software program (Inus Technology, Seoul, South Korea). Mesiodistal tooth widths and dental arch
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
833
Fig 2. Surface registration analysis based on a best-fit algorithm displaying the deviations between the 2 model shells.
Table I. Mean maxillary and mandibular arch-length discrepancy values in the groups and ICCs for the comparisons
between the caliper measurements and the digital model systems Caliper measurements
Maxillary crowding (mm) Mandibular crowding (mm)
Mean 4.9 4.7
SD 3.8 3.1
Emodels Mean 4.4 4.3
SD 3.3 3.0
segments were measured on both the plaster models and the STL files to the nearest 0.01 mm. Maxillary and mandibular arch-length discrepancies were calculated based on these measurements. STL files from all 3 groups of digital models were also paired and digitally registered to create 3 groups that reflected the differences in the surface areas of these models. Surface registration of the model pairs was accomplished using the best-fit algorithm of the Rapidform software (Fig 2). After the registration protocol, average linear differences between the 2 STL files were recorded using the measurement tool of the same software. Additionally, surface overlap correlations between the file pairs were evaluated at 3 tolerance levels: 0.25, 0.5, and 1 mm. Statistical analysis
Two examiners (S.A. and C.S.), working independently, repeated all measurements twice 3 to 4 weeks later. The intraobserver repeatability and reproducibility of these observers were evaluated with intraclass correlation coefficients (ICCs). One-way analysis of variance (ANOVA) was used to compare the mean values of the measurements between the 4 groups. Agreement between the digital platforms compared with the caliper measurements was evaluated with both ICCs and the Tukey mean difference test (Bland-Altman analysis).
Virtual scanned models ICC 0.91 0.89
Mean 4.7 4.5
SD 3.3 3.0
ICC 0.95 0.96
CBCT-generated models Mean 4.5 4.2
SD 3.0 3.2
ICC 0.85 0.85
Statistical analyses were carried out with both SPSS statistics (version 20; IBM, Armonk, NY) and XLSTAT Mac (version 2012; Addinsoft, New York, NY). RESULTS
Both the intraobserver repeatability and the reproducibility of the observers showed near perfect agreement (ICCs . 0.92). According to 1-way ANOVA, no significant differences were found between the groups for the arch-length discrepancy measurements (Table I). The ICCs of the digital platforms compared with the caliper measurements are listed in Table I. Although all 3 digital model groups had good correlations with the caliper measurements, the virtual scan models had the highest correlation with the manual method (ICCs . 0.95). The findings from the Tukey mean difference test supported this finding and are summarized in Table II. Tukey mean difference plots can be seen in Figure 3. This analysis showed no consistent bias of 1 approach vs the others compared with the caliper measurements, and random errors were detected in all comparisons. For the estimations of arch-length discrepancies in both jaws, the mean bias (0.24 6 0.67 mm) of the scanned virtual models compared with the caliper measurements was smaller than the mean biases for the emodels and the CBCT-generated models. All 3
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
Akyalcin et al
834
Table II. Summary of the Tukey mean difference test
for the estimation of arch-length discrepancies in both arches Caliper measurements compared with
Bias SD 95% CI
Emodels 0.49 0.81 2.11 to 1.23
Scanned virtual models 0.24 0.67 1.62 to 1.1
CBCT-generated models 0.45 1.10 2.65 to 1.74
digital platforms provided similar measurements to the caliper measurements. However, from a clinical perspective, the measurements on the scanned virtual models provided better agreement with the manual caliper measurements, as shown by the smaller confidence interval (agreement limit) range (Table II). Table III presents the summary of the surface registration analysis of the matched pairs of maxillary and mandibular digital models. The best surface overlap correlation was observed between the virtual scanned models and the emodels. The mean linear distances between the STL shells of these 2 model systems were 0.14 and 0.13 mm for the maxillary and mandibular arches, respectively. However, when the virtual scanned models or the emodels were paired with CBCT-generated models, the mean surface difference increased by almost 0.2 mm in both cases. A 95% surface overlap correlation was observed between the virtual scanned models and the emodels at a tolerance level of 0.5 mm (Table IV). This means that if a 0.5-mm average linear distance between the 2 STL files could be tolerated, 95% of the surface area would coincide. However, the same amount of surface overlap correlation (95%) did not exist in the paired groups, including the CBCT-generated models, until the tolerance level was doubled. At the 1-mm tolerance level, the virtual scanned models and the emodels displayed 99% correlations. DISCUSSION
Previous studies have investigated the measurement accuracy of digital model systems with their brandspecific softwares.4-12 This could lead to conflicting results because of software differences and potential operator calibration issues with the use of multiple programs. Digital dental models that are basically 3D files should ideally be manipulated and analyzed with high precision, independent of the software interface. Therefore, in our study, it was intended to compare the 3D STL shells of the same dental anatomy from various digitizing sources using the same software program. However, landmark identification in any 3D
December 2013 Vol 144 Issue 6
software will likely differ from actual physical measurements because there is no caliper landing on physical points during the process. Additionally, no measurement of a physical object is exact. Some degree of error is associated with the measured data in all experiments. This might explain some of the statistical variability between the mean values of caliper and digital measurements. However, because both the repeatability and reproducibility of the results were confirmed, the diagnostic accuracy of the digital model systems was established using both correlations and agreement limits. According to our results, all 3 digital systems investigated here showed good correlations with manual caliper measurements when diagnosing the archlength discrepancy using ICCs. Although ICCs are a good retest measure,13 determining method reliability requires a comprehensive set of statistical methods.14,15 This is why the Tukey mean difference test was also applied to the data. In essence, the digital platforms showed a similar pattern of systematic errors that can be observed on the Tukey mean difference plots (Fig 3). However, potential differences between the construction phases of the digital models led to varying degrees of bias and agreement limits as they were compared with the manual measurements. As can be evidenced by smaller mean bias and agreement limits, the scanned virtual models slightly outperformed the emodels and the CBCT-generated models. Studies comparing emodels with conventional plaster models have suggested that digital models offer interchangeable information with plaster models when used for orthodontic diagnostic purposes.3,7,8 In our study, the emodels had less bias than did the CBCTgenerated models. Recent studies have demonstrated that linear measurements from CBCT-generated digital models have similar levels of accuracy when compared with other digital model systems.16-19 However, a high level of accuracy as well as high-quality image detail are required before CBCT-generated models can generally be accepted. As demonstrated by our findings, CBCT-generated models are not perfect yet. In our study, they had the largest mean bias and confidence interval. In some instances, almost 3-mm differences were observed for the estimation of dental crowding between the measurements on the CBCT-generated models and the caliper measurements. This much difference might have clinical implications, especially in severely crowded dentitions.20 The second part of our study, evaluating the correlation between the surface areas of the STL files obtained from the digital sources, yielded a parallel outcome to the measurement accuracy findings. In a
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
835
Fig 3. Tukey mean difference plots for comparison of digital model systems with direct caliper measurements. Table III. Mean linear distances between the surface
areas of the maxillary and mandibular digital model pairs
Maxillary model pairs Scanned virtual models and emodels Scanned virtual models and CBCT- generated models CBCT-generated models and emodels Mandibular model pairs Scanned virtual models and emodels Scanned virtual models and CBCT-generated models CBCT-generated models and emodels
n
Mean (mm)
SD
30 30
0.14 0.32
0.04 0.06
30
0.34
0.07
30 30
0.13 0.31
0.04 0.06
30
0.32
0.07
recent report, Park et al21 demonstrated that surface registration was a simpler, more reproducible and reliable method than superimposition based on anatomic structures. Therefore, we used a regional registration with surface-to-surface matching (best-fit method) in our study. According to our results, a 95% surface overlap correlation did not exist in the comparisons of the CBCT-generated models with the other 2 digital model systems until 1.00 mm of difference between the surface areas of the 2 shells was tolerated. However, the surface overlap correlation between the virtual scanned models and the emodels was more than 95% at a 0.5mm tolerance for both the maxillary and mandibular model sets. This finding suggests that the surface areas of the CBCT-generated models could lack the finer details such as occlusal pits and fissures. The lack of detail in this group can even be observed with the naked eye when examining the 3 digital model systems closely (Fig 1). This is partly due to the segmentation process of maxillary and mandibular teeth during the construction phase of the CBCT-generated models. Digital models synthesized from CBCT images are constructed from scans with the patient's teeth in occlusion. This inevitably leads to overlapping of the images and
makes the segmentation process arduous and timeconsuming.22,23 Second, spatial resolution might play an important role on the accuracy of 3D tooth reconstructions. In a study that compared 3D tooth reconstructions using CBCT and microcomputed tomography data, the most important differences occurred mainly where the tooth was thinner, either near the cervix of the still-developing enamel cap or at the incisal edges of the anterior teeth.24 This finding points out the importance of isotropic voxel size when capturing finer details on physical models. It was also reported that field-of-view selection has a significant influence on the quality of the 3D surface models of the dental arches from 1 CBCT system.22,25 Al-Rawi et al22 evaluated the accuracy of 3D CBCT reconstructions of the tooth crowns and occlusal surfaces with 3 field-of-view sizes and concluded that the choice of a large field of view reduced the visibility of the occlusal surfaces compared with small and medium fields of view. This is due to the decrease in voxel size in scans with larger fields of view. Recently, Lightheart et al18 compared CBCTgenerated digital models with OrthoCAD models to evaluate surface overlap correlation between them. They demonstrated fair agreement between the 2 digital model systems, where only 90% matching occurred at the 1-mm to 1.25-mm tolerance level range. Accordingly, they signified the importance of gingival tissues, since digital models obtained by optical laser scanning differ from CBCT-generated-models in these areas. It is clearly impossible to capture the gingiva with CBCT technology. Therefore, they suggested that tasks related to gingival surface areas—eg, design of a vacuumformed appliance—should not be performed on CBCTgenerated models. Our results are in agreement with these findings. Although the types of files were standardized and the linear measurements were all done on STL files in our study, accuracy of impression-based digital
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
Akyalcin et al
836
Table IV. Surface overlap (%) for the maxillary and mandibular digital model pairs at various tolerance levels Scanned virtual models and emodels Maxillary Tolerance level (mm) 0.25 0.50 1.00
Mean 84 95 99
SD 8 2 0
Mandibular Mean 86 96 99
SD 9 4 1
Scanned virtual models and CBCT-generated models Maxillary Mean 51 78 95
models might have been limited by the impression or the stone material. According to Torassian et al,26 the alginate material we used in our project showed a statistically and clinically significant change in all dimensions within 72 hours. In our study, all impressions were poured immediately. Therefore, distortion during the construction phase of the plaster models, if any, was kept minimal. However, it could have been a better approach to use alginate substitutes that are dimensionally accurate over an extended period of time. Perhaps eliminating the impression step completely and acquiring digital models with intraoral scanners will soon be the gold standard to retrieve 3D models of dental arches. However, intraoral scanners need to be developed to overcome their initial shortcomings such as restricted scanning time and the difficulties in scanning contact points and the vertical dimensions of gingivae in the dental arch.27 Based on the limitations of our study, optical laser scanning of the impression or the plaster model is the best method for digitizing patient records. From a radiation-protection point of view, a CBCT scan of the patient must never be the primary source for constructing digital models.28,29 In other words, use of ionizing radiation to generate digital models as a substitute for other digitizing sources is not acceptable. In our study, we used patients who already had their CBCT scans taken because of conclusive evidence suggesting a diagnostic need: eg, surgical intervention. This approach made it possible for us to evaluate the precision of the CBCT-generated models along with other digitizing sources. However, CBCT-generated models are not recommended for routine clinical use. Under justified circumstances, their generation will require a good-quality scan with no movement artifacts and with adequate spatial resolution. Based on our findings and what we know from the previous studies, a voxel size \0.3 mm should give adequate diagnostic information.15,24 However, CBCT-generated models with smaller voxel sizes should be further evaluated for their surface details and precision.
December 2013 Vol 144 Issue 6
Mandibular SD 8 8 2
Mean 52 80 96
SD 9 8 2
CBCT-generated models and emodels Maxillary Mean 48 77 95
SD 9 8 3
Mandibular Mean 52 80 96
SD 10 9 3
CONCLUSIONS
Arch-length discrepancy measurements made on digital model files obtained from different sources including emodels, a desktop scanner (virtual scanned models), and CBCT-generated models had similar patterns of random error when compared with direct caliper measurements. However, scanned virtual models provided better agreement with plasters casts. The best surface overlap correlation was observed between the virtual scanned models and the emodels, indicating a strong match between the STL shells of these 2 systems. REFERENCES 1. Peluso MJ, Josell SD, Levine SW, Lorei BJ. Digital models: an introduction. Semin Orthod 2004;10:226-38. 2. Akyalcin S. Are digital models replacing plaster casts? Dentistry 2011;1:e102. 3. Fleming PS, Marinho V, Johal A. Orthodontic measurements on digital study models compared with plaster models: a systematic review. Orthod Craniofac Res 2011;14:1-16. 4. Goonewardene RW, Goonewardene MS, Razza JM, Murray K. Accuracy and validity of space analysis and irregularity index measurements using digital models. Aust Orthod J 2008;24:83-90. 5. Leifert MF, Leifert MM, Efstratiadis SS, Cangialosi TJ. Comparison of space analysis evaluations with digital models and plaster dental casts. Am J Orthod Dentofacial Orthop 2009;136: 16.e1-4. 6. Quimby ML, Vig KW, Rashid RG, Firestone AR. The accuracy and reliability of measurements made on computer-based digital models. Angle Orthod 2004;74:298-303. 7. Mullen SR, Martin CA, Ngan P, Gladwin M. Accuracy of space analysis with emodels and plaster models. Am J Orthod Dentofacial Orthop 2007;132:346-52. 8. Stevens DR, Flores-Mir C, Nebbe B, Raboud DW, Heo G, Major PW. Validity, reliability, and reproducibility of plaster vs digital study models: comparison of peer assessment rating and Bolton analysis and their constituent measurements. Am J Orthod Dentofacial Orthop 2006;129:794-803. 9. Horton HM, Miller JR, Gaillard PR, Larson BE. Technique comparison for efficient orthodontic tooth measurements using digital models. Angle Orthod 2010;80:254-61. 10. Keating AP, Knox J, Bibb R, Zhurov AI. A comparison of plaster, digital and reconstructed study model accuracy. J Orthod 2008; 35:191-201. 11. Santoro M, Galkin S, Teredesai M, Nicolay O, Cangialosi TJ. Comparison of measurements made on digital and plaster models. Am J Orthod Dentofacial Orthop 2003;124:101-5.
American Journal of Orthodontics and Dentofacial Orthopedics
Akyalcin et al
12. Tomassetti JJ, Taloumis LJ, Denny JM, Fischer JR Jr. A comparison of 3 computerized Bolton tooth-size analyses with a commonly used method. Angle Orthod 2001;71:351-7. 13. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86:420-8. 14. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8:135-60. 15. Rankin G, Stokes M. Reliability of assessment tools in rehabilitation: an illustration of appropriate statistical analyses. Clin Rehabil 1998;12:187-99. 16. Kau CH, Littlefield J, Rainy N, Nguyen JT, Creed B. Evaluation of CBCT digital models and traditional models using the Little's index. Angle Orthod 2010;80:435-9. 17. Creed B, Kau CH, English JD, Xia JJ, Lee RP. A comparison of the accuracy of linear measurements obtained from cone beam computerized tomography images and digital models. Semin Orthod 2011;17:49-56. 18. Lightheart KG, English JD, Kau CH, Akyalcin S, Bussa HI, McGrory KR, et al. Surface analysis of study models generated from OrthoCAD and cone-beam computed tomography imaging. Am J Orthod Dentofacial Orthop 2012;141:686-93. 19. Tarazona B, Llamas JM, Cibrian R, Gandia JL, Paredes V. A comparison between dental measurements taken from CBCT models and those taken from a digital method. Eur J Orthod 2013;35:1-6. 20. Redich M, Weinstock T, Abed Y, Schneor S, Holdstein Y, Fischer A. A new system for scanning, measuring and analyzing dental casts based on a 3D holographic sensor. Orthod Craniofac Res 2008; 11:90-5. 21. Park TJ, Lee SH, Lee KS. A method for mandibular dental arch superimposition using 3D cone beam CT and orthodontic 3D digital model. Korean J Orthod 2012;42:169-81.
837
22. Al-Rawi B, Hassan B, Vandenberge B, Jacobs R. Accuracy assessment of three-dimensional surface reconstructions of teeth from cone beam computed tomography scans. J Oral Rehabil 2010; 37:352-8. 23. Fourie Z, Damstra J, Schepers RH, Gerrits PO, Ren Y. Segmentation process significantly influences the accuracy of 3D surface models derived from cone beam computed tomography. Eur J Radiol 2012;81:e524-30. 24. Maret D, Molinier F, Braga J, Peters OA, Telmon N, Treil J, et al. Accuracy of 3D reconstructions based on cone beam computed tomography. J Dent Res 2010;89:1465-9. 25. Hassan B, Couto Souza P, Jacobs R, de Azambuja Berti S, van der Stelt P. Influence of scanning and reconstruction parameters on quality of three-dimensional surface models of the dental arches from cone beam computed tomography. Clin Oral Investig 2010; 14:303-10. 26. Torassian G, Kau CH, English JD, Powers J, Bussa HI, Marie SalasLopez A, et al. Digital models vs plaster models using alginate and alginate substitute materials. Angle Orthod 2010;80:474-81. 27. Cuperus AMR, Harms MC, Rangel FA, Bronkhorst EM, Schols JGJH, Breuning KH. Dental models made with an intraoral scanner: a validation study. Am J Orthod Dentofacial Orthop 2012; 142:308-13. 28. Pauwels R, Beinsberger J, Collaert B, Theodorakou C, Rogers J, Walker A, et al., SEDENTEXCT Project Consortium. Effective dose range for dental cone beam computed tomography scanners. Eur J Radiol 2012;81:267-71. 29. American Dental Association Council on Scientific Affairs. The use of cone-beam computed tomography in dentistry: an advisory statement from the American Dental Association Council on Scientific Affairs. J Am Dent Assoc 2012;143:899-902.
American Journal of Orthodontics and Dentofacial Orthopedics
December 2013 Vol 144 Issue 6
