Accuracy of 3D-Printed Dental Models of Different Tooth Surfaces

Background: Few studies have been reported regarding the accuracy of 3D-printed models for orthodontic applications. The aim of this study was to assess the accuracy of 3D-printed dental models of different tooth surfaces. Methods (cid:0) Thirty maxillary dental models were produced by means of a stereolithography-based 3D printer. Each printed model was digitally scanned and compared with the oral-scanned STL le via superimposition analysis, with a colour map used to assess the accuracy of different surfaces of anterior and posterior teeth. Results: Statistically signicant differences were found in the average deviations of different tooth surfaces. The mean average absolute deviations of the occlusal surfaces of posterior teeth were greater than those of other surfaces. Percentages of points beyond the upper and lower limits of different tooth surfaces displayed the same results. Conclusions: The occlusal surfaces of posterior teeth on 3D-printed dental models show greater deviations than those of other regions, especially in pits and ssures.


Background
Rapid prototyping (RP) technologies provide the possibility for a physical dental model to be generated from digital data economically and conveniently. These three-dimensionally printed dental models have saved dentists from the dilemma of storage space, risks of damage, and the inconvenience of miscommunication. The 3D-printed dental model is being increasingly used in orthodontic diagnosis and treatment. However, it is subject to deviations due to the accuracy of the 3D printer.
Currently, the most commonly used techniques for orthodontics applications are PolyJet and stereolithography (SLA). Due to its high printing resolution and fast forming speed, there has been an increase in the number and availability of SLA-based printers in recent years. 1 Digital light processing (DLP) is a subset technology of SLA printing that uses a projected planar image of light to photopolymerise resin. 2 During this process, the printing plate moves up in a given Z-axis distance (for example, 0.025 mm, 0.05 mm, 0.1 mm), and photopolymerising resin is projected to the light of a speci c wavelength (405 nm) that cures a cross-section layer by layer. This printing process is repeated until the model is complete. It usually takes up to several hours for a dental model to be printed with this technology, depending on the Z-axis print layer thickness set by the printer. 3 However, little research has been reported regarding the accuracy of 3D-printed models for orthodontic applications. Most of the literature compares linear measurements of 3D-printed models with those of stone models. [4][5][6][7] Zhang et al. compared the accuracies of 3D-printed dental models using three types of DLP and SLA printers at different thicknesses and found that the printing accuracy of all printers was higher at 50 μm. 8 Park and Shin 9 compared the accuracy and reproducibility of conventional dental casts and 3D-printed models fabricated by three types of printers. The results showed that the volumetric changes in casts made by the conventional method were signi cantly smaller than those of 3D-printed casts. Brown et al. 10 compared the tooth and arch measurements of three model types (digital, DLP, and PolyJet) with those of stone models Their results indicated high degrees of agreement among all types of models for all measurements, except the crown height measurements between the stone and DLP models. However, linear measurements indicate that the print model is consistent with the plaster model, but the results of linear measurements can demonstrate only that the 3D-printed model is suitable only for use in orthodontics study. More and more orthodontic appliances are being manufactured based on 3D-printed models; therefore, accuracy of tooth surfaces is critical. Models with inaccurate 3D-printed surfaces may lead to the poor t of an orthodontics appliance. The purpose of this study was to assess the accuracy of 3D-printed dental models of different tooth surfaces.

Methods
A sample of 30 maxillary dental models was included in this study. Thirty volunteers who met the criterion for inclusion (having a complete permanent dentition) and without the exclusion criteria (having obvious dental anomalies in size and shape and severe dental crown defects) were randomly selected from the hospital. First, digital dentition data were obtained by intraoral scan (iTero, Align Technology Inc., San Jose, CA). These scanned data were set as prime STL reference models. Thirty STL reference models were printed with DLP rapid prototyping technologies (DentLab One, SHINING 3D, Hangzhou, China), with a build layer thickness of 0.05 mm. 11 The 3D-printed models were then scanned with the same intraoral scan as the STL test models.
The digital models generated from both intraoral scan and 3D-printed models were exported to Geomagic Control software (3D Systems, Rock Hill, SC) as STL format les for model superimposition. Each tooth of the 3D-printed model was segmented separately into buccal, lingual, and occlusal surfaces and superimposed with the corresponding tooth of the STL reference model, then merged for calculation of the deviation of the STL reference model (Fig. 1). With 'canine to canine' as the anterior region and ' rst premolar to second molar' as the posterior region, the superimposition was done separately for the buccal surfaces of anterior teeth (BA), lingual surfaces of anterior teeth (LA), occlusal surfaces of anterior teeth (OA), buccal surfaces of posterior teeth (BP), lingual surfaces of anterior teeth (LP), and occlusal surfaces of anterior teeth (OP). The segmenting procedure was done twice independently by two researchers.
Observer consistency was calculated for reliability. Each 3D-printed model scanning le (labeled as the STL test model) was then individually superimposed on the oral-scanning le (labeled as the STL reference model) by means of an automated best-t algorithm (Fig. 2). Geomagic software showed the means and standard deviations of different tooth surface distances between the STL reference models and the STL test models. A 0.10-mm threshold parameter was set as the critical value for the analysis of deviations between the STL reference model and the STL test model. Any points in the test le deviating from the reference le by more than 0.10 mm in the positive or negative direction were considered to be beyond the upper or lower limits, accordingly. Reports were generated for separate calculation of the total positive and negative deviations.
IBM SPSS Statistics 20.0 (IBM, Chicago, IL) was used for statistical analysis. The Tukey test was used to evaluate the differences between the superimposition of the STL test model and the STL reference model. P values less than 0.05 were considered to be signi cant.

Results
The repeatability of the measurements in this study was high, with an intraclass correlation coe cient value of 0.878. The colour map showed the deviations of the STL test model and the STL reference model (Fig. 2). A 0.10-mm threshold parameter was set as the critical value for the analysis of deviations between the reference le and each test le. The darker the colour, the larger the variance, and the lighter the colour, the smaller the variance, i.e. red and blue showed greater deviations than yellow and green. It could be clearly seen from the colour map that the colours of occlusal surfaces of posterior teeth were much darker than those of other surfaces.
The average deviations of 3D-printed dental models of different tooth surfaces were then collected and compared ( Table 1). The average deviation of the anterior region was generally smaller than that of the posterior region, which can be seen clearly from the box chart (Fig. 3). The means and standard deviations of the occlusal surfaces of posterior teeth were greater than those of other surfaces, with a statistically signi cant difference (P < 0.05). Percentages of points beyond the upper and lower limits of different tooth surfaces were also compared ( Table 2). The variance between the different tooth surfaces was more pronounced. The occlusal surfaces of posterior teeth appeared to have higher percentages of points beyond the upper and lower limits, which showed greater deviation from the reference models (P < 0.05) (Fig. 4).  tooth, arch, and occlusal measurements. 12 They found that the difference was larger in the posterior region than in the anterior region in superimposed 3D digital models, which is consistent with the results of this study.

Discussion
The results of our research clearly demonstrated that the occlusal surfaces of posterior teeth of 3Dprinted models deviated primarily from those of the intraoral scan model. Both the average absolute deviations and percentages of points beyond the upper and lower limits of different tooth surfaces provided con rmatory evidence for this discovery. Although, in this study, the average difference between the 3D-printed model and the reference model appeared as a small inclusion occlusal surface, the deviation in the region of posterior pits and ssures was obvious. The dark red colour in this region showed the deviation here to be even more than 0.1 mm, which may have some signi cant clinical effects. For example, a 3D-printed template or tray cannot be tted well onto this 3D-printed working model in the laboratory.
Printing errors in the 3D-printed model can arise from each link of the printing process and the parameters thereof. These include residual polymerisation of the resin, effects of support structures, print resolution (X and Y planes), layer thickness (Z plane), and surface nishing. 13  During the rapid prototyping process, the photosensitive resin, which is sticky and requires manual cleaning, will be cured by the ultraviolet laser. 15 The liquid adhesive can ow along the smooth surface but can easily remain on the pits and ssures of the occlusal surfaces of posterior teeth. If it is not cleaned completely or not cleaned in time, the material at the bottom of the groove will cure by itself, as occurs during pit and ssure sealing. The deeper the ssure, the greater the deviation (Fig. 5) However, few investigations have focused on deviations of the different tooth regions (anterior and posterior teeth) and surfaces. This study compared deviations of 3D-printed models and reference models on buccal, lingual, and occlusal surfaces of anterior and posterior teeth. No study has drawn rm and reliable conclusions as to whether the deviations between 3D-printed models and a reference model are clinically acceptable. It remains controversial whether differences in dimensions between the reference model and the 3D-printed models affect the accuracy of orthodontic appliances. Kasparova et al. compared traditional plaster casts, digital models, and 3D-printed models and found 3D-printed models to have advantages over traditional plaster casts due to their accuracy and price. 6 Wan Hassan et al. compared the accuracy of measurements made on rapid prototyping and stone models with different degrees of crowding. 4 They found signi cant differences for all planes in all categories of crowding except for crown height in the moderate crowding group and arch dimensions in the mild and moderate crowding groups. They concluded that the rapid prototyping models were not clinically comparable with conventional stone models.
Intraoral or extraoral scanning is becoming more and more common and may even replace traditional models in the future, but there is no clear evidence as to whether digital models and 3D-printed models can take the place of stone models to produce some orthodontic appliances in the laboratory. Even designed and produced with digital models, those appliances still need to be tried on the 3D-printed models. Further, due to the relatively obvious print errors on the occlusal surfaces of posterior teeth, some appliances and templates made with digital models cannot be fully placed on 3D-printed models. 17 Some measurement differences that occur in these 3D-printed dental models will affect the accuracy of manufactured orthodontic appliances, especially the t on occlusal surfaces. Deviations in the occlusal template for orthognathic surgery will affect the precision of the surgery. 3D-printing technology is also widely used in the design and manufacture of clear aligners. As is well-known, differences between sequential aligners are only 0.2-0.3 mm, so errors over 0.3 mm may in uence the expression of tooth movement. Cole et al. examined the accuracy of 3D-printed retainers compared with conventional vacuum-formed and commercially available vacuum-formed retainers. 18 The results showed the least deviation from the original reference models in the conventional vacuum-formed retainers and the greatest deviation in the 3D-printed retainers. However, the deviation was clinically acceptable. Further research is needed to con rm the precision and clinical acceptability of 3D-printed models in orthodontics clinics.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests. Authors' contributions DT performed the main study and drafted the manuscript. XLG and WXT helped to acquire data. YLJ analyzed and interpreted the data. YNS was a major contributor in designing the manuscript. FB designed the main scheme of the research and modi ed the manuscript. All authors read and approved the nal manuscript. Figure 1 Schematic gure, illustrating the design of this study. Scanning models generated from intraoral scans and 3D-printed models were exported to Geomagic Control software. Tooth surface segmentation was performed on 3D-printed scanning models, and each tooth surface was individually superimposed on the reference model.