2.1.

Model Fabrication

The anatomic model of a maxillary full denture, based on standard working models (frasaco GmbH, Tettnang, Germany), was created with CAD and commercially available software (Meshmixer, Autodesk Inc., San Rafael, California). The parallel cylinders, with a predefined inner diameter and a depth of 4 mm, were placed at tooth positions 17 (C1), 21 (C2), and 27 (C3) as reference elements (see photograph in Fig. 1). A characteristic crown preparation at tooth 23, and an inlay preparation at tooth 16, simulated a normal prosthetic situation. The master model was milled out of an industrially manufactured polyetheretherketone (PEEK) block (Denseo PEEK blank, Denseo GmbH, Aschaffenburg, Germany) on a five-axis computerized-numerical-control milling machine (SilaMill 5, vhf camfacture AG, Ammerbuch, Germany). PEEK is a high-performance polymer, and it is known to be dimensionally stable ²² and is also used in medical applications, see, e.g., Ref. 23. An extension on the back of the model served for mounting purposes.

Fig. 1

The PEEK model’s size corresponds to that found in the human body, as displayed by the photographs on the left. Three well-defined hollow cylinders were incorporated to determine the precision of the IOSs. The other photograph shows the placement of the model on the rotation stage in the CT-system nanotom^® m (phoenix|x-ray, GE Sensing & Inspection Technologies GmbH, Wunstorf, Germany).

In addition, two other models were fabricated using the same design. A photopolymer resin, namely dental model resin (Formlabs, Somerville, Massachusetts), was used within the SLA printers form 2 (Formlabs, Somerville, Massachusetts) and with a layer thickness of $25\mu \mathrm{m}$. Models fabricated by the stereolithography technology were washed with isopropyl alcohol for a period of 15 min. Subsequently, they were postcured with UV light under an inert gas atmosphere. The density of the models was experimentally determined to $1.59\mathrm{g}/{\mathrm{cm}}^3$, which is significantly higher than the value of the raw material at $1.12\mathrm{g}/{\mathrm{cm}}^3$. This result is an indicator of cross-linking at the 405-nm wavelength.

2.2.

Reference Generation

The high-resolution CT ($\mu \mathrm{CT}$) scans of the master model, described already, served as the reference data. Tomographic data were recorded using the advanced conventional system nanotom^® m (phoenix|x-ray, GE Sensing & Inspection Technologies GmbH, Wunstorf, Germany), as shown in Fig. 1. This system is equipped with a nanofocus tube with a maximal acceleration voltage of 180 kV, which produces power of up to 15 W. For data acquisition, we employed the maximal acceleration voltage and a beam current of $30\mu \mathrm{A}$. To shift the mean photon energy to higher values, a 0.5-mm-thin copper film was placed behind the transmission target. We recorded 1600 radiographs throughout 360 deg. The exposure time for the first two datasets was set to 3 s per radiograph, whereas it was increased to 9 s for the third and to 24 s for the fourth dataset. To investigate the repeatability of the $\mu \mathrm{CT}$-system and the impact of the cone beam, the following source–sample distances (SSDs) and source–detector distances (SDDs) were positioned. For the first and the second scans, SSD and SDD were 78.75 and 225.00 mm, respectively. For the third scan, we used SSD 157.50 mm and SDD 450.00 mm, and for the fourth scan SSD 210.00 mm and SDD 600.00 mm were used. Exposure times were adapted to maintain comparable intensities, and the effective pixel length corresponded to $35\mu \mathrm{m}$. These approaches keep the magnification constant. A large SSD limits the angle of the cone beam and might improve the accuracy but leads to significantly longer exposure times.

In a previous study,² we have captured a dental model using the nanotom^® m and the tactile coordinate measuring machine Leitz PMM 864 (Hexagon Metrology GmbH, Wetzlar, Germany). Both approaches provided true micrometer resolution. Consequently, the micrometer precision of the nanotom^® m system for centimeter distances was validated.

For visualization purposes, the acquired volume data were processed by means of VGStudio MAX (Volume Graphics, Heidelberg, Germany). In addition, the data size was reduced and converted into the STL format. This conversion enabled us to compare the tomography data with data for the intraoral scans.

2.3.

Data Acquisition, Using Intraoral Scanners

This study comprises five IOS systems, namely the 3M™ True Definition Scanner (3M ESPE, St. Paul, Minnesota), the TRIOS^® 3 (3shape, Copenhagen, Denmark), the CS 3600 (Carestream, Atlanta, Georgia), the Medit i500 (Medit corp., Seongbuk-gu, South Korea), and the Emerald™ (Planmeca Oy, Helsinki, Finland). With each system, the master model was scanned 10 times to obtain the necessary statistics. For this purpose, the model was mounted in the anatomically correct upright position. One trained examiner (M.S.) performed all of the scans in identical conditions (light, temperature, etc.), and scanner handling, the use of powder, and the scan path were carried out according to the manufacturer’s guidelines. 3M recommended the use of powder to improve the scanning results. To prepare the model as recommended, a thin layer of powder (3M Powder Sprayer; 3M, St. Paul) was applied to the model. The scans with the other four systems were taken without the application of powder, i.e., in a powder-free fashion.

2.4.

Data Procession and Evaluation

The measurements and analyses were performed using the well-established software GOM Inspect (GOM GmbH, Braunschweig, Germany). The positions of the three hollow cylinders were determined using their center points, identified via the Gaussian best-fit method, see Fig. 2. The three distances were derived from tomography and IOS data individually, see Fig. 3.

Fig. 2

Scheme for determining the center points of the hollow cylinders via the Gaussian best-fit method.

Fig. 3

Scheme for specifying the three distances. Distance $d_1$ characterizes the length between 17 and 21, distance $d_2$ is the length between 21 and 27, and distance $d_3$ corresponds to the length between 17 and 27.

To compare accuracy and precision, these data were matched to the reference. The global best-fit deviation was calculated for each scan, in order to visualize the displacement field, see what follows.

3.1.

Measurements with nanotom^® m

To confirm the well-known stability of the $\mu \mathrm{CT}$-system and the reproducibility of the tomographic data acquisition, the first two scans were performed in identical conditions. As expected, differences were within the micrometer range, and even for distances between 4 and 5 cm, as given by the spacing between the hollow cylinders, the differences were well below the voxel length of $35\mu \mathrm{m}$. For scan 1, we found $d_1=45.444\mathrm{mm}$, $d_2=41.622\mathrm{mm}$, and $d_3=48.098\mathrm{mm}$. For scan 2, the selected distances were determined to $d_1=45.450\mathrm{mm}$, $d_2=41.627\mathrm{mm}$, and $d_3=48.101\mathrm{mm}$. This result, i.e., differences of 4 to $6\mu \mathrm{m}$ or relative deviations by about ${10}^{-4}$, indicates that imaging modality and the method for determining longitude are appropriate choices for precision measurements. As a consequence, we assumed an error bar of $6\mu \mathrm{m}$ for establishing the spacing between the hollow cylinders.

To evaluate the impact of the cone-beam geometry in the nanotom^® m on the accuracy, two scans with modified sample positions with respect to source and detector were performed. The experimentally determined distances between the hollow cylinders of scan 3 were $d_1=45.468\mathrm{mm}$, $d_2=41.640\mathrm{mm}$, and $d_3=48.127\mathrm{mm}$. The values for scan 4 amounted to $d_1=45.480\mathrm{mm}$, $d_2=41.647\mathrm{mm}$, and $d_3=48.134\mathrm{mm}$. Compared to the set value, given by the STL file with $d_1=45.503\mathrm{mm}$, $d_2=41.676\mathrm{mm}$, and $d_3=48.192\mathrm{mm}$, we identified values only 23, 29, and $58\mu \mathrm{m}$ smaller. Although the discrepancies amount only to 1/10th of a percent and less than two pixel lengths, the phenomenon was still detectable. Therefore, tomographic data acquisition for the three-dimensionally printed models was performed with the parameters in scan 4. It should be noted, however, that the potential gain in accuracy meant that data acquisition took seven times longer to complete than the standard approach.

3.2.

Geometry of the PEEK Model

The PEEK model was generated from the STL-data by a milling machine, it is therefore unknown how far the model corresponds to the desired geometry. Comparing the STL data with the tomography data, one obtains a superposition of deficiencies from milling and imaging. Previous studies with the nanotom^® m, however, demonstrate true micrometer accuracy, which is further supported by the measurements described above. Therefore, we can reasonably estimate the accuracy of the PEEK machining through a direct comparison of STL-data with high-resolution tomography data.

Figure 4 elucidates the differences between the design and the measured data using the color code on the color bar. The red color indicates positions, where material more than $100\mu \mathrm{m}$ thick should be further removed. The blue color represents positions, where the milling tool removed more than originally desired, and the green color shows positions in perfect agreement. Since the hollow cylinders are almost identical, their axes and related center points are not affected. Measurement of the distances described already is therefore meaningful.

Fig. 4

3D representation of the differences between the desired geometry, given by the STL-file and the data from scan 4 and according to the color bar and the related values on the right. Perfect agreement is represented by the green color. Obviously, the milling tool provided a reasonable result, although in some areas more material than desired was removed (blue color), while other areas show excess material (red color). One can further observe that the hollow cylinders are larger in size than planned, but because we only consider the center points, the determination of distances is hardly affected.

3.3.

Geometry of the SLA Models

In specific situations, the digital workflow has to be modified, which might require a physical model to be generated from the IOS data. Here, the method of choice is stereolithography,²⁴ but current clinical experience shows that these SLA-models are less accurate than conventional impressions. To quantify this level of accuracy, we produced two SLA-models and used tomographic imaging and registration.

Based on the high-resolution tomography measurements performed as described already, the distances $d_1$, $d_2$, and $d_3$ were extracted. For the first SLA-model, the three distances corresponded to 45.611, 41.830, and 48.210 mm, whereas for the second SLA model, we found 45.753, 41.831, and 48.247 mm, meaning that the two models were not exactly the same in terms of geometry. The differences were in the submillimeter range, see Fig. 5. Comparison with STL-data, however, is the benchmark. The average deviation along the full arch amounted to $(49\pm 9)\mu \mathrm{m}$. For the distances considered within this study, the deviations accounted for $\mathrm{\Delta }{d}_1=108\mu \mathrm{m}$, $\mathrm{\Delta }{d}_2=154\mu \mathrm{m}$, and $\mathrm{\Delta }{d}_3=18\mu \mathrm{m}$ (first SLA model), and $\mathrm{\Delta }{d}_1=250\mu \mathrm{m}$, $\mathrm{\Delta }{d}_2=155\mu \mathrm{m}$, and $\mathrm{\Delta }{d}_3=55\mu \mathrm{m}$ (second SLA model). These differences between the desired design and the actual physical models clearly indicate the limitations of the stereolithography printers currently used in dental offices.

Fig. 5

3D representation of the differences between the desired geometry, given by the STL-file, we used for the model preparation (photograph on the left) and the high-resolution tomography data of the two 3D-printed models. The green color shows parts without differences, whereas the red and blue colors indicate excess and deficiency of material, respectively.

3.4.

Accuracy of the Intraoral Scanners

The results obtained from the 10 independent experiments with each of the IOS systems are listed in Table 1. These numbers show the reproducibility of the individual devices, when an experienced dentist has performed data acquisition according to the guidelines provided by suppliers. The data scattered within a few tens of micrometers.

Table 1

Measured distances d1, d2, and d3 (in mm) obtained from the IOS systems used. The data determined from the μCT scans (d1=45.480  mm, d2=41.647  mm, and d3=48.134  mm) can be regarded as ground truth.

To determine the accuracy of the IOS systems in measuring the three selected distances, we compared these values with tomography data for the PEEK model. For the 3M™ True Definition Scanner (3M ESPE, St. Paul, Minnesota), which is based on pulsed blue light and active wavefront sampling,⁸ the deviations for $d_1$ were $(80\pm 14)\mu \mathrm{m}$, for $d_2$ $(62\pm 8)\mu \mathrm{m}$, and for $d_3$ we found $-(20\pm 7)\mu \mathrm{m}$. For the TRIOS^® 3 (3shape, Copenhagen, Denmark), a structured light scanner using confocal microscopy principle,⁹ the length deviations amounted to $(57\pm 18)\mu \mathrm{m}$, $(82\pm 16)\mu \mathrm{m}$, and $(124\pm 97)\mu \mathrm{m}$, respectively. The CS 3600 (Carestream, Atlanta, Georgia), a system applying four-color structural light,¹⁰ provided $(81\pm 22)\mu \mathrm{m}$, $(41\pm 20)\mu \mathrm{m}$, and $(40\pm 64)\mu \mathrm{m}$, respectively. Using the Medit i500 (Medit Corp., Seongbuk-gu, South Korea), based on triangulation, we found $-(93\pm 54)\mu \mathrm{m}$, $-(76\pm 33)\mu \mathrm{m}$, and $-(41\pm 247)\mu \mathrm{m}$. Finally, the Emerald™ (Planmeca Oy, Helsinki, Finland), a system applying tree-color projected pattern for triangulation,¹⁰ yielded $d_1=(14\pm 48)\mu \mathrm{m}$, $d_2=-(80\pm 54)\mu \mathrm{m}$, and $d_3=(110\pm 291)\mu \mathrm{m}$. The error bars corresponded to the standard deviations derived from the 10 measurements. Whereas the data were generally within a tenth of a millimeter, one recognizes some trends. The results of the Medit i500 scanner, for example, gave rise to values below the selected ground truth.

Considering the combination of the three selected distances, one can directly compare the performance in length measurements of the five IOS systems by means of the median values and the related variances. From the lowest to the highest median amplitudes, the Emerald™ gained $(2\pm 34)\mu \mathrm{m}$, the CS 3600 achieved $(58\pm 2)\mu \mathrm{m}$, the 3M™ True Definition Scanner scored $(59\pm 4)\mu \mathrm{m}$, the TRIOS^® 3 reached $(79\pm 4)\mu \mathrm{m}$, and the Medit i500 led to $-(98\pm 45)\mu \mathrm{m}$.

The average values, however, do not represent the full story. Therefore, the color-coded deviation fields are shown in Fig. 6. Although the Emerald™ IOS reproduced the three selected distances perfectly well, the strong color gradients indicate the challenging handling and a relatively weak reproducibility. Consequently, the CS 3600, the 3M™ True Definition Scanner, and the TRIOS^® 3 might be the better choice concerning reproducible and accurate measurements within the oral cavity of the patient. The results displayed in Fig. 6 also show that the Medit i500 is less precise than other IOS systems.

Fig. 6

Color-coded deviation field of the IOS derived from the difference to the selected ground truth, i.e., the high-resolution tomography experiments using nanotom^® m. The green color shows agreement, whereas the red and blue colors point the locations of positive and negative deviations.

Using the software GOM Inspect (GOM GmbH, Braunschweig, Germany), one finds a similar result. This software provided the following precision values: TRIOS^® 3—$35\mu \mathrm{m}$, CS 3600—$43\mu \mathrm{m}$, and 3M™ True Definition Scanner—$46\mu \mathrm{m}$. The other two systems yielded less precise data: Medit i500—$93\mu \mathrm{m}$ and Emerald™—$97\mu \mathrm{m}$. If only a single quadrant is considered, one finds another picture; therefore, the selection of the best available IOS system depends on the clinical case, the training of the dentist, and the convenience.

3.5.

Accuracy of Single-Crown Preparation

To analyze the single tooth preparation and the IOS-accuracy on a smaller length scale, Tooth 16 was considered. The mean deviation was determined in the same manner as for the full arch, i.e., $n=10$. The IOS-data of the single-tooth preparations demonstrated the expected performance of the systems included into the study. The accuracy is almost perfect, because the number of images to be stitched is limited. For the tooth stump (tooth 16), we found the following precision values: TRIOS^® 3—$(9\pm 5)\mu \mathrm{m}$, CS 3600—$(10\pm 5)\mu \mathrm{m}$, Medit i500—$(11\pm 1)\mu \mathrm{m}$, and 3M™ True Definition Scanner—$(12\pm 1)\mu \mathrm{m}$. Only one system yielded a significantly less precise dataset: Emerald™—$(47\pm 9)\mu \mathrm{m}$.

3.6.

Limitations Related to Clinical Conditions

It should be noted that the results of IOS cannot be fully judged from an in vitro study. The dentists have to treat also edentulous patients, to consider the difficulties in controlling the salivary flow, and to account for the presence of restorative and prosthetic works in the dental arch. The usability of the selected IOS device depends the learning curve of the dentist and the size of intraoral-camera size with respect to the space and accessibility of the oral cavity as well as the available and necessary scanning time to record the 3D surface data.