Recent studies have demonstrated comparable trueness of direct IOS digitization of full-arches to indirect digitization with the conventional impression being poured and digitized by a laboratory scanner. In vitro, a polymeric full-arch model was replicated by either indirect method using polyether impression material or by True Definition scanner (TD) [34]. The 3D linear shift between the two endpoints of the scan (between second molars) was 122 μm for TD scanner, and it was 174 μm for the indirect method. Using a similar approach for evaluation the trueness [12] the 3D linear shift between the two endpoints of the scan was 287 μm for iTero Element scanner (the version was not given), and it was 257 μm for the indirect method in vitro, but in vivo, the intraoral scan resulted in better trueness (305 μm) than the indirect method (517 μm). The method used in these papers resembles our identical point method in terms of that it measures the deviation of predefined points at the most distant region; that is why the values for the iTero Element were comparable to our result (246 μm). In another study [6], the trueness of indirect digitization of the full arch model was compared to seven IOS. The indirect digitization method resulted in significantly better trueness (16 μm) than any other IOSs; 49 μm for Trios 3, 58 μm for Carestream 3600, 89 μm for Medit i500, 63 μm for iTero Element 2, 48–90 μm for Omnicam models. This data was considerably lower than our results, probably due to the traditional surface comparison method. Similar results were found in another study [35]. A new method was developed to measure full-arch trueness in patients by bonding four spheres onto the teeth (around second molars and first premolars) surface with known, constant distances. After intraoral scan or indirect digitization, the local best fit was applied on the spheres and the surface deviation was calculated. The indirect method yielded the best trueness (15 μm) followed by TD (23 μm), Trios (37 μm), CEREC Omnicam (214 μm).
Our result employed in a dentate human cadaver arch showed no increase in deviation as a function of distance for physical impressions, which is consistent with a previous study [24]. The authors found little increase (from 15 μm to 67 μm) in deviation from the theoretical scan origin (left second upper molar) to the end of scanning (contralateral molar) for physical impression, but a more substantial increase was observed for IOS. The values ranged from 15 μm to 205 μm for the CS 3600 (with a mean of 119 μm) and from 10 μm to 227 μm for the Trios 3 (with a mean of 184 μm). They were somewhat different from the present results (365 μm and 156 μm respectively) for this specific scanner, but we used cadaver instead of a stone model, and the identical point method was applied instead of the surface comparison method which makes a considerable difference (Fig. 7.). Similarly to the present study, the trueness averaged across the full arch for Planscan found to be 661 μm measured by the identical point method with the consideration of scanning origin, and this was three times more than with the full arch surface comparison (197 μm) [27]. Thus, the difference between the above study and the present one could be explained by the differences between method, i.e., the surface versus identical point deviation.
If the alignment of the actual scan to the reference model is done at the tooth of scanning origin, then the deviation grows as distance increases from the origin [24, 27]. The surface alignment at one tooth involves fewer pixels; therefore, it could be speculated that the local best-fit algorithm limited to one tooth has less accuracy compared to full arch alignment. It may cause an artifact of a constant linear increase in deviation from the starting point. However, the increase was neither linear nor constant. In this study, a physical impression captured by a laboratory scanner was also analyzed by fixing the experimental model to the control model at what would be the scan origin during alignment. The present result confirms that accumulated deviation is not an artifact but instead is caused by the stitching mechanism, which is specially related to IOS, but not to the laboratory scanners [16, 36]. This problem is likely the main reason that physical impressions continue to have higher trueness for full arches versus direct intraoral scanning [6, 35, 37, 38]. The tooth wise kinetics of the deviation suggests that the actual deviation at a specific tooth could be much more than the average full arch deviation. Thus initiation of scanning should be done as close to the restoration site as possible to minimize model distortion.
The novel method could also perform the measurements separately along three different axes according to the dental planes on the arch (occlusal, sagittal, and transversal). Scans were started on the occlusal, resulting in axis Z showing the distance between the occlusal surface and the scanner. The investigated IOSs use different optical mechanisms in order to determine depth information [39,40,41]. The deviation measured along the z-axis can indicate the differences between various types of hardware. This sets an internal coordinate system for stitching further images together. The hardest step is defining the depth parameter, which differentiates a 3D scanner from a 2D camera. Similarly to another study [34], we found that axis Z is the most inaccurate dimension.
Previously, the raw data used in this study was analyzed by the surface comparison method and revealed no statistical difference among scanners in terms of trueness and precision for full-arch imaging with the exception of the Planscan, which showed significantly higher deviation [13]. The authors concluded that it might require a higher sample size to reveal statistical differences. Our novel method using identical points with alignment at the scanning origin on the compared objects was capable of distinguishing between the trueness of scanners investigated using the same raw data. That is similar to a previous study showing this novel method illustrated a significant effect of the scanning patterns on trueness when compared to the commonly used traditional best-fit surface comparison method [27]. This finding confirms the increased sensitivity and statistical power of the new method (Fig. 6.). The higher specificity could prove increasingly valuable as the newer generation of IOS is more and more accurate [4, 6]. In this study, newer versions from the same manufacturer produced better trueness (Element 1 vs. Element 2 and Planscan vs. Emerald) similarly to other studies [6, 24]. Apart from hardware research, manufacturers make significant investments in performing software R&D to continually improve the accuracy of IOS (both trueness and precision), which can also be experienced by users after a version update [27, 42]. Furthermore, many factors can influence accuracies such as scanning pattern [5, 43, 44] or substrate reflectance [29, 30] software settings and upgraded versions [6, 42]. Accuracy studies must be reevaluated from time to time due to this constant development.
According to our results and others [6, 13, 28], it is not surprising that precision and trueness had a positive correlation. The precision is statically equal to the standard deviation of the trueness; therefore, a non-significant difference in trueness between scanners does not necessarily mean that one scanner is better than the other. High standard deviation (low precision) could mask the differences in trueness values. In this study, the best example is that the difference between CS 3600 and the physical impression was not significant due to the high standard deviance (low precision) of the CS 3600, but we cannot conclude that this is a perfect scanner. Precision also has a high clinical relevance because a restoration needs to fit in every case regardless of which scan, usually only one, is it created from.
Determining the precision of the IOS from in vivo clinical assessment is a fairly simple method, as the 3D models generated through the repeated, consecutive scanning of the dental arch can easily be compared with appropriate analytical software that can superimpose these models with each other [6, 12, 13, 24, 44,45,46]. However, direct evaluation of the trueness value is not possible in vivo, as the reference model can only be created using indirect extraoral methods with physical or optical tracking devices. A dental arch constrained within a human’s head cannot be transferred into these devices. Therefore, trueness values are usually assessed in vitro, on the manufactured physical samples [6, 24, 44,45,46]. Recently, it has been demonstrated that the optical properties of various materials influence the accuracy of IOS, which is likely also true of mucosal tissues and other restorative and optical impression material combinations [29, 30]. Therefore, this study represents a further step toward a more clinically relevant analysis by using a cadaver with representative tissues compared to artificial materials and substrates.
However, other clinical conditions were not simulated in our study. The substantial differences in the deviation between intraoral scan and extraoral scanning using the same protocol were explained by the patient movement, limited intraoral space, intraoral humidity, and saliva flow in another study [36]. The cadaver tissue in our study was removed from the body make the scanning easier for the operator, and it was free of saliva, but it kept wet by water. In a study [47], where artificial saliva was applied on the model surface, the intraoral scans were distorted considerably compared to other in vitro studies [6, 24], probably due to the bubbles formed on the surface. Another limit of our method could be that although the mean surface deviation for the local best fit applied at a single tooth was very low (mean = 18 μm, standard deviation = 13 μm, n = 574), in some case (e.g., a Planscan case) it went up to 146 μm. This deviation could be considered the error in the automatic copy of the points from the reference scan to the test scan. However, this deviation is much less than the differences between the deviation measured on the surface using non-identical points (conventional method), and the deviation measured between identical points regardless of full arch alignment or alignment at scan origin was used. The manual selection of these points may also involve some uncertainty in the novel method. The points were selected on the cusps on the molars and the premolars and on the incisal edge on the anterior. These areas are special in terms of having high deflection. If we had selected some points on the smooth lateral surfaces on the tooth, we might have been closer deviation to the conventional surface comparison method. Contrary to a pre-designed geometrical object (i.e., an engineered object such as an implant scan body), the teeth are amorphous. Consequently, there are no special points on it. Another limitation that this method still suffers from the same problem as the conventional method, such as there is no physical limit of superimposition. The test scan could go in and out of the reference scan resulting in a negative (under the surface) and positive (above the surface) deviation values. In reality, it is not possible as the designed framework cannot sink into the virtual model (similarly to the physical stone model). Recently it was demonstrated [48] that this kind of virtual superimposition method may underestimate the more real physical alignment. This problem only partly eliminated by the alignment at the scan origin.