1. Introduction
Success in orthognathic surgery depends not only on the technical aspects of the operation but to a larger extent on the formulation of a precise surgical plan, consistency, and capability of achieving predictable and stable results [
1]. Technological advances have been developed in the last few decades to improve surgery: intraoperative navigation, imaging software, CAD-CAM technologies, three-dimensional (3D) printing, augmented reality, etc. These technologies help shorten preoperative surgical planning, and then this information is transferred to the operative setting through CAD/CAM splints, navigation, or surgical guides, reducing overall surgical time.
Virtual surgical planning has facilitated accurate diagnoses and detailed treatment planning through better visualization of 3D phenotypic changes [
2], allowing for a precise preoperative surgical plan using a computerized 3D environment. In addition, over the past 10 years, the development of 3D printed models and patient-specific guides has improved surgical planning as well as the transfer of the surgical plan into the operating room for better surgical results [
3].
Conventional planning of orthognathic surgery was conducted based on a radiographic cephalometric analysis and mock surgery on plaster-cast dental models mounted in a semi-adjustable articulator [
4]. Data were obtained from different studies (radiographs, models and articulators, face bow, etc.) and were interpreted before being able to develop a treatment plan [
5]. This cast model-based surgery is complex and arduous, and difficulties may be encountered when correcting occlusal cant and facial asymmetries.
The development of computer-aided surgical simulation represented a paradigm shift in surgical planning for patients with cranio-maxillofacial deformities. When using 3D planning, all the necessary information is provided in images that can be manipulated on a computer. Three-dimensional imaging-based planning systems enable the surgeon to establish necessary osteotomy planes preoperatively and assess different surgical scenarios [
1]. Since this is done virtually, surgery is performed in a more predictable way.
The surgical occlusion set-up is a vital step in virtual planning for orthognathic surgery [
2] using intraoral scanning and a full digital set-up. With 3D printing, this information is transferred to the operating room. The clinical applications of 3D printers in orthognathic surgery include the production of occlusal splints, osteotomy/cutting guides, repositioning guides, spacers, fixation plates/implants, and 3D printed models [
3]. Most studies have agreed that CAD/CAM occlusal splints with no modifications compared to traditionally manufactured occlusal splints provided a reliable substitute in orthognathic surgery [
3]
There are many commercially available software programs for virtual surgical planning and simulation [
6]. However, surgical planning may not necessarily reflect the actual surgical outcomes, and its accuracy and predictability must be established.
Most of the methods that have been proposed to assess the accuracy of the postoperative outcome versus the 3D surgical planning in the literature are based on the identification of cephalometric landmarks and computation of the differences between the planning model and the actual result [
7].
The purpose of this study was to evaluate predictability in orthognathic surgery and the accuracy of surgical outcomes using virtual surgical planning. The postoperative results of bimaxillary orthognathic patients were compared to the preoperative virtual surgical plan using different commercially available software.
3. Results
Table 1 shows the median and the 25th and 75th percentiles of the differences between the virtual planning and postoperative results for each cephalometric point in the three axes for all patients. The median distances are also displayed in
Figure 6.
The overall analysis showed good accuracy, with all median differences below 1 mm. The highest median differences were seen at point A and Pog in the anteroposterior direction, which were 0.835 mm and 0.780 mm, respectively.
The highest accuracy was observed at point B and Pog in the mediolateral direction (0.070 mm and 0.079 mm, respectively), followed by point B in the vertical direction (0.150 mm). The overall accuracy in the mediolateral direction was very good for all cephalometric points. The differences in the bone margin over the upper first molars were slightly higher compared to the mandibular first molars: 0.770 mm and 0.665 mm for the right and left upper first molars, and 0.310 mm and 0.580 mm for the right and left lower molars, respectively.
Then, we computed the absolute 3D distance for point A, point B, and Pog. The median and 25th and 75th percentiles for such values were calculated and are presented in
Table 2. The median 3D values for point A, point B, and Pog were 0.934 mm, 0.613 mm, and 1.034 mm, with the highest differences found at Pog.
Furthermore, we calculated the weight of each axis to the overall 3D distance using multivariate analysis for point A, point B, and Pog (
Figure 7). For point A, the axis that weighed the most was the
y axis (anteroposterior), with a coefficient of 0.885 (meaning that for each unit that the distance increased in the
y axis, the 3D distance increased by 0.885 mm), which was statistically significant (
p < 0.001). For point B, it was the
z axis (vertical) that had the most influence on the 3D distance, with a coefficient of 0.813, which was also statistically significant (
p < 0.001). In the case of Pog, the highest input was given by the
y axis as well (anteroposterior), with a coefficient of 0.726, which was statistically significant (
p < 0.001).
4. Discussion
Virtual surgical planning enables the precise analysis of a 3D model that represents the clinical situation and facilitates diagnosis and treatment planning [
18,
19]. Since Swennen [
19] initiated 3D cephalometric analysis and treatment planning, 3D virtual planning has replaced 2D cephalometric analysis.
Recent trends in orthognathic surgery have evolved to minimize the period of preoperative orthodontic treatment and to combine 3D technology in the process of surgical planning to improve accuracy [
2]. Several technological advances are being used widely to reduce preoperative planning and surgical operating time, including CAD/CAM technologies, 3D printing, intraoperative navigation, and augmented reality. Using 3D virtual models, all the procedures for diagnosis and surgical splint production can be simulated, intermediate assessment can be conducted, and the intermediate and final surgical splints can be produced. Therefore, errors encountered during laboratory procedures are minimal [
20].
Virtual surgical planning has facilitated accurate diagnoses and detailed treatment planning through better visualization of 3D phenotypic changes [
2]. In addition, virtual surgical planning allows for soft tissue evaluation and prediction. Available virtual planning software, such as Dolphin (Dolphin Imaging & Management Solutions, Chatsworth, CA, USA) or Maxilim (Medicim NV, Mechelen, Belgium), allow for an approximate prediction of the soft tissue response after maxillomandibular repositioning as a simple geometric transformation, not considering accurate mechanical modeling and tissue properties.
Multiple studies have analyzed computational models of soft tissue prediction in orthognathic surgery, using complex finite element analyses and algorithms.
Alcañiz et al. [
21] studied 10 patients who underwent orthognathic surgery and presented a simulation methodology for the planning of orthognathic surgical interventions, paying special attention to soft tissue simulation. The handling of several tissue couplings, i.e., soft tissue and bone, demonstrated a high complexity, which requires high-resolution meshes and long computation times to ensure accurate results. Furthermore, they found a tendency toward negative/cold errors, which was slightly higher on coarse meshes.
Furthermore, the use of 3D printed surgical guides and pre-bended fixation plates on 3D printed models based on virtual surgical planning increases the accuracy of the performed osteotomies and repositioning of the bone segments, shortening surgical time and improving the overall surgical results [
2,
3]. The use of 3D printers in orthognathic surgery is widely extended and includes the production of splints, surgical guides, pre-bend plates, patient-specific implants and plates, and 3D models. Compared to the traditional method, the digital-based occlusal splint provides high accuracy, reliability, and consistency, as well as improved quantitative control and efficiency [
3].
Numerous studies have demonstrated that 3D printing technologies help the clinician shorten the operative time, increase surgical safety, and improve the predictability of surgical outcomes [
3]. Hernandez-Alfaro et al. [
22] used an intraoral digital scanner to obtain surface images of the dental arches. After fusing the scans with the CBCT images of the patients for CAD/CAM intermediate splint generation, the accuracy and reliability of the protocol were assessed, showing an error below 1.5 mm between the virtual intermaxillary position and the intraoperative intermaxillary relationship and revealing a high overall accuracy.
Lin et al. [
3] performed a systematic review regarding applications of 3D printed technology in orthognathic surgery. They identified 78 articles where different applications were described. They found that most articles described the use of occlusal splints, osteotomy/cutting guides, positioning guides, spacers, fixations plates/implants, and 3D printed models. This technology was demonstrated to be beneficial to both the clinician and the patient, reducing the preoperative planning time and overall surgical time.
A surgical plan and surgical splint generated by a computer have the advantage that they can be shared with engineers from a Computer-Aided Design/Computer-Aided Manufacturing (CAD/CAM) center through the internet. Surgical splints can be fabricated by an additive manufacturing or milling technique and applied during surgery to transfer the virtual surgical planning to the operating field. For occlusal splints, previous studies have proposed the use of CAD/CAM occlusal splints as a reliable substitution to address the flaws of laboratory-based methods, including non-controllable errors, inter-laboratory differences, high costs, and time inefficiency [
3].
Nevertheless, differences still exist between the virtual planning and surgical execution due to the complexity of movements and transfer of data throughout splints from the virtual planning to the surgery. The reliability and predictability of virtual surgical planning must be tested.
In the literature, many studies have tried to evaluate the accuracy of virtual planning in orthognathic surgery using different protocols. These protocols and methodologies vary from linear/angular measurements, surface-to-surface differences, or virtual triangles. The most frequently used approach is linear and angular measurements, which rely on accurately identifying cephalometric landmarks and are prone to human error, especially when it must be done both on the preoperative and postoperative model [
23]. Color-coded distances maps are visual analytical tools that display the distance between two 3D surface meshes and are generally included in most software.
Tucker et al. [
16] evaluated the accuracy of virtual planning based on the surface distance differences between the plan and the actual outcome on 11 different regions of the maxilla and mandible. The method was done using a surface-to-surface best fit of the two virtual models, aligning the base of the skull and measuring the distance between the planned and actual outcome post-operatively. They found no statistically significant difference between the simulated and the actual surgical models in all 11 regions of interest, with distances of less than 0.5 mm except for the left lateral maxilla (0.536 mm).
In 2016, Zhang et al. [
24] analyzed the accuracy of virtual surgical planning and 3D printed surgical templates in orthognathic bimaxillary surgery in 30 patients. They studied the linear and angular differences in different cephalometric points. The mean linear difference was 0.81 mm and the overall mean angular difference was 0.95 degrees. Furthermore, they found that 3D printed surgical templates worked better on the maxilla than on the mandible (0.71 mm vs. 0.91 mm for the mean linear difference) and showed better control of the deviation from the midfacial plane (0.55 mm) than the FHP (0.92 mm) and the coronal plane (0.97 mm).
Stokbro et al. [
4] studied 30 patients who had undergone bimaxillary orthognathic surgery with and without segmentation and genioplasty, and found all mean linear differences for the maxilla, mandible, and chin segment to be within 0.5 mm. They also found that the mean precision, measured as the standard deviation, was smallest in the superoinferior direction, followed by the mediolateral deviation, and finally, anteroposterior. Precision was also the most accurate in the mandible, slightly less in the maxilla, and least in the chin segment, probably due to the mandible-first sequence.
Cevidanes et al. [
25] and Hajeer et al. [
26,
27] quantified 3D displacement using the
X,
Y, and
Z vectors of landmark displacement, similarly to our study. Kawamata et al. [
27] described methods referring to both linear and angular measures. However, these measures do not reflect what happens along the whole surface [
28]. For this reason, color-coded maps are needed to display information on what is happening along the whole surface model.
In their systematic review, Alkhayer et al. [
29] analyzed 12 papers regarding the accuracy analysis of virtual surgical planning. The accuracy values for the pitch, yaw, and roll (°) were (<2.75, <1.7, <1.1) for the maxilla, respectively, and (<2.75, <1.8, <1.1) for the mandible. They observed that the calculation of the linear and angular differences between the virtual plans and postoperative outcomes was the most frequented method used for accuracy assessment, and a difference of less than 2 mm/° was considered acceptable. They concluded that virtual planning appears to be more accurate, especially in terms of frontal symmetry, similar to our results.
On the other hand, Baan et al. [
7] used a different method for validation and accuracy of results in bimaxillary orthognathic surgery. They used a tool called “OrthoGnathicAnalyser” to analyze the postoperative outcomes with regard to the virtual planning. The main difference lies in the fact that no landmark identification is needed because the relevant translational and rotational movements of each jaw segment could be computed from the rotation matrices of the jaw segments during the registration process using the OrthoGnathicAnalyser tool. They found that the left/right translation showed the lowest absolute mean difference between the 3D planning and the surgical result for both the maxilla and mandible, at 0.49 mm, and 0.71 mm, respectively, whilst the vertical positioning of the maxilla and mandible suggested the lowest accuracy. In line with previous studies, the interocclusal wafer provided less intra-operative control in the vertical dimension.
The method we applied in this study can express more 3D shape information in comparison to those that were based solely on the calculations of linear and angular distances, offering valuable information in the three axes and three dimensions for the cephalometric points selected and visual data in the color-coded maps.
In the literature, differences of less than 2 mm between the virtual surgical planning and the actual postoperative results have been considered clinically acceptable by many authors [
4,
16,
30]. However, we believe that this statement must be taken cautiously since it highly depends on the quantity of movement (an advancement of 4 mm is not as clinically irrelevant as 8 mm).
In this study, overall good accuracy was found, with all differences for point A, point B, Pog, and both the upper and lower first molars being below 1 mm. The greatest values for the rest of the landmarks were found in the y (anteroposterior) axis at point A (0.835 mm), point B (0.480 mm), and Pogonion (0.780 mm), with overall worse accuracy in the anteroposterior direction for all the landmarks studied. On the contrary, distances found in the mediolateral direction were low (0.239 mm for point A, 0.070 mm for point B, and 0.079 mm in Pog). This means that the accuracy provided by intraoperative interocclusal splints is good in the mediolateral direction but poorer in the anteroposterior direction. The authors believe this can be explained because the thickness of the splint is considerable, and the segments may slide and be minimally displaced in the anteroposterior direction. Higher differences in the anteroposterior direction may also be explained because the magnitude of movement in this plane is usually the greatest when performing orthognathic bimaxillary surgery.
Vertical distances were also low (<0.3 mm), with the lowest at point B (0.150 mm) and slightly higher at point A (0.280 mm). The latter is manually controlled by the surgeon by measuring the distance from a bone-fixed point (a screw at the nasion in this series) and not by the splint, which may explain the higher difference.
The differences in the bone located over the upper first molars were found to be higher compared to those of the mandibular first molars. Again, the authors believe this difference may be explained by the fact that surgery is performed in a mandible-first sequence, and a final splint to position the maxilla is not used and it is positioned manually instead.
Regarding the 3D distances, the highest difference was found at Pog, with the lowest at point B, where the highest accuracy was found. When performing the multivariate analysis to examine the influence of each axis on the overall 3D distance, it was observed that for point A and Pog, it was the y axis (anteroposterior) that showed the greatest coefficient, meaning that for point A and Pog, the anteroposterior direction had the strongest influence on the overall 3D distance. Thus, it was in the anteroposterior direction that accuracy was the lowest. On the other hand, for point B, it was the vertical axis that had the highest effect on the 3D distance.
In conclusion, we found that in this study, the positioning of the mandible compared to the maxilla was more accurate. In addition, the lowest accuracy was found in the anteroposterior direction, whilst the highest accuracy was observed in the mediolateral direction. Further studies are needed to determine if future refinements of interocclusal splints may improve these results. Furthermore, in the vertical direction, all median differences were very accurate, especially in point B, where the distances were the lowest. Vertical distances were the highest at point A, which is intraoperatively controlled by the surgeon by osseous references and may explain these results. Differences in the mediolateral direction were also low for point A, point B, and Pog, reflecting a high accuracy when positioning both the maxilla and mandible in the midline, especially the mandible.
The authors believe that a mandible-first sequence allows the surgeon to overcome and correct problems of the centric relation and achieve a more predictable position. It is less prone to errors caused by the incorrect position of the mandible during CBCT examination. Special attention is paid to properly seat the condyles in the fossa during the stabilization of the proximal segment of the mandible and fixation since maxillary position can be inaccurate if the mandible is not correctly positioned. The authors’ preference is also not to use the final splint, and the maxilla is positioned according to correct dental interdigitation and occlusion, therefore eliminating interferences from the splint. However, a palatal splint is used in cases of segmental maxillary surgery. It is important to consider this when discussing these results.
Regarding the limitations of this study, it is a retrospective study and, as stated by many authors, the need for a manual selection of the cephalometric landmarks for analysis, which need to be identified multiple times on the models, is prone to human error and may be a source of inaccuracy in this study. Two solutions can be provided to overcome the landmark identification bias—either the fully automated identification of landmarks or the elimination of cephalometric landmark identification, as proposed by Baan et al. [
7]. Another source of limitation may be correlated to erroneous data on the surface mesh (for example, streak artifacts or surface roughness), which would have a marked effect on the measurements [
31]. In addition, postoperative CBCT was acquired without the use of the final splint. This may lead to a different occlusion than planned due to occlusal interferences, showing higher differences between the 3D planned and surgically achieved mandibular position.
As improvements to this study, measurements in more cephalometric and dental landmarks could be taken to conduct a more exhaustive and detailed analysis. Different assessment methods could also be used to overcome observer-dependent landmark identification errors. It would also be interesting to analyze these results considering the quantity of movement performed and display these results as percentages.