Accuracy of virtual planning in orthognathic surgery: a systematic review

Background The elaboration of a precise pre-surgical plan is essential during surgical treatment of dentofacial deformities. The aim of this study was to evaluate the accuracy of computer-aided simulation compared with the actual surgical outcome, following orthognathic surgery reported in clinical trials. Methods Our search was performed in PubMed, EMBASE, Cochrane Library and SciELO for articles published in the last decade. A total of 392 articles identified were assessed independently and in a blinded manner using eligibility criteria, out of which only twelve articles were selected for inclusion in our research. Data were presented using intra-class correlation coefficient, and linear and angular differences in three planes. Results The comparison of the accuracy analyses of the examined method has shown an average translation (< 2 mm) in the maxilla and also in the mandible (in three planes). The accuracy values for pitch, yaw, and roll (°) were (< 2.75, < 1.7 and < 1.1) for the maxilla, respectively, and (< 2.75, < 1.8, < 1.1) for the mandible. Cone-beam computed tomography (CBCT) with intra-oral scans of the dental casts is the most used imaging protocols for virtual orthognathic planning. Furthermore, calculation of the linear and angular differences between the virtual plan and postoperative outcomes was the most frequented method used for accuracy assessment (10 out of 12 studies) and a difference less than 2 mm/° was considered acceptable and accurate. When comparing this technique with the classical planning, virtual planning appears to be more accurate, especially in terms of frontal symmetry. Conclusion Virtual planning seems to be an accurate and reproducible method for orthognathic treatment planning. However, more clinical trials are needed to clearly determine the accuracy and validation of the virtual planning in orthognathic surgery. Supplementary Information The online version contains supplementary material available at 10.1186/s13005-020-00250-2.


Background
Two-dimensional (2D) radiographs and manual model surgery are essential parts of the preoperative planning for orthognathic surgery. However, this approach has its limitations, especially in the case of patients with major facial deformity or asymmetry [1], as 2D cephalometric images cannot provide full information about the 3D structures.
When conventional 2D surgical plans are executed, unexpected problems, such as a bony collision in the ramus area, the discrepancy in pitch, roll and yaw rotation, midline difference and chin inadequacy may occur [2].
When two-jaw surgery is performed, an inter-occlusal splint is fabricated to work as an intermediate guide for repositioning the maxilla relative to the intact mandible [3]. Any variation between the plan and the plaster model surgery could lead to a poorly fabricated wafer, which in turn could lead to unexpected (and often undesirable) results, regardless of how skillfully and carefully the surgery is performed [3].
These examples illustrate that the elaboration of a precise pre-surgical plan is of utmost importance when it comes to correcting dentofacial deformities.
In particular, the visualization of skeletal complexities within an asymmetric dentofacial deformity has been greatly enhanced through three dimensional (3D) modeling, which can demonstrate the extent of yaw rotation in the maxilla and mandible, occlusal plane canting and differential length of a mandibular body or the ramus [1,10,11]. The 3D simulation method has been accepted for planning in orthognathic surgery and led to significant improvements in surgical outcomes [1,9,12]. Intraoperative efficiency has also improved with the fabrication of the templates and jigs to reproduce gaps or spacing between the osteotomies depicted in the virtual plan. These jigs may reinforce intraoperative accuracy of the clinical movement of the virtual plan and aid in orienting and positioning bony segments [10,[13][14][15][16][17][18]. Thus, the aim of this systematic review is to assess the accuracy of computer-aided planning in orthognathic surgery.

Methods
A systematic search was conducted of electronic and printed articles that have been published in the period (2007-2017) on virtual planning for orthognathic surgery and in the English language. The databases used were PubMed, EMBASE, Cochrane Library and SciELO.
Keywords and Boolean operators ('OR' and 'AND') were used to join the terms related to orthognathic surgery and virtual planning. The same search strategy was applied to the Cochrane Library since this also uses MeSH terms.

Search strategy
For the search of EMBASE, the entry terms 'orthognathic surgery' AND 'virtual planning surgery' were used to carry out a specific search.
Health sciences descriptors were used to search the SciELO databases, 'orthognathic surgery' AND 'virtual planning' were performed.

Eligibility of the studies
The eligibility of the studies was determined by the author (A.A.), observing the following criteria: (1) the main theme of the paper had to focus on virtual planning for orthognathic surgery; (2) the study had to be original and interventional; (3) the surgical procedure had to be virtually planned with a virtual surgical splint; (4) accuracy measures had to be presented for the surgical procedure; (5) the sample size of the trial had to be ≥10. The latter criterion was determined somewhat arbitrarily, as a reasonable minimum, given the small sample sizes of these studies in general.

Main search
Three hundred and sixty-seven articles were found in PubMed, 84 in EMBASE, 7 in Cochrane Library and 16 in SciELO. Duplicate papers were removed, leaving a total of 392 possible studies, that have been read and 31 of these were chosen for full-text reading (Fig. 1).

Eligibility assessment
As part of the eligibility assessment, 31 studies were read in full. At the end of this analysis, only twelve papers were included in the sample for our systematic review. The other 19 studies were excluded for the following reasons: virtual surgical planning for orthognathic surgery was not the main focus of the paper [19],the paper was not an intervention study [17], or it was not original [5,20,21], the surgical procedure did not involve a computer-assisted virtual surgical splint [22][23][24], the accuracy measurements for the surgical procedure were not provided [25][26][27][28][29] and the sample size was less than 10 [16,22,[30][31][32][33].

Quality assessment of the included articles
The quality of the papers was assessed using an adaptation of the bias analysis proposed by Clementini and colleagues [34]. The criteria were the presence or absence of the following: sample randomization, blind assessment, statistical analysis, defined inclusion and exclusion criteria and reporting of follow-up. With respect to the risk of bias for each analyzed study, papers containing all the above items were considered low risk, studies lacking one or two items were missing were deemed medium risk, and investigations that lacked three or more items were considered high risk.

Results
Descriptive data of the included studies (sample size, age, gender and type of facial deformity) are presented in (Table 1).
The imaging protocols and the software used for surgical planning varied substantially among the studies, These variations are shown in ( Table 2).
The included studies also varied in the type of surgical plan and virtual splints, as well as in the method used for the assessment of accuracy. These variations we summarized in (Table 3).
The actual accuracy values are presented in detail in Additional file 1 (Table S1).
Finally, the papers included in this review were assessed as being medium quality, since the risk of bias was considered medium in ten studies of the twelve. The risk of bias assessement for the included studies are presented in (Table 4).

Discussion
The use of computerized methods for diagnosis and treatment planning in orthodontics and orthognathic surgery has evolved substantially [42], which is confirmed by the 392 papers on this topic that have appeared in the major databases in the period (2007-2017).
Hsu and colleagues reported that computer-aided techniques enable the accurate correction of maxillary malformations with yaw deviation, alignment of proximal and distal segments and restoration of mandibular symmetry [6].
Lin and co-workers concluded that virtual orthognathic planning yields aesthetically favorable results, a high level of patient satisfaction, accurate translation of the treatment plan and thus making the operation itself easier and safer [20,44].
The analyzed studies used both the CT and CBCT imaging modalities (two of them worked with both). Better identification of soft tissue and less image distortion where metallic elements are present are obvious advantages of CT over CBCT, while disadvantages include image quality, the supine position of the patient during the test (especially because of mandibular retrusion) and larger radiation doses [45][46][47]. Mandibular retrusion in the supine position during CT image capture was attenuated using central occlusal registry [6,42]. The major disadvantage of CBCT is the occasional appearance of metal artifacts, but this is diminished by scanning the plaster casts [37,40,42], intraoral scanning of the dental arches [30,37], scanning occlusion with reference points [6,7] or by a triple scan procedure [39,41].
Thus, the fusion of facial CT images and dental arch scans is important in computer-aided planning and it is more accurate when reference points are reproducible for both modalities [8].

Evaluation of the accuracy of the virtual planning methods used in Orthognathic surgery
One of the most frequently used methods to evaluate the accuracy of virtual planning is the use of the mean error differences in superimposition between the virtual plan and the postoperative outcomes. Baan and colleagues used this technique to assess the degree of correspondence between the planned and performed positions. They also assessed the repeatability of the surgical procedure performed by different surgeons, and noticed that the discrepancy between the 3D planning and the postoperative results was the greatest regarding the vertical positioning of the maxilla and mandible, suggesting a less accurate intra-operative vertical control of virtual planning [39].
On the other hand, Franz and co-workers suggested that the use of the mean error as an only endpoint to measure the degree of accuracy can limit the generalizability of the studies. They also suggested that the confidence interval does not describe the real range of the method error but defines only the range of values that the mean error can assume from a statistical perspective [23].
Ho and colleagues calculated the accuracy of computer-aided orthognathic planning by evaluating the root-mean square difference (RMSD) of the 3D simulation and postsurgical CBCT images and found that the errors were acceptable, with RMSD (0.63 ± 0.25) mm for the maxilla and (0.85 ± 0.41) mm for the mandible [1].
De Riu and co-workers also suggested that the simple superimposition of the simulation and the cephalometric results is an unsatisfactory method, as it fails to consider the magnitude of the surgical manipulation leading to an error of a given magnitude. For instance, a slight CT computed tomography, CBCT cone beam computed tomography, 3D three dimensional, NA data not provided by the authors, CBCT cone beam computed tomography Bimaxillary surgery Inter-occlusal wafer was milled based on the virtual planning.
Intraclass correlation coefficient (ICC) was calculated to evaluate the interobserver and intra-observer variability for the rotational and translational measurements of the maxilla and mandible. Bimaxillary surgery planning through maxilla Occlusal splint Linear and angular distance between reference points on the x (pitch), y (roll), and z (yaw) planes, 3D imaging (surface-best-fit), 3ds Max (Autodesk Inc., USA) Zinser et al. 2013 Germany [42] Clinical and 3D analysis Bimaxillary surgery (28), planning through maxilla Occlusal splint, Bone splint (maxilla and mandibular condyle) Linear distance between the reference points for the x, y, and z planes in 3D imaging (voxel-based) Centenero and Hernández-Alfaro .2012, Spain [43] (15) Bimaxillary surgery, (1) Single maxillary surgery Occlusal splint Intra-class correlation coefficient (ICC) of the reference lines and angles; concordance level 3D imaging (NA) 3D three-dimensional, NA no information provided by the authors, FHP Frankfort horizontal plane, CP coronal plane, MFP midfacial plane, N nasion point positional error can be completely acceptable for large manipulations, but would be unacceptable when the manipulation takes place at a small scale and thus needs to be extremely precise [35]. The accuracy of the translation of the maxilla with computer-assisted planning for orthognathic surgery was < 1 mm in the study of Hsu and colleagues, indicating that this type of planning is accurate for the maxilla [6].
The Stokbro group found that the mean linear differences for the maxilla, mandible and the chin segment in all three planes were within 0.5 mm, while the mean precision, measured as the standard deviation, had the smallest deviation superoinferiorly, followed closely by mediolateral deviation, and finally the largest deviation was found anteroposteriorly [38].
De Riu and co-workers found that virtual surgical planning presented a high degree of accuracy for most of the parameters assessed, with an average error of 1.98 mm for linear measurements and 1.19°for angular measurements. At the same time, they observed significant differences between planned and achieved anterior facial height (p = 0.033). Without genioplasty, no significant difference was observed (U test; p = 0.45). The authors concluded that the problem was caused by the virtual model of the soft tissues, which made it difficult to manage the vertical dimension [35].
It has been also shown in the study of Baan and colleagues that the right /left translation has the lowest absolute mean difference between the 3D planning and the surgical results for both the maxilla and mandible (0.49 mm and 0.71 mm, respectively). Furthermore, they noticed that in 7 out of 10 cases, the maxilla was positioned more posteriorly than in the 3D plan, with an absolute mean difference of 1.41 mm. The same tendency was found in the sagittal position of the mandible, where in 8 out of 10 cases the mandible was positioned more posteriorly than planned with absolute mean difference of 1.17 mm [39]. Lee and colleagues suggested that the condylar position might have been changed during surgery by muscle tone and gravity as the patient was placed in the supine position, which affects the optimal condylar seating [48]. Stokbro et al. (2016) are of the same opinion about this issue.
The clinical analysis of Sun and colleagues, of the twenty three patients, using the OrthoGnathic Analyser, showed an adequate position of the maxilla and mandible in the left/right direction with a deviation of 0.32 mm and 0.75 mm, respectively. It was found that the maxilla had a lower RMSD (0.6 mm) than did the mandible (0.85 mm) [19]. Zhang et al. showed that the overall mean linear difference was (0.81 mm), and the overall mean angular difference was (0.95°) [40], which was an improvement as compared with their previous study, as a result of surgical experience, 3D printing technology, and improvement of the elasticity modulus of 3D-printed surgical templates [49].
On the other hand, Baan et al. observed that the accuracy of the pitch of the maxilla (2.72°) and the mandible (2.75°) showed the highest discrepancy between the 3D plans and the actual postoperative status. This variance could be the result of bone conflict between the pterygoid plate and the osteotimized maxilla [39]. Stokbro et al. came to similar conclusions [38].

Comparison of the accuracy between classical and virtual planning methods
A lot of studies compared computer-assisted planning with classical planning and found favorable accuracy results in all bony segments for computer-aided planning [36,41,42,50]. Ziesner and colleagues reported that the Defined inclusion and exclusion criteria mandibular condyle maintained a central position in the temporomandibular joint, which did not occur when classic planning was used [42]. Hsu et al. compared the two types of interventions in the chin and found highly favorable accuracy results for computer-aided planning in this bone segment, with the largest difference recorded for translation in the sagittal plane (2.5 mm) and rotation pitch (3.68°). They explained these differences by the fact that classical planning does not use surgical splints; surgeons are guided by their experience, some internal reference points and the chin plate [6].
Ritto and colleagues reported on a similar level of precision in all evaluated regions when assessing the vertical positioning of the maxilla, but virtual surgical planning (VSP) was more accurate for the anteroposterior position of the maxilla. As for transverse positioning, conventional model surgery (CMS) yielded higher precision only for the upper midline position. However, there was no statistically significant difference between the groups, and the mean imprecision was also < 2 mm for all regions evaluated [36].

Risk of Bias assessment
The papers included in this systematic review were classified as medium quality, since the risk of bias was considered medium in ten studies [1, 6, 7, 35-37, 39, 41-43], that is, the majority.

Conclusions
In conclusion, the results of this systematic review suggest that computer-aided planning is an accurate method for orthognathic surgery of the maxilla and the mandible.
We found that CBCT with intraoral scan of the dental cast is the most frequently used method for virtual orthognathic planning, and SimPlant (Materialise, Leuven, Belgium) and Dolphin (Dolphin Imaging, USA) are the most widely used software.
Despite its limitations, the calculation of the linear and angular differences between the virtual plan and the postoperative status is still the most frequently used method for accuracy assessment, and differences < 2 mm/°are considered acceptable.
Additional file 1: Table S1. Virtual planning accuracy of the included studies.