Accuracy of Soft Tissue Prediction in Skeletal Type III Relationship Using a Computer-Aided Three-Dimensional Surgical Simulation Planning Program

Abstract

Using a computer-aided, three-dimensional surgical simulation planning program, soft-tissue planning can help achieve adequate facial esthetics and patient satisfaction after orthognathic surgery. This study aimed to assess the Simplant O&O software’s soft tissue prediction accuracy. Fourteen skeletal type III patients who underwent orthognathic surgery by the same surgeons were included in this prospective study, and they were separated into two groups: the one-jaw (n = 5) and two-jaw (n = 9) groups. The software was used to analyze the preoperative (T₀) and 4-month postoperative computed tomography data (T₁), as well as intraoral scans. Data from cone-beam computed tomography and stereolithography from a scanned dental cast were used to reconstruct a composite skull model. Based on the presurgical CT data, the program generated a predicted soft tissue image (T_P), which was then superimposed on the T₁. The distances between seven T₁ and T_P landmarks were measured and evaluated using a one-sample t-test. In the one-jaw group, the mean error for all linear measurements was 1.73 ± 1.14 mm, whereas the mean error of the two-jaw group was 1.03 ± 0.83 mm, and both measurements were within clinically acceptable limits. Pronasele had the best correlation (mean error of 0.63 ± 0.45 mm) while soft tissue pogonion and soft tissue point B had the worst correlations (mean error of 2.87 ± 2.22 mm and 1.31 ± 0.98 mm, respectively). Even though there were some limitations, it was possible to conclude that the ability to accurately predict soft tissue changes using Simplant O&O for skeletal type III patients makes it adequate for use in clinical practice.

1. Introduction

Orthognathic surgery combined with orthodontics is the only option for patients whose orthodontic problems are so complex that neither growth modifications nor camouflage can correct their malformations [1]. Because of maxilla, mandible, and chin repositioning, orthognathic surgery can significantly improve facial esthetics and proportions. These procedures also aim to achieve balanced occlusion and proper function. However, seventy percent of patients cite facial esthetics as their primary motivation, highlighting the significance of soft tissue [2]. Soft tissue treatment plans must be carefully considered to achieve adequate facial esthetics and patient satisfaction, and the placement of skeletal parts will be dictated [3]. As a result, orthodontists and surgeons encounter significant challenges in treatment planning and predicting soft tissue outcomes.

In the past, clinicians have traditionally used acetate tracing paper and conventional two-dimensional imaging. These methods provide limited information due to the distortion caused by projecting complex three-dimensional structures onto two-dimensional surfaces, resulting in inevitable magnification errors, especially when dealing with patients who have significant facial asymmetry [4,5,6]. Computer-aided surgical simulation utilizing 3D images from multi-slice computed tomography or cone beam computed tomography (CBCT) used for surgery planning has gained popularity in recent years, and they have the potential to solve the aforementioned problems, assisting clinicians in predicting more accurate surgical results and being used in patient communications [7]. Many commercial 3D planning programs have been released on the market, each with its own set of details and physical models. Simplant O&O is a three-dimensional digital software planning package that can be used for treatment planning, surgical prediction, and patient education. This program has the advantage of being able to simply build and superimpose STL files generated from CBCT data, which are convenient and reduce the superimposition discrepancies caused by using other programs. The accuracy of Simplant O&O’s soft tissue prediction is still unknown in the current literature. Therefore, the purpose of this study was to evaluate the soft tissue prediction accuracy of Simplant O&O software (Materialize Dental, Leuven, Belgium) by comparing its prediction results with actual postoperative soft tissues after surgery.

2. Materials and Methods

2.1. Subjects

Fourteen patients who required orthognathic surgery from January 2021 to November 2022 were included in this study. All patients had undergone surgery at the Department of Orthodontics and the Department of Oral and Maxillofacial Surgery, Faculty of Dentistry, Mahidol University. The Mahidol University Institutional Review Board (MU-IRB) approved the study (COA.No.MU-DT/PY-IRB 2021/010.0102), and all participants were asked to sign an informed consent agreement. The patients were selected based on the following inclusion criteria: (1) skeletal class III dentofacial deformities, (2) no congenital craniofacial deformities or trauma and no history of head and neck surgery, and (3) availability of high-quality CBCT and intraoral scans before surgery. All surgical procedures were performed by a single skilled surgeon. Depending on the operation they underwent, the participants were divided into two groups: the one-jaw and two-jaw groups.

2.2. Data Collection and Analysis

All patients underwent CBCT and intraoral scanning on the same day using the same device for both the preoperative (T₀) and postoperative (T₁) images (the latter of which were taken at least 4 months after surgery). The CBCTs were obtained using a Veraview X800 (J. Morita, Kyoto, Japan) and saved as .dcm files, whereas the dental scans were obtained using a 3shape TRIOS intraoral scanner and saved as .STL files. Both sets of CBCT data were separately imported into the planning software, Simplant O&O (Materialise Dental NV, Leuven, Belgium). The 3D virtual models and soft tissue volumes were reconstructed and oriented using a Frankfort horizontal plane (FHP; orbital—middle point of porion) parallel to the floor. Since all patients had fixed appliances in place at the time the CBCTs were taken, metal artifacts from the metal brackets were scattered all over the tooth surfaces, obscuring the details of occlusion. An intraoral scan was then imported and superimposed over the original blurred CT dentition using occlusal landmarks and voxel-based matching methods to obtain a clear vision of the dentition (Figure 1). For the T₀ images, virtual skulls were separated into independent segments according to the procedures that the patient underwent, and these were prepared for surgical movement. The software’s automatic segmentation will distinguish a specific component and eliminate all other structures for better visualization and analysis, and this is less time-consuming compared to manual segmentation [8]. The preoperatively virtually osteotomized bone segments were moved the same amount as the actual surgical movement in both the anteroposterior and vertical directions. The actual surgical movements for each patient were obtained using the voxel-based matching method after importing and superimposing the virtual skull and soft tissues from the T₁ onto the T₀ project (Figure 2). The native Simplant algorithm was used to generate a 3D soft tissue prediction image (T_P) (Figure 3). The distances between seven landmarks (

Table 1. Soft tissue landmarks.

Point	Definition
Pronasale (P)	Most anterior midpoint of the nasal tip
Subnasale (Sn)	The midpoint on the nasolabial soft tissue contour between the columella crest and the upper lip
Soft tissue point A (A’)	Point of greatest concavity on the contour of the upper lip
Stomion superior (Stou)	Midpoint of the lower border of the upper lip
Stomion inferior (Stoi)	Midpoint of the upper border of the lower lip
Soft tissue point B (B’)	Point of greatest concavity on the contour of the lower lip
Pogonion (Pg)	The most anterior midpoint of the chin

) of the T₁ and the T_P were measured. All of these points were standard anthropometric or cephalometric landmarks [9,10,11]. All distance testing was completed by a single investigator to reduce inaccuracies between viewers. All measurements were taken twice, two weeks apart.

2.3. Statistical Analysis

The result data were arranged using Microsoft Excel (Version 14.0.1, Redmond, WA, USA), entered into SPSS (Version 26.0, Chicago, IL, USA), and subsequently analyzed. The intraclass correlation coefficient was calculated to evaluate the intra-observer variability for the measurements. In each group, the mean and standard deviation for the differences measured at each landmark were calculated for both absolute and vector values. The discrepancies of the individual landmarks between the simulated and post-surgical images were assessed and calculated by a one-sample t-test. The Pearson correlation coefficient was used to assess the relationship between the surgical movement distances and the soft tissue landmark discrepancies. A p value of 0.05 was chosen as the statistical significance level.

To interpret the prediction accuracy results, a linear difference between the prediction and the postoperative outcome of less than 2 mm was considered clinically insignificant as this magnitude of difference has been proposed as the threshold for a visually perceptible facial difference [12]. The accuracy percentage was calculated by dividing the number of patients with a discrepancy of less than 2 mm by the total number of samples.

3. Results

Fourteen patients (

Table 2. The study sample’s characteristics.

ID	Sex	Age	Procedure	Surgical Movement (Sagittal)
				Maxilla (mm)	Mandible (mm)
One-jaw group
1	M	24	BSSO	N/A	−6.55
2	M	27	BSSO	N/A	−13.11
3	M	29	BSSO	N/A	−5.6
4	F	37	BSSO	N/A	−6.14
5	F	30	BSSO + genioplasty	N/A	−0.95
Two-jaw group
6	M	27	LFI + BSSO + genioplasty	1.63	−8.61
7	M	28	LFI + BSSO	0.76	−8.7
8	M	25	LFI + BSSO	1.14	−2.6
9	F	23	LFI + BSSO	0.96	−6.34
10	F	35	LFI + BSSO	4.08	−2.14
11	F	32	LFI + BSSO	3.76	−1.5
12	F	26	LFI + BSSO + genioplasty	1.31	−3.43
13	F	21	LFI + BSSO	0.68	−5.99
14	F	24	LFI + BSSO + genioplasty	1.99	−3.03

M, male; F, female; positive value, anterior movement; negative value, posterior movement.

) consisting of six males and eight females were included in this study. The mean age was 27.7 years old (ranging from 21 to 37 years old). All patients had skeletal type III malocclusion and were classified into two groups based on the procedures they underwent. Patients in the one-jaw group were given BSSO alone (n = 5) and those in the two-jaw group underwent Le Fort I osteotomies combined with mandibular setbacks using BSSO (n = 9). Genioplasty was performed on a few of the individuals in both groups. On average, the T₁ data were collected 6.5 ± 2.5 months (range of 4 to 13 months) after surgery. The mean distance of the sagittal mandibular setback in the one-jaw group was 6.5 ± 3.3 mm (range of 1.0 to 13.1 mm) while the average surgical movements in the sagittal plane of the two-jaw group were as follows: 1.8 ± 1.3 mm for maxilla advancement (range of 0.7 to 4.1 mm) and 4.5 ± 3.1 mm for mandibular setback (range of 1.5 to 8.7 mm). The intraexaminer agreement was high, with an intraclass correlation coefficient greater than 0.9.

3.1. Absolute Value

For the one-jaw group, absolute values were used to calculate the linear measurement differences between the prediction images (T_P) and the postoperative result (T₁). With a test value of zero, all landmarks showed significant differences (

Table 3. Discrepancies between the predicted soft tissue images and the actual postoperative soft tissue images for the one-jaw group.

Point	p-Value	Mean Difference	SD	95% Confidence Interval of the Difference		Accuracy Rate, %
				Lower	Upper
P	0.006 *	0.479	0.201	0.229	0.729	100
Sn	0.007 *	1.633	0.721	0.738	2.529	60
A’	0.030 *	1.115	0.760	0.172	2.058	80
Stou	0.046 *	1.435	1.123	0.041	2.830	60
Stoi	0.020 *	2.733	1.621	0.720	4.746	60
B’	0.037 *	1.813	1.315	0.180	3.446	40
Pg	0.044 *	2.868	2.215	0.118	5.619	40

*, statistical significance at p = 0.05; SD, standard deviation.

). The mean error for all linear measurements was 1.73 ± 1.14 mm. Pronasele had the highest correlation between the T_P and T₁ images (mean error of 0.48 ± 0.21 mm) while soft tissue pogonion was the least accurate (mean error of 2.87 ± 2.22 mm). Table 3 also shows that the predictions for two landmarks (soft tissue point B and pogonion) had achieved accuracies of approximately 40%, with an average absolute error of less than 2 mm, while predictions for the other five had achieved accuracies greater than 60%.

The mismatch between the postoperative result (T₁) and the prediction images (T_P) in the two-jaw group is shown in

Table 4. Discrepancies between the predicted soft tissue images and the actual postoperative soft tissue images for the two-jaw group.

Point	p-Value	Mean Difference	SD	95% Confidence Interval of the Difference		Accuracy Rate, %
				Lower	Upper
P	0.004 *	0.719	0.534	0.309	1.130	100
Sn	0.003 *	1.184	0.839	0.539	1.830	77.8
A’	0.017 *	0.933	0.927	0.221	1.646	77.8
Stou	0.006 *	0.720	0.585	0.271	1.170	100
Stoi	0.002 *	1.247	0.832	0.608	1.887	88.9
B’	0.004 *	1.313	0.984	0.556	2.070	77.8
Pg	0.016 *	1.126	1.104	0.278	1.975	88.9

*, statistical significance at p = 0.05; SD, standard deviation.

. All landmarks revealed significant differences at p = 0.05. For all linear measurements, the average error was 1.03 ± 0.83 mm. Soft tissue point B was the least accurate area (mean error of 1.31 ± 0.98 mm) while pronasele had the highest correlation between the T_P and T₁ images (mean error of 0.72 ± 0.53 mm). Table 4 also demonstrates that all landmarks had accuracies greater than 70%, with an average absolute error of less than 2 mm.

The relationship between the actual surgical movement distances and the prediction discrepancy is presented in

Table 5. Correlations between the surgical movement distances in sagittal axis and the prediction discrepancies.

	Group		P	Sn	A’	Stou	Stoi	B’	Pg
Maxilla	T	Pearson Correlation	0.772 *	−0.092	−0.152	0.499	−0.529	−0.440	−0.532
		p-value	0.015	0.814	0.696	0.172	0.143	0.236	0.140
Mandible	O	Pearson Correlation	0.156	0.374	0.705	0.355	0.896 *	0.683	0.587
		p-value	0.802	0.535	0.184	0.558	0.040	0.204	0.298
	T	Pearson Correlation	−0.141	−0.422	−0.427	−0.198	0.393	0.595	0.703 *
		p-value	0.718	0.258	0.252	0.609	0.295	0.091	0.035

*, correlation is significant at the 0.05 level (two-tailed); O, one-jaw group (n = 5); T, two-jaw group (n = 9).

. Stomian inferius was discovered to have a significant correlation with the sagittal mandibular movement in the one-jaw group whereas, in the two-jaw group, pronasele and pogonion were found to have significant correlations with the sagittal maxillary and mandibular movements, respectively. The other landmarks revealed no statistically significant differences.

3.2. Vector Values

The discrepancies were also interpreted as vector values with directional properties. The positive means indicated that the actual postoperative soft tissues are located more anteriorly than predicted. Four points (pronasele, stomian inferius, soft tissue point B, and pogonion) had a negative mean value for both groups, indicating that the prediction images were more anterior than the reality for these areas. Other maxillary landmarks, on the other hand, had positive values, as shown in

Table 6. Means and standard deviations of the vector values.

Group		P	Sn	A’	Stou	Stoi	B’	Pg
O	Mean	−0.309	0.605	0.063	0.261	−2.099	−1.535	−1.402
	SD	0.204	0.824	0.652	0.866	1.035	0.761	1.565
T	Mean	−0.041	0.867	0.739	0.843	−1.062	−1.486	−0.735
	SD	0.450	0.559	0.536	0.297	0.258	0.481	0.920

O, one-jaw group (n = 5); T, two-jaw group (n = 9).

and Figure 4.

4. Discussion

Satisfactory esthetics following orthognathic surgery are the result of using reliable and accurate prediction software [13,14]. Therefore, there have been several attempts to clarify the accuracy of such programs. As the popularity of 3D orthognathic treatment planning using virtual models has increased, many commercial 3D planning programs with varying algorithms and models have been released on the market. Dolphin imaging software (Chatsworth, CA, USA) was one of the most widely used programs in the previous study, with plenty of evidence supporting its accuracy. However, Dolphin cannot superimpose or compare STL files. A majority of studies employed another program to address this issue [7,15]. Simplant O&O, on the other hand, can both create and superimpose STL files, removing the need for additional software and reducing the error introduced by the imaging registration process. As a result, the goal of this study was to assess the soft tissue prediction accuracy of Simplant O&O software. Previously, many articles investigated the precision of this program in predicting osseous structures and its ability to transfer virtual planning to an actual surgical outcome, and they discovered that this program provided satisfactory results [16,17]. However, there is a scarcity of literature on soft tissue prediction. To the best of our knowledge, this is the first study to assess Simplant O&O’s soft tissue prediction accuracy in patients with type III skeletal malocclusion.

According to our findings, the Simplant prediction images had an overall linear mean errors of 1.73 and 1.03 mm for the one-jaw and two-jaw groups, respectively. Pronasele had the highest accuracy of both groups while pogonion and soft tissue point B had the lowest. The simulation errors in the mandible area were higher for two reasons. First, the mandible has more bony movements than the maxilla. The average sagittal advancement of the maxilla was 1.8 ± 1.3 mm, while the average mandibular setback distances were 4.5 and 6.5 mm for the one-jaw and two-jaw groups, respectively. This discrepancy corresponded to the surgical movement distance. Second, for the two-jaw group, the greater complexities in two-jaw surgeries may have resulted in small errors in the maxilla and larger errors in the mandible [18]. Even though five patients of the one-jaw group had only mandibular setbacks, all landmarks showed some discrepancies, indicating that there were also soft tissue changes in the upper jaws, even when no interventions on the maxilla were performed. According to the findings of Jokic et al. [19] and Naoumova et al. [20], the thickness of the upper lip decreases after BSSRO because the surgery eliminates the pseudo-position of the lip that occurs as a result of compensation in class III cases.

In both groups, all of the landmarks showed significant differences when compared to the actual postoperative CBCTs. Simplant O&O’s soft-tissue prediction errors were expected to be less than 2 mm. Differences in the human face less than 2 mm are not visible to the human eye, according to Kaipatur et al. [12]. Furthermore, many researchers in the literature have stated that prediction errors of less than 3 mm are not clinically significant [21,22,23,24]. Therefore, despite the fact that all of the landmarks showed significant differences, the mean values did not exceed the 2 mm accuracy threshold except for those of the stomian inferius and pogonion in the one-jaw group, as this is generally considered to be a visually perceptible facial difference, indicating that this error was not clinically significant.

There was also a correlation between the surgical distances and the landmark discrepancies. The sagittal mandibular movements in the one-jaw group were significantly correlated with stomian inferius whereas the sagittal maxilla and mandibular movements in the two-jaw group were significantly correlated with pronasele and pogonion, respectively. As a result, it could be implied that the greater the surgical movement, the lower the correlation expected from these areas.

Aside from the interpretation of the absolute values, the vector values represented the error directions. Pronasele, stomian inferius, soft tissue point B, and pogonion were all negative for both groups whereas other maxillary landmarks had positive values. As a result, this program’s soft tissue predictions for the maxilla, except for pronasele, are more posterior than they should be. The mandibular predictions, on the other hand, were more anterior than in reality. This could have occurred because orthodontic treatment after surgery can alter the Pg position due to the vertical force of elastic or tooth extrusion, which induces downward and backward mandibular rotational movements. The soft tissue pogonion will also move backward in relation to the underlying hard tissue [25]. Other orthodontic tooth movements, such as upper or lower anterior retraction, can also affect lip position. Ramos et al. [26] revealed that upper lip retraction due to maxillary incisor retraction occurred at a mean ratio of 1:0.75, and Kusnoto et al. [27] found that every millimeter of mandibular incisor retraction resulted in 0.4 mm of upper lip retraction and 0.6 mm of lower lip retraction. Al-Abdwani et al. [28] discovered that each 10-degree change in maxillary and mandibular incisor inclinations result in average horizontal changes of 0.4 mm and 0.3 mm for point A and point B, respectively. However, tooth extraction to relieve crowding or to correct abnormal inclination is usually required prior to surgery; therefore, the extensive changes in tooth movements occurred and were completed before taking the preoperative CBCTs. Only minor tooth movements occurred during the postoperative period, and they had a minor influence on the soft tissue changes after surgery, with no significant impacts on the results interpretation.

There are some differences between the results of Simplant O&O and those of other well-known programs. Maxilim 3D software (Medicim—Medical Image Computing, Mechelen, Belgium) was studied by Liebregts et al. [13] and Shafi et al. [29]. Both authors concluded that Maxilim’s 3D soft tissue prediction was clinically acceptable in general. Liebregts et al. found that the lower lip region had the lowest accuracy and the sub-nasal region had the highest accuracy when used for patients who had undergone BSSO for mandibular advancements. Shafi et al., on the other hand, studied 13 patients who underwent Le Fort I surgeries and concluded that the upper lip area was the most unpredictable. For Dolphin software studies, Resnick et al. [15] and Knoops et al. [7] studied patients who underwent only Le Fort I maxillary movements, and they identified subnasale and subalar areas as having the poorest midline agreement. Meanwhile, Elshebiny et al. [30] studied patients who underwent bimaxillary surgery and discovered that the most underpredicted areas for Dolphin 3D were nasolabial angle, soft tissue point A, and subalar area, which are roughly the same areas identified in the two studies above. As a result, we can conclude that 3D soft tissue prediction using Dolphin has acceptable accuracy, though with limitations in the subalar region. Our findings were more in line with Maxilim than with Dolphin. It is possible that this was due to the fact that both Simplant O&O and Maxilim software use the mass tensor model algorithm (MTM) [31,32] whereas Dolphin 3D imaging employs the region growing algorithm for segmentation and the sparse landmark-based photographic morphing algorithm for virtual model prediction, which was originally developed for two-dimensional predictions and then adapted to three-dimensional predictions [7,33]. All of the aforementioned programs rely on semi-automatic segmentation tools that necessitate user intervention to locate seed points or initiate contours. Recently, artificial intelligence (AI) systems or fully automatic segmentation have been validated and found to be accurate for the goal of maximizing efficiency and reducing operator variability, and they are capable of improving prediction errors [8]. However, more studies are required prior to determining actual clinical relevance.

There were several limitations in this study. First, the small sample size was a major impediment to reaching a definitive conclusion, especially for the one-jaw group. Because the inclusion criteria were quite strict, the number of subjects included was limited. The patients had skeletal type III malocclusion, and they also underwent orthognathic surgery with the same surgeon. To eliminate magnification discrepancies, both preoperative and postoperative CBCTs must be obtained from the same device. An intraoral scan and CBCT must be performed on the same day to reduce tooth movement caused by orthodontic force, which influences the fusion of the dental model to the virtual skull using voxel-based matching methods. Furthermore, only five of the fourteen patients had single-jaw surgeries. Bimaxillary surgery for skeletal type III correction has been on the rise in recent centuries, while mandibular setback surgery has been on the decline [34], which is due to the discovery that mandibular setback surgery may result in increased upper airway resistance and a narrowing of the size and proportion of the pharyngeal airway more than would result from bimaxillary surgery, leading to airway obstruction and contributing to the development of obstructive sleep apnea syndrome [33,35]. As a result, combined maxillary and mandibular osteotomies are frequently used for patients who have a significant anteroposterior discrepancy, resulting in a limited number of single-jaw samples. A larger study sample size may enable stronger conclusions to be drawn.

Second, the sample was not perfectly homogeneous. As previously stated, in addition to the fewer samples in the one-jaw group, some participants in both groups underwent genioplasty. Considering that this procedure only affects pogonion landmarks, we established the results without performing the subset analysis. To eliminate this confounding factor, a future study with definitive homogeneous samples is required.

Third, this study only examined midline landmarks and did not evaluate lateral landmarks. Since this study used soft tissue volumes and the postoperative CBCTs were collected at least 4 months after surgery, the fixed appliances remained in place. Metal artifacts from the metal brackets were scattered and had strong impacts on the landmark identifications in the lateral areas, making it difficult to produce reliable results from the lateral areas of the face. The use of a facial scan imported and fused with a virtual model, rather than a soft tissue volume created from a CBCT, can possibly solve this problem and produce a more definitive conclusion.

Finally, the time interval between surgery and postoperative CBCT varies between 4 and 13 months. Due to an ethical issue, there is no protocol for taking CBCTs six months after surgery or after completing orthodontic treatment, and this is to be accomplished in a way that does not expose a patient to additional radiation. Therefore, the postoperative CBCTs used in the study were taken when the surgeon needed to check bone healing and remove plate fixations, which occurred approximately 4–6 months after surgery. According to Van der Vlis et al. [36], only 20% of the initial edema remained after three months. However, significant reductions in soft tissue swelling continue to occur 6–12 months after surgery. As a result, some participants may have residual soft tissue swelling, which alters the interpretation of the results.

5. Conclusions

In conclusion, the purpose of this study was to assess the accuracy of the planning software’s predictions of soft tissue profiles to assist clinicians by providing a better understanding of both the program’s strengths and weaknesses. Considering the study’s limitations, it is possible to conclude that Simplant O&O has adequate accuracy for use in clinical practice for both single and bimaxillary osteotomy predictions. Some limitations have been anticipated in the mandible rather than the maxilla. The predicted image of the lower part of the face will be more anterior in the sagittal plane than the actual one, especially when surgical movement distances are extensive. Further investigation with larger samples is required to reach a definitive conclusion and assist the software in making more precise predictions by improving the underlying algorithm.