Deep Learning‐Based Three‐Dimensional Analysis Reveals Distinct Patterns of Condylar Remodelling After Orthognathic Surgery in Skeletal Class III Patients

ABSTRACT

Objective

This retrospective study aimed to evaluate morphometric changes in mandibular condyles of patients with skeletal Class III malocclusion following two‐jaw orthognathic surgery planned using virtual surgical planning (VSP) and analysed with automated three‐dimensional (3D) image analysis based on deep‐learning techniques.

Materials and Methods

Pre‐operative (T1) and 12–18 months post‐operative (T2) Cone‐Beam Computed Tomography (CBCT) scans of 17 patients (mean age: 24.8 ± 3.5 years) were analysed using 3DSlicer software. Deep‐learning algorithms automated CBCT orientation, registration, bone segmentation, and landmark identification. By utilising voxel‐based superimposition of pre‐ and post‐operative CBCT scans and shape correspondence, the overall changes in condylar morphology were assessed, with a focus on bone resorption and apposition at specific regions (superior, lateral and medial poles). The correlation between these modifications and the extent of actual condylar movements post‐surgery was investigated. Statistical analysis was conducted with a significance level of α = 0.05.

Results

Overall condylar remodelling was minimal, with mean changes of < 1 mm. Small but statistically significant bone resorption occurred at the condylar superior articular surface, while bone apposition was primarily observed at the lateral pole. The bone apposition at the lateral pole and resorption at the superior articular surface were significantly correlated with medial condylar displacement (p < 0.05).

Conclusion

The automated 3D analysis revealed distinct patterns of condylar remodelling following orthognathic surgery in skeletal Class III patients, with minimal overall changes but significant regional variations. The correlation between condylar displacements and remodelling patterns highlights the need for precise pre‐operative planning to optimise condylar positioning, potentially minimising harmful remodelling and enhancing stability.

1. Introduction

After orthognathic surgery, condylar remodelling may occur as a postoperative consequence due to condylar displacement within the articular fossa, particularly if accurate preoperative analysis is not performed. Virtual Surgical Planning (VSP) has become a pivotal tool in orthognathic surgery, offering an advanced approach to preoperative planning [1]. Compared to traditional two‐dimensional (2D) imaging, three‐dimensional (3D) assessments provide a more comprehensive view, significantly reducing intraoperative risks and errors [2]. Achieving successful outcomes relies on accurately translating the virtual plan into the operating room, ensuring optimal jaw repositioning, skeletal stability and harmonious function [1, 3]. One of the most critical factors in orthognathic surgery is precise preoperative planning and intraoperative repositioning of the mandibular proximal segments, as postoperative condylar remodelling or adaptation can significantly affect long‐term outcomes [3, 4, 5]. The mandibular condyles play a key role in maintaining proper jaw function, and significant displacement increases the risk of maladaptive remodelling. This may lead to complications such as bone resorption, condylar deformation, pain, altered jaw mechanics, postoperative relapse or temporomandibular joint disorders.

Understanding condylar adaptations remains essential for evaluating long‐term outcomes [4, 5] after orthognathic surgery. Advanced 3D imaging technologies, such as Cone Beam Computed Tomography (CBCT), have significantly improved the evaluation of mandibular condyle position [6, 7]. However, these tools are primarily used for preoperative simulation rather than actual prediction of treatment outcomes based on data evidence. Although AI‐based models that predict treatment outcomes are not yet available, automated AI tools, including deep‐learning algorithms, now allow for the careful construction of databases documenting treatment outcomes [1]. These AI‐based tools now allow for automated 3D image analysis, including automated orientation, segmentation, registration, and landmark identification. Databases analysed with these tools can provide valuable insights into postoperative changes, such as condylar remodelling, and may potentially lead to the development of decision support tools. By integrating VSP with deep learning, we can bridge the gap between preoperative planning and postoperative evaluation, creating a comprehensive framework for assessing surgical success and ensuring improved patient outcomes.

While several studies have examined the impact of condylar positioning on surgical outcomes, there is limited data on mid‐term post‐operative condylar remodelling using an automated voxel‐based approach for 3D imaging analysis [1]. Previous studies have demonstrated significant condylar remodelling following surgery, typically assessed using non‐automated methodologies that analysed specific regions of the condyle. These studies found that while both bone resorption and apposition occur, resorption tends to predominate, with maximum values of 2.63 ± 1.23 mm and a decrease in condylar height [8]. Voxel‐based registration (VBR) is the latest technique for automated superimposition of CBCT scans, utilising specific mathematical algorithms to compare pre‐ and post‐surgical condylar morphology following mandibular osteotomies [9]. Moreover, no data are currently available regarding the correlation between different types of condylar remodelling and post‐surgical condylar displacement. A comprehensive three‐dimensional analysis of orthognathic patients using a fully digital virtual planning protocol can provide valuable insights into post‐operative condylar morphology, allowing for the assessment of changes in condylar surfaces and the distinction between areas of bone apposition and bone resorption.

The aim of this study was to evaluate the qualitative and quantitative morphometric changes in the mandibular condyles of patients with skeletal Class III malocclusion who underwent two‐jaw orthognathic surgery, using an automated three‐dimensional methodology. Specifically, the analysis was conducted through voxel‐based superimposition of 3D volumetric maps derived from pre‐ and post‐operative CBCT scans to assess overall changes in condylar morphology, as well as specific modifications (bone apposition/bone resorption) on the superior articular surface, lateral and medial poles of the condyles. Additionally, the study aimed to investigate the correlation between bone apposition or bone resorption and the extent of actual condylar movements.

2. Material and Methods

This retrospective study was conducted using previously collected data. The research adhered to medical protocols and the ethical standards outlined in the Declaration of Helsinki, with approval from the Institutional Review Board (HUM00224130). Informed consent was obtained from all participants, allowing the use of their radiologic data for research purposes.

The study utilised de‐identified CBCT scans (FOV 17 × 20 cm; 110 kV; 59 mSv) from non‐growing patients, aged 18 to 32, who had Class III dentoskeletal malocclusion (A point‐Nasion‐B point angle [ANB°] < 0°; unilateral or bilateral Angle Class III molar relationship). All patients underwent bimaxillary orthognathic surgery performed by the same surgeon through a conventional method, with pre‐operative virtual surgical planning performed using Dolphin imaging software (Dolphin Imaging and Management Solutions, Chatsworth, CA, USA). The CBCT scans were securely stored in a Dropbox folder on the laboratory's protected server. Patients who had undergone single‐jaw surgery used a surgery‐first approach, or had a history of mandibular trauma, pre‐operative temporo‐mandibular disorders, TMJ surgery, craniofacial syndromic anomalies, systemic diseases or osteometabolic disorders were excluded from the study.

Pre‐operative CBCT scans, taken 1 month prior to surgery (T1), and mid‐term post‐operative scans, acquired 12–18 months after surgery (T2), were evaluated. The 3D analysis was performed using tools based on deep‐learning algorithms incorporated in the 3D Slicer software. Each T1 scan was standardised by an automated orientation according to the Frankfurt and midsagittal planes [10]. For each patient, voxel‐based registration was automatically performed, aligning the T2 scan with the T1 image, using the cranial base as the reference for superimposition, since it remains stable post‐surgery [10]. Automated bone segmentation was used to generate virtual models of the mandible, and its accuracy was verified through ITK‐SNAP software (version 3.8.0; http://www.itksnap.org) [11]. Automated identification of specific condylar landmarks (RCo and LCo, right and left‐most superior points of the condylar contour; RLCo and LLCo, right and left‐most lateral central points; RMCo and LMCo, right and left‐most medial central points) was performed and subsequently verified by an experienced orthodontist to analyse and quantify condylar displacement post‐surgery [12].

For the analysis of condylar remodelling, the region of interest was defined specifically on the condylar region (Figure 1). The total segmentation of the mandible was then cut using the automated tool AutoCrop3D for both the right and left sides. Automated voxel‐based regional registration of the T2 condyles onto the T1 condyles was performed for each patient. The registered T1 and T2 condyles were simultaneously cut at the same lower level using the Easy‐Clip tool to ensure that variations in the lower condylar level did not affect the quantification of condylar remodelling on the superior surface, as well as the medial and lateral poles. Shape analysis was conducted by subtracting the post‐surgical (T1–T2) changes between corresponding surface meshes, utilising shape correspondence analysis (SPHARM‐PDM; 3D Slicer software). Three‐dimensional point‐wise linear distances between each time point were calculated to determine remodelling changes in millimetres (Model to Model Distance; 3D Slicer software). Shape correspondence enabled the propagation of regional points across time points within a radius of two mesh points to identify three representative regions on each condyle in the 3D models (superior, medial and lateral surfaces) using the Pick n Paint in 3D Slicer software. The landmarks were initially identified using a deep‐learning‐based tool, and their positions were subsequently verified by an experienced orthodontist after a qualitative evaluation of the condylar registration. These regions were automatically propagated through different time points, allowing for comparison of the same anatomical region over time. Distances between corresponding points on the surface from T1 to T2, including 21 regional measurements on each region, were used for statistical analysis. The 15th and 85th percentiles of changes in these regions were selected to quantify differences in the overall surface, excluding outliers. The 15th percentile represents negative values, indicating inward movement (bone resorption), while the 85th percentile represents positive values, indicating outward movement (bone apposition).

FIGURE 1.

Automated workflow for condylar remodelling analysis: Identification of the region of interest (condylar area) at T1 for automated cropping; automated segmentation of the condyle; automated regional registration of the cropped T1 condyle (yellow) with the segmented T2 mandible (red); identification of specific condylar regions, including the superior, lateral, and medial poles.

Statistical analysis was carried out using R software (version 4.3.1; http://www.r‐project.org). A pilot study was conducted to determine the appropriate sample size, which concluded that 15 patients were necessary (μ1 − μ2 = 1.47; SD = 1.85; α = 0.05; β = 0.8). The Kolmogorov–Smirnov test was applied to assess the normality of the distributions. Descriptive statistics included absolute and relative frequencies for categorical variables, mean and standard deviation for normally distributed continuous data, and median with interquartile range (IR) for non‐normally distributed data. For bivariate analysis, a two‐tailed Student's t‐test was used for normal distributions, while the Wilcoxon test and Mann–Whitney U test were employed for non‐normal distributions. Linear regression modelling was performed to examine the correlation between overall remodelling and changes in each specific condylar region, as well as the relationship between remodelling and condylar movement. A significance level of α = 0.05 was used.

3. Results

Seventeen patients (9 females; 8 males; mean age: 24.8 ± 3.5 years) with skeletal Class III malocclusion were included in the study sample. The mean duration of post‐operative follow‐up was 15.4 ± 2.9 months. No signs or symptoms of temporomandibular dysfunction were reported during follow‐up, indicating functional stability in all patients. The mean overall condylar remodelling at 15th percentile was −0.84 ± 0.4 and − 0.89 ± 0.4 mm on the left and right, respectively. The mean overall condylar remodelling at 85th percentile was 0.88 ± 0.4 and 0.92 ± 0.4 mm on the left and right, respectively. For both the overall condylar surface and each individual pole, the majority of patients exhibited a mean remodelling of < 1 mm on both the right and left sides (Figure 2). No significant gender‐related differences were observed for the overall condylar surface or any individual pole on either the right or left condyles (p > 0.05). Details of condylar remodelling are presented in (Table 1 and Figure 3). The superior articular surface exhibited bone resorption on both the right and left sides, whereas the lateral pole consistently showed bone apposition on both condyles. At the medial pole, mean bone apposition was observed on the left side, while mean bone resorption occurred on the right. Table 2 presents the linear regression model analysing the correlation between overall condylar remodelling and each individual pole on the right and left sides.

FIGURE 2.

Frequences distribution of condylar remodelling 1 year after surgery on the left (A) and right (B) side: The overall surface at 15th percentile (15) and 85th percentile (85); the lateral pole at 15th percentile (L_15) and 85th percentile (L_85); the medial pole at 15th percentile (M_15) and 85th percentile (M_85); the superior articular surface at 15th percentile (S_15) and 85th percentile (S_85).

TABLE 1.

Condylar remodelling analysis: Bone resorption and apposition patterns.

Condylar remodelling a	Mean	SD	Min	25th percentile	50th percentile	75th percentile	Max
Left
Superior articular surface (mm)	−0.149	1.436	−2.903	−0.971	0.294	1.116	1.755
Lateral pole (mm)	0.123	1.108	−1.666	−0.627	−0.445	1.002	2.374
Medial pole (mm)	0.096	1.283	−1.658	−0.781	−0.251	0.877	2.469
Right
Superior articular surface (mm)	−0.381	1.536	−2.736	−1.126	−0.611	0.646	2.344
Lateral pole (mm)	0.375	1.174	−1.541	−0.808	0.567	1.240	2.106
Medial pole (mm)	−0.532	1.237	−2.122	−1.407	−0.702	−0.261	2.552

Abbreviations: Max, maximum; Min, minimum; mm, millimetres; SD, standard deviation.

^a Negative values indicate bone resorption; positive values indicate bone apposition.

FIGURE 3.

Qualitative analysis showing semi‐transparent overlays of a representative patient, highlighting right (A) and left (B) condylar remodelling one‐year post‐surgery. The focus is on bone resorption at the superior and medial poles, and bone apposition at the lateral pole.

TABLE 2.

Linear regression of overall condylar remodelling and specific condylar regions.

	Estimate	Standard error	p
Left condylar surface
Intercept	0.002	0.028	0.9
Medial pole	0.063	0.026	0.03
Superior articular surface	0.147	0.025	< 0.001
R‐squared: 0.7
Right condylar surface
Intercept	0.006	0.036	0.8
Superior articular surface	0.071	0.029	0.03
R‐squared: 0.4

Table 3 reports the correlation between condylar remodelling and post‐surgical condylar displacement. On average, a medial, posterior and lower post‐operative displacement was observed in both right and left condyles. On the left side, bone apposition at the lateral pole was significantly correlated with medial and posterior condylar displacement, while superior articular surface resorption was significantly correlated with medial displacement (p < 0.05). Additionally, resorption of the left condylar surface was significantly associated with medial displacement (p = 0.027). On the right side, bone apposition at the lateral pole was significantly correlated with medial, anterior and lower condylar displacement (p < 0.05). Resorption at the medial pole was significantly correlated with posterior and lower displacement (p < 0.05). Superior articular surface resorption was significantly linked to medial and posterior condylar displacement (p < 0.05). Bone apposition on the right condylar surface was significantly correlated with medial displacement (p < 0.05).

TABLE 3.

Linear regression of condylar remodelling and post‐surgical displacement by region.

	Estimate	Standard error	p
Left lateral pole
Intercept	−0.669	0.255	0.03
LLCo_R‐L	−1.966	0.459	0.002
LMCo_A‐P	−0.682	0.264	0.027
LMCo_R‐L	1.116	0.408	0.02
R‐squared: 0.7
Left superior articular surface
Intercept	−0.550	0.364	0.15
LLCo_R‐L	−1.622	0.732	0.04
R‐squared: 0.33
Left condylar surface
Intercept	−0.116	0.057	0.06
LLCo_R‐L	−0.249	0.099	0.027
R‐squared: 0.51
Right lateral pole
Intercept	−0.138	0.183	0.5
RLCo_A‐P	0.469	0.127	0.003
RMCo_R‐L	1.141	0.329	0.004
RMCo_S‐I	−0.223	0.082	0.017
R‐squared: 0.57
Right medial pole
Intercept	−0.712	0.299	0.036
RCo_A‐P	1.0	0.396	0.028
RLCo_S‐I	0.388	0.168	0.04
RMCo_R‐L	−1.316	0.494	0.02
R‐squared: 0.38
Right superior articular surface
Intercept	0.393	0.361	0.3
RCo_ R‐L	−0.993	0.414	0.04
RMCo_A‐P	1.338	0.460	0.017
R‐squared: 0.52
Right condylar surface
Intercept	−0.098	0.035	0.015
RCo_ R‐L	0.063	0.027	0.036
RMCo_R‐L	0.287	0.058	< 0.001
R‐squared: 0.67

4. Discussion

This study leverages deep‐learning techniques for a comprehensive 3D analysis of condylar remodelling in skeletal Class III patients following orthognathic surgery. The findings revealed distinct patterns of condylar remodelling following orthognathic surgery, particularly in the superior and lateral poles of the condyles. These results provide critical insights into the bone resorption and apposition that occur during post‐surgical adaptation, highlighting the importance of localised mechanical stress in shaping these changes.

Despite minimal overall changes in condylar morphology—< 1 mm on both sides—significant regional variations were evident at the 12‐ to 18‐month follow‐up. Bone resorption at the superior articular surface and apposition at the lateral pole were key findings, consistent with previous studies indicating that condylar height is primarily affected by remodelling in the superior or anterosuperior region due to post‐surgical forces [13, 14, 15]. This suggests that condylar remodelling is not uniform but rather driven by localised mechanical conditions following surgery. After orthognathic surgery, the condyles experience compression, leading to pressure loading on both the condyle and the glenoid fossa [16]. While bimaxillary surgery may reduce the risk of condylar resorption compared to single‐jaw surgery by minimising stress on the pterygomasseteric sling, some degree of condylar remodelling remains possible [13, 14, 15, 16]. Limited studies have evaluated 3D condylar changes after BSSO using voxel‐based registration [3, 17, 18, 19]. Although post‐surgical condylar displacement appears inevitable, minor morphological changes are generally considered clinically insignificant [1, 20]. In severe cases, condylar remodelling may alter mandibular positioning, potentially leading to long‐term skeletal relapse [3, 5]. However, the underlying mechanisms remain unclear to date [5]. One hypothesis suggests that the condyle's adaptive capacity is overwhelmed by mechanical loading [21]. Another theory points to reduced blood flow in the condyle [22]. Additionally, sex hormones may play a role, as condylar resorption is more frequently observed in women [23]. However, this study found no significant differences in condylar remodelling between male and female patients, aligning with the findings of recent analyses by Park and colleagues [13].

Regarding the nature of condylar remodelling, the superior articular surface consistently exhibited bone resorption on both the right and left sides, while the lateral pole showed bone apposition across both condyles. In contrast, the medial pole presented with mean bone apposition on one side and mean bone resorption on the other. It is noteworthy that the correlation analysis between overall condylar remodelling and each specific pole revealed a significant, direct relationship between overall remodelling and changes at the superior articular surface for both the right and left condyles. Despite the relatively minor extent of morphological changes, bone resorption at the superior articular surface may influence the total condylar surface area. Although previous studies have primarily reported a dominance of bone resorption, the presence of bone apposition was also observed, consistent with the findings of this study [14, 15, 16, 24, 25]. Factors such as condylar axis rotation, altered mechanical loading and rigid fixation can impact the condylar head, leading to varied patterns of bone resorption and apposition due to changes in the trabecular structure and mineralisation [14, 15, 16, 26, 27, 28]. Considering the minimal condylar modifications recorded in this study, these encouraging results may be attributed to the implementation of precise virtual surgical planning, which allowed for the preoperative determination of the proximal mandibular segments' positions and helped to minimise intraoperative challenges.

The detailed analysis of post‐surgical condylar displacement along the axial, coronal and sagittal planes revealed a lower, posterior and medial shift in the condylar position at the mid‐term follow‐up compared to the pre‐surgical assessment. In terms of the correlation between condylar remodelling and post‐surgical displacement, the medial rotation of the condyles in the axial plane showed a significant association with the overall remodelling of the condylar surface. This result is in accordance with previous studies that suggested inward rotation of the condyle in the axial plane after mandibular setback surgery could play a pivotal role in condylar remodelling, independent of bone resorption or apposition patterns [15, 16, 24, 28]. Furthermore, medial condylar displacement after surgery was significantly correlated with bone apposition at the lateral pole and bone resorption at the superior articular surface.

This study's findings emphasise the value of virtual surgical planning (VSP) and advanced 3D imaging techniques in reducing surgical discrepancies and optimising outcomes. By employing automated 3D analysis and voxel‐based superimposition, high precision in measuring condylar changes was achieved, offering a reliable, reproducible method for evaluating post‐surgical outcomes. This approach minimises operator‐dependent variability and enhances the ability to detect subtle morphological changes that are often missed using traditional techniques. The integration of virtual surgical planning and advanced 3D imaging techniques has become indispensable in the treatment of skeletal malocclusions, mainly because a precise condylar repositioning after mandibular osteotomy remains a significant challenge [1, 29]. Two‐dimensional radiographs have been limited in capturing angular changes and providing standardised measurements for condylar evaluation [30]. In contrast, three‐dimensional imaging combined with voxel‐based superimposition offers a more precise and objective approach for assessing condylar movements and morphological changes [1]. Although virtual surgical planning enhances the positioning of the mandibular proximal segments, positional changes can exert physical pressure on the condylar surface, leading to condylar remodelling as an adaptive response [16]. Although previous studies have examined postoperative condylar remodelling following orthognathic surgery using 3D reconstruction and superimposition techniques, none have employed an automated 3D imaging analysis capable of enhancing outcome assessment [14, 15, 16, 24, 31].

Ensuring consistency in defining reference planes and landmarks across 3D models is complex, but the use of an automated voxel‐based protocol and open‐source software driven by convolutional neural networks allowed for accurate qualitative and quantitative evaluation of post‐surgical changes. By leveraging AI‐based tools for condylar analysis, this automated approach minimises operator‐related variability, providing a reliable method for evaluating postoperative outcomes. While specific data on the performance of AI‐based tools under significant morphological changes are not yet available, additional quality control measures were implemented to ensure consistency. These steps included a thorough evaluation to confirm the robustness of the automated process, maintaining its suitability for analysing condylar displacement.

While this study provides valuable insights, the small sample size and focus on skeletal Class III malocclusion are limitations. Future research should include larger cohorts, diverse facial growth patterns and prospective study designs that incorporate both radiological and clinical assessments. Such studies will provide a more comprehensive understanding of post‐operative outcomes, contributing to the refinement of surgical techniques. Moreover, given the observed posterior displacement of the condyles post‐operatively, examining changes in the posterior condylar surface and glenoid fossa would be an important direction for future research. Although the current study focused on the superior surface and medial/lateral poles due to the minimal mean antero‐posterior displacement (< 1 mm), investigating these additional joint components would provide a more comprehensive understanding of the adaptive response to condylar displacement. Regarding ethical considerations, CBCT imaging does expose patients to ionising radiation. However, its use 12–18 months after surgery serves a critical clinical purpose beyond research. Specifically, CBCT enables the assessment of skeletal stability in the maxillary and mandibular bones, which is a key component of post‐surgical follow‐up in orthognathic patients. Despite these limitations, this study offers a strong foundation for advancing our understanding of post‐surgical condylar remodelling and improving long‐term outcomes in orthognathic treatment.

5. Conclusion

The automated 3D analysis revealed distinct patterns of condylar remodelling following orthognathic surgery in skeletal Class III patients, with minimal overall changes but significant regional variations. Specifically, bone resorption was observed at the superior articular surface, while bone apposition occurred at the lateral pole. The correlation between condylar displacements and specific remodelling patterns, particularly at the superior articular surface and lateral poles, suggests that optimising condylar positioning during surgery may minimise potentially deleterious remodelling.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.