Artificial intelligence-based upper airway segmentation for evaluating volume changes following genioplasty in patients with obstructive sleep apnea

Artificial intelligence-based upper airway segmentation for evaluating volume changes following genioplasty in patients with obstructive sleep apnea | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Artificial intelligence-based upper airway segmentation for evaluating volume changes following genioplasty in patients with obstructive sleep apnea Jiaqi Zheng, Yupeng Ruan, Xirui Wang, Jianhua Liu, Tingwei Bao This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8567392/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background : Obstructive sleep apnea (OSA) is strongly linked to upper airway anatomical compromise, with mandibular retrognathia being a risk factor. Genioplasty is commonly performed for both aesthetic and functional enhancement. Convolutional neural networks (CNNs) enable reliable segmentation of cone-beam computed tomography (CBCT) images. This study aimed to evaluate AI-based upper airway segmentation from CBCT in OSA patients who underwent genioplasty. Methods : A total of 170 CBCT images were utilized, divided into a training/validation set (n=110) and a test set (n=60). The test set consisted of 30 matched preoperative(T0) and postoperative(T1) image pairs from OSA patients with microgenia who underwent advancement sliding genioplasty. A SegResNet CNN model was employed for fully AI-based segmentation of subregional upper airway volumes, with performance assessed via dice similarity coefficient (DSC), volume similarity (VS), and 95 percentile Hausdorff Distance (95% HD). Correlations between clinical indicators, volume changes, and model metrics were analyzed. Results : The model exhibited a mean DSC value of 0.900-0.907, a mean VS value of 0.949-0.950 and a mean 95%HD of 1.485-1.588. Postoperatively, both subregions showed significant volume increases (velopharynx: 8888.19 ± 3106.34 vs. 10615.96 ± 3501.67; oropharynx: 6330.92 ±3218.49 vs. 7905.11 ± 4413.17, p<0.05), and oropharyngeal expansion weakly correlated with chin advancement magnitude. Conclusions : The present SegResNet-based model achieved fast and accurate upper airway segmentation from pre- and postoperative CBCT scans of OSA patients underwent genioplasty, establishing a basis for developing efficient analytical models to predict surgical outcomes for OSA patients. Clinical trial number : not applicable. Convolutional neural network (CNN) Cone-beam computed tomography (CBCT) Obstructive sleep apnea (OSA) Microgenia Genioplasty Upper airway Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Obstructive sleep apnea (OSA) is a sleep-disordered breathing condition characterized by intermittent obstruction of the upper airway. Anatomical and pathological variations in the upper airway region, which are largely influenced by the surrounding skeletal structures, are closely associated with the development of OSA.[1] Several studies have demonstrated the correlation between mandibular retrognathia and upper airway narrowing, indicating that mandibular retrognathia is an independent risk factor for the development of OSA.[2–4] Genioplasty has become a widely adopted surgical option for mandibular retrognathia, particularly for patients who seek aesthetic improvement. With ongoing refinements in osteotomy design, the procedure has encompassed a range of functional applications. Increasing the upper airway volume is reported to be one of the major functions of genioplasty.[5] Cone beam computed tomography (CBCT) scans have been shown to be a reliable method of examining the morphology of the upper airway. The three-dimensional(3D) reconstruction from CBCT scans enables detailed monitoring of the upper airway by separating surrounding anatomical structures. Software provides semi-automatic image segmentation and 3D reconstruction capabilities, which are still too technique-dependent and time-consuming to be routinized in daily practice. Furthermore, the anatomical boundaries that are used for subdividing the upper airway are subject to individual variations and surgical influences, which can complicate the precise delineation of reference points within the pharynx.[6,7] Convolutional neural networks (CNNs), a prominent branch of deep learning, are increasingly applied to analyze medical imaging data. These automatic methods enable rapid, precise segmentation of craniofacial anatomical structures, thereby reducing observer-related variability and improving result reproducibility. The reliability of the effectiveness of automatic segmentation models has been validated in studies of the analysis of orthodontic treatment and maxillofacial surgery.[8,9] In the present study, we aimed to evaluate the accuracy of fully AI-based segmentation of upper airway from CBCT images before and after treatment with genioplasty in OSA patients using a MONAI Label framework, which is a medical open network for artificial intelligence. Materials and methods This retrospective study received ethical approval from the Clinical Research Ethics Committee of The First Affiliated Hospital, Zhejiang University School of Medicine. Informed consent was obtained from patient. The study was conducted in accordance with the ethical standards of the responsible committee on human experimentation and the Declaration of Helsinki. Data from patients underwent genioplasty between January 2023 and September 2025, were collected. Furthermore, all CBCT images were screened to exclude those with significant artifacts or distortions that could compromise analytical quality. Based on a prior power analysis, a sample of 170 scans was randomly selected from the 230 eligible datasets that met the inclusion criteria for final analysis. The inclusion criteria were as follows: 1. Images containing the epiglottis base and hyoid bone. 2. Images captured in accordance with a standardized natural head and neck posture (maintaining a natural head position, eyes looking straight ahead, the jaw in a resting position, and the tongue at rest); 3. Images devoid of any discernible artefacts; The exclusion criteria were as follows: 1. Images showing severe distortion of the upper airway morphology due to inability to identify the epiglottis or abnormal tongue posture; 2. Images taken under different magnification ratios; The dataset was partitioned into a training/validation set (n = 110) and an independent test set (n = 60). The test set comprised matched preoperative and postoperative CBCT scans from 30 OSA patients presenting with concomitant microgenia. In comparison, the training/validation set included preoperative and postoperative images from patients exhibiting a diverse range of dentomaxillofacial deformities. Data acquisition All patients underwent a standardized imaging protocol for genioplasty assessment, which included photographic documentation and CBCT scans at two time points: preoperative (T0) and 6 months postoperatively (T1). CBCT imaging was acquired using a NewTom VGi scanner (Verona, Italy) with a consistent scanning protocol. The parameters were set as follows: tube voltage 110 kV, tube current 3.5 mA, exposure time 3.6 s, and voxel size 0.3 mm. The field of view extended from the superior orbital rim to the inferior border of the mandible. All CBCT datasets were archived in Digital Imaging and Communications in Medicine (DICOM) format. Semi-automatic segmentation CBCT datas were imported to the Mimics 21.0 software (Materialises Interactive Medical Image Control System, Leuven, Belgium). The regions of interest (ROIs) of upper airway were segmented based on a predefined Hounsfield unit (HU) window (-1000 to -500). The resulting binary volume mask was then generated and converted into Standard Triangle Language (STL) format for three-dimensional representation. All digitized landmarks were first transformed into three-dimensional coordinates using 3-Matic software (Version 12.0, Materialises NV, Leuven, Belgium). Subsequently, nine skeletal and soft tissue landmarks, along with six reference planes, were constructed based on this coordinate system (Table 1 ). Briefly, bilateral orbitales (OrL, OrR) and right porion (PR) was selected to generate Frankfort horizontal plane (FH plane), and the midsagittal plane (MSP) was defined as a plane passing through Nasion(N) and Sella(S) perpendicular to FH plane. Crown plane (CP) was then established at N, perpendicular to the FH plane and the MSP. Posterior nasal spine (PNS), Soft palate tip (SPt), Epiglottis base (Eb) and pogonion (Pog) were defined in MSP. Velopharynx (VP) and oropharynx (OP) were defined by the PNS-plane, SPt-plane, and Eb-plane which were parallel to FH plane (Fig. 1 ). To ensure reliability, the process was repeated twice at an interval of two weeks, and the ground truth was derived from the average of the two repeated measurements. Table 1 skeletal and upper airway landmarks and reference planes Landmarks: N Nasion: the most anterior point of the nasofrontal suture S Sella: the center of the sella image OrL、OrR Orbitales in both side: The lowest point of the left and right inferior orbital margin PR Porion: The uppermost point of the right external ear canal PNS Posterior nasal spine on the midsagittal sectional image SPt Tip of the soft palate on the midsagittal sectional image Eb Base of the epiglottis on the midsagittal sectional image Pog Pogonion: The anterior point of the chin Reference planes: FH plane Frankfort horizontal plane MSP Midsagittal plane: plane passing through N and S perpendicular to FH plane CP Crown plane: plane passing through N and perpendicular to FH and MSP PNS plane Plane passing through PNS and parallel to FH plane SPt plane Plane passing through SPt and parallel to FH plane Eb plane Plane passing through Eb and parallel to FH plane AI-based segmentation The 3D SegResNet architecture was shown as Fig. 2 . The model employs a 3D encoder-decoder architecture with residual connections. The encoder is structured around four successive downsampling stages, each comprising ResNet-style residual blocks. Each block contains two 3×3×3 convolutional layers, each followed by batch normalization and a ReLU activation. A max-pooling operation (stride = 2) reduces spatial dimensions between stages, while the number of feature channels doubles sequentially from an initial size of 32. Dropout (p = 0.2) is applied within residual blocks to mitigate overfitting. The decoder mirrors the encoder through three upsampling stages, where feature channels are halved and spatial resolution is recovered. Skip connections concatenate corresponding encoder and decoder feature maps to preserve fine-grained anatomical details. The final layer utilizes a 1×1×1 convolution to map features to two output classes, followed by a softmax function to generate voxel-wise segmentation probabilities.[10,11] MONAI Label, an open-source framework for medical image annotation and deep learning, was utilized to train the AI-based segmentation model via its active learning extension.[12] The 110 CBCT datasets comprising the training/validation set were imported into the 3D Slicer software platform (version 5.8.1, Harvard, USA). Three-dimensional models of the upper airway subregions, which were generated via a semi-automatic segmentation process, were imported into the software as labeled "segments". An AI-based segmentation model was thereafter developed and trained on these annotations within the MONAI Label Radiology application (version 0.8.5, https://monai.io/ ). During the training phase, the algorithm randomly selected 20% of the 110 available CBCT scans to constitute an independent validation set, which was utilized for the ongoing assessment of model performance. The model training required a total of 19.59 minutes. Training was terminated after 137 epochs using an early stopping criterion, as the validation loss plateaued and showed no further convergence. Upon completion of the model training, the 60 CBCT scans comprising the test set were processed in 3D Slicer, where they underwent AI-based segmentation by the trained model. The output 3D models were then saved in STL format.(Fig. 3 ) Evaluation metrics The performance of the AI-based segmentation was quantitatively evaluated against the semi-automatic ground truth by first deriving the four fundamental cardinalities: true positive (TP), false positive (FP), true negative (TN), and false negative (FN) voxels. These cardinalities served as the basis for calculating a comprehensive set of three evaluation metrics: the Dice Similarity Coefficient (DSC), volume similarity (VS), the 95 percentile Hausdorff Distance (95% HD). These metrics were selected to provide complementary assessments: the DSC evaluates overall spatial overlap, the VS refers to the similarity of segmented masque volumes between the two segmentation methods and the 95% HD measures the maximum boundary discrepancy while robustly excluding outliers. $$\:DSC=\frac{2TP}{2TP+FP+FN}$$ $$\:VS=1-\frac{|FN-FP|}{2TP+FP+FN}$$ The volumes of upper airway subregions were obtained through automatic measurement. The magnitude of chin advancement was quantified as the linear difference in the distance from the Pog to CP between preoperative and postoperative states. Statistical analysis The mean ± SD was calculated for data with a normal distribution; the median, interquartile range, minimum and maximum were calculated for data that did not have a normal distribution. The intraclass correlation coefficient (ICC) was calculated to provide intra-examiner reliability between repeated measurements. According to the results of the Kolmogorov–Smirnov test performed to assess the normal distribution of the datasets, the t-test and ANOVA were used to compare data between and within groups, respectively while non-parametric tests were used to analyze data that did not meet the conditions required for other analyses. Values of p<0.05 were considered statistically significant. Result The test set comprised 30 patients, including 25 females and 5 males, with an age range of 22–42 years and a mean age of 26.93 years. The mean body mass index (BMI) was 19.81 ± 2.04 kg/m². The mean follow-up period was 14.13 ± 5.58 months, ranging from 6 to 22 months. All patients underwent advancement sliding genioplasty, leading to a chin advancement of 6.26 ± 1.66 mm. Seventeen patients (56.67%) experienced numbness, but the symptom resolved spontaneously within 9 months postoperatively. The study demonstrated excellent agreement with the manually performed ground truth, with mean ICCs of 0.954 preoperatively and 0.968 postoperatively. The average duration for subregional segmentation with the model was 17.87 ± 2.86 s, while semi-automatic segmentation took 22.31 ± 6.29 min. As detailed in Table 2 , the model attained mean DSC scores of 0.900 ± 0.029 at T0 and 0.907 ± 0.021 at T1. No statistically significant differences in performance were found between the T0 and T1 scans for any airway subregion. Analysis of individual airway subregions revealed consistent regional differences, with significantly higher agreement observed in the velopharyngeal region compared to the oropharyngeal region at both T0 and T1 (p < 0.05). VS showed no statistically significant differences between the two ROIs in either preoperative or postoperative comparisons. Furthermore, the 95% HD remained below 2 mm for all subregions at both T0 and T1 (p > 0.05). Table 2 Model performance of the upper airway subsections. DSC VS 95%HD T0 T1 p T0 T1 p T0 T1 p VP 0.907 ± 0.027 0.911 ± 0.020 0.484 0.949 ± 0.037 0.948 ± 0.035 0.875 1.574 ± 0.701 1.537 ± 0.679 0.779 OP 0.888 ± 0.026 0.896 ± 0.026 0.308 0.948 ± 0.034 0.953 ± 0.034 0.607 1.688 ± 0.999 1.412 ± 0.691 0.13 Mean 0.900 ± 0.024 0.907 ± 0.021 0.196 0.949 ± 0.030 0.950 ± 0.025 0.920 1.588 ± 0.599 1.485 ± 0.641 0.267 p 0.020* 0.048* - 0.975 0.847 - 0.828 0.769 - Multiple comparison VP > OP VP > OP - - - - - - - *: p < 0.05 DSC: Dice Similarity Coefficient ; VS: volume similarity; 95% HD: 95 percentile Hausdorff Distance; VP: velopharynx; OP: oropharynx In volumetric comparisons (Table 3 ), the VP region demonstrated a significant volume increase of 1965.87 ± 2118.15 mm³, while the OP region increased by 1760.79 ± 2638.38 mm³. The calculated volume changes (ΔV) for each ROI showed no statistically significant differences between the semi-automatic ground truth and the model-derived measurements (ΔVP: 1965.87 ± 2118.15 mm 3 vs. 2134.64 ± 2128.84 mm 3 , p = 0.446; ΔOP: 1760.79 ± 2638.38 mm 3 vs. 1878.31 ± 2981.84 mm 3 , p = 0.057) (Fig. 4 ). Table 3 Comparison of the volume between AI-based segmented subregions preoperatively and postoperatively Volume(mm 3 ) T0 T1 p VP 8888.19 ± 3106.34 10615.96 ± 3501.67 < 0.001*** OP 6330.92 ± 3218.49 7905.11 ± 4413.17 0.0028** Total 15219.11 ± 5900.22 18521.07 ± 7529.41 < 0.001*** *: p < 0.05; **: p < 0.01; ***: p < 0.001 VP: velopharynx; OP: oropharynx Correlations between clinical indicators, model performance, and volume changes are shown in Table 4 . BMI was not correlated with model performance. Similarly, chin advancement exhibited no significant association with model performance on postoperative images. The DSC of the VP region at T0 showed a positive correlation with patient age. Furthermore, a positive correlation was observed between the magnitude of chin advancement and the volumetric change of the OP region. Table 4 correlation between clinical indicators, volume changes and model performance Age BMI Chin advancement VP DSC(T0) -0.400* -0.246 0.143 OP DSC(T0) -0.218 -0.275 0.196 VP 95%HD(T0) 0.346 0.235 -0.257 OP 95%HD(T0) 0.028 0.172 -0.231 VP VS(T0) -0.264 0.124 0.086 OP VS(T0) -0.248 -0.015 0.499** VP DSC(T1) -0.292 0.055 0.052 OP DSC(T1) -0.072 0.075 0.214 VP 95%HD(T1) 0.176 0.210 -0.099 OP 95%HD(T1) 0.149 0.164 0.007 VP VS(T1) -0.091 0.105 0.226 ΔVP volume 0.112 -0.104 0.136 ΔOP volume -0.214 -0.279 0.480** ΔTotal volume -0.074 -0.215 0.350 *: p < 0.05; **: p < 0.01; ***: p < 0.001 DSC: Dice Similarity Coefficient ; VS: volume similarity; 95% HD: 95 percentile Hausdorff Distance; VP: velopharynx; OP: oropharynx Discussion Mandibular retrognathia constitutes a significant risk factor for airway obstruction in non-obese OSA patients, often necessitating surgical intervention.[13,14] OSA is driven by two synergistic mechanisms: sleep-related reductions in upper airway muscle tone and inherent anatomical narrowing of the upper airway. The pharyngeal dilator muscles, particularly the genioglossus and palatopharyngeal muscles, are essential for maintaining airway patency. Kobayashi et al.[15] reported that the genioglossus counteracts pharyngeal collapse by producing anterior tongue displacement upon its retraction. Malhotra et al.[16] demonstrated that genioglossus responsiveness to negative pressure is enhanced during supine sleep compared to wakefulness but diminishes during lateral sleep, indicating a posture-specific compensatory mechanism against airway collapse. Altered upper airway muscle tone is another key pathological feature. Mezzanotte et al.[17] observed increased baseline submental muscle activity during wakefulness in OSA patients, suggesting a compensatory neuromuscular mechanism that fails during sleep, resulting in airway collapse. Advancement genioplasty is considered as a treatment option for OSA due to its ability to promote anterior displacement of the genioglossus muscle. The most common technique, sliding genioplasty, was the main focus of this study. To effectively advance the genioglossus muscle, the osteotomy must be planned to be located above the mental spine, which increased the risk of intraoperative traction on the mental nerve, resulting in an elevated incidence of temporary paresthesia of the lower lip and chin. [18] We developed a robust CNN model based on SegResNet using the MONAI Label framework for the AI-based segmentation of the upper airway before and after genioplasty in patients with OSA. Existing evidence indicates that genioplasty predominantly influences the glossopharyngeal and hypopharyngeal regions of the upper airway.[6,19] Consequently, the nasopharynx was excluded from the ROIs to avoid computational complexity and potential errors associated with segmenting the intricate anatomical structures of the nasal cavity.[20] To further enhance robustness against postural variations that disproportionately affect small volumetric spaces, the glossopharynx and hypopharynx were merged into a single segment for analysis.[21,22] Compared to manual segmentation, CNN-based methods for medical image analysis offer advantages in terms of efficiency and consistency, yet require specialized technical expertise to develop. MONAI Label addresses this barrier by integrating directly with well-established clinical platforms like 3D Slicer, which is widely used in stomatology. This allows clinicians to apply advanced segmentation without programming skills. As an open-source, plug-and-play tool, it reduces both financial costs and technical complexity, facilitating the translation of research into clinical practice. And it has been widely applied in imaging assessments across neurosurgery, orthopedics, and dentistry.[23–25] The application of CNN-based models for automatic upper airway segmentation has facilitated a more efficient delineation workflow. Consequently, a growing body of research has been dedicated to exploring and validating these methodologies within craniofacial imaging. Fernanda Nogueira-Reis et al[26] developed an automatic framework integrating six 3D U-Nets to segment craniofacial structures, including the pharyngeal airway. Their model achieved superior performance to the semi-automatic approach, with a higher mean DSC (0.996 vs. 0.883) and markedly faster processing (1.1s vs. 48.4 min). Kaan Orhan et al.[27] developed a model for automatic upper airway segmentation, which was applicable to both OSA and non-OSA patients. The AI-based results showed high agreement with manual segmentations across key metrics, including the airway's narrowest point, cross-sectional area, and volume. Dong-Yul Kim et al.[9] employed a 3D U-Net for fully automatic pharyngeal segmentation, assessing subregional volume changes following orthognathic surgery in Class III patients. The model achieved a mean DSC of 0.89, demonstrating strong concordance with manual semi-automatic segmentations on both preoperative and postoperative CBCT images. Additionally, their comparative performance analysis revealed that segmentation accuracy was higher for preoperative nasopharyngeal volumes than that in their postoperative counterparts. In this study, the model achieved a mean DSC of 0.900, a mean VS of 0.949, and a mean 95% HD below 2 mm, reflecting robust segmentation accuracy. Concurrently, the automatic segmentation approach demonstrated superior efficiency in processing time. Notably, segmentation performance remained consistent between preoperative and postoperative scans, likely due to the incorporation of both image types during training. Although the minimum DSC for oropharyngeal segmentation exceeded the accepted threshold for radiographic analysis (> 0.88), performance in this region was significantly lower than in the velopharynx. The observed reduction in segmentation accuracy may stem from the complex craniofacial anatomy characteristic of OSA patients with concurrent microgenia. In these patients. These individuals often present with an elongated soft palate that blends radiographically with the base of the tongue, obscuring the SPt landmark.[28,29] Additional anatomical variability is introduced by common variations in epiglottic morphology and position, which contribute to structural heterogeneity around the Eb landmark.[30,31] Similar to the report in previous literature, we found significant increases in velopharynx volume and oropharynx volume. A positive correlation was observed between the degree of chin advancement and the increase in oropharyngeal volume, which can be attributed to the anterior displacement of the genioglossus and the movement of the hyoid bone. Dos Santos Junior et al.[32] demonstrated that sliding genioplasty for genioglossus advancement significantly increased the posterior airway space and reduced the AHI from 12.4 to 4.4 in patients with mandibular retrognathia. Chen et al.[5] found that a modified advancement genioplasty significantly increased total airway volume by 1,528 ± 638 mm³, with the hypopharynx being the primary site of expansion, and observed concomitant anterior-superior displacement of the hyoid bone. Valls-Ontañón et al.[6] reported that aesthetic osseous genioplasty increased total upper airway volume by 9.89% (2817 ± 7256 mm³) through the upward and forward movements of the hyoid bone, with the most significant gain occurring at the oropharyngeal level by 16.8%. Kauke-Navarro et al.[19], in their systematic review, synthesized evidence indicating that isolated genioplasty leads to an average total airway volume increase of approximately 8.5% (1511 mm³) and a mean posterior airway space gain of 2.9 mm, correlating with improvements in AHI and snoring severity. This study has some limitations. First, this study was limited by its reliance on CBCT scans acquired from a single center and a single scanner model under relatively standardized acquisition protocols. Consequently, the generalizability of the developed model to heterogeneous multi-center datasets obtained from different CBCT devices with varying imaging parameters requires further validation. Expanding the sample size to include more diverse imaging sources would strengthen the model's robustness. Second, deep learning models function as "black boxes", which lake intuitive visualization of the decision-making process underlying segmentation—for instance, the specific image features that determine airway boundary delineation. Further work should integrate explainable AI techniques, such as gradient-weighted class activation mapping, to improve model interpretability and enhance greater clinical trust. Additionally, the anatomical landmarks used to define the upper airway can vary depending on patient-specific craniofacial morphology and the site of obstruction, particularly in individuals with microgenia. Since the quality of ground truth in the training dataset is inherently dependent on clinician expertise, assessing inter-examiner reliability among multiple experienced raters would help quantify annotation consistency and further enhance model reliability.[20] Conclusion This study demonstrated the potential of a SegResNet-based CNN model for quantitatively assessing subregional upper airway changes following genioplasty, with robust segmentation of the velopharynx and oropharynx in CBCT images of OSA patients with microgenia supporting this conclusion. The results revealed significant increases in both velopharyngeal and oropharyngeal volumes after advancement sliding genioplasty. Future research should focus on expanding the sample size by including data from multiple centers and different CBCT devices, validating the model's generalizability with the aim of developing a more efficient analytical tool. Declarations Acknowledgements We appreciate the editor and reviewers for their valuable feedback, which has contributed to improving this paper. Funding This work was supported by the Basic Public Welfare Research Project of Zhejiang Province, China (grant no. LGF20H140004). Consent for publication All authors consent to publicate the manuscript in COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE. CRediT authorship contribution statement Jiaqi Zheng: Writing – original draft, Visualization, Software, Formal analysis, Data curation. Yupeng Ruan: Visualization, Investigation, Validation, Data curation. Xirui Wang: Visualization, Investigation, Validation. Jianhua Liu: Supervision, Resources, Project administration. Tingwei Bao: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Funding acquisition. Availability of data and materials The datasets generated and/or analysed during the current study are not publicly available due [The clinical data of all participants in this study belongs to The First Affiliated Hospital, Zhejiang University School of Medicine and we need to obtain the approval of the hospital's medical department when obtaining it] but are available from the corresponding author on reasonable request. Ethics approval and consent to participate This retrospective study was approved by the Ethical Committee of The First Affiliated Hospital, Zhejiang University School of Medicine (Approve number: [2025B] IIT Ethics Approval No.1101. Approve date: 2025.09.23). All methods were carried out in accordance with relevant guidelines and regulations. Written informed consent was obtained from all the patients above 18 years old. Consent for publication Written informed consent for publication of identifying images or other personal or clinical details was obtained from all the patients above 18 years old. Competing interests The authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article. References R. Rojas, R. Chateau, C. Gaete, C. Muñoz, Genioglossus muscle advancement and simultaneous sliding genioplasty in the management of sleep apnoea, Int. J. Oral Maxillofac. Surg. 47 (2018) 638–641. https://doi.org/10.1016/j.ijom.2017.10.008. R. Kochhar, V. Modi, N. de Silva, S. Gueye-Ndiaye, J.M. Neugarten, M.J. Ward, L.M. Gerber, A. An, E. Mauer, H. Veler, Polysomnography-guided mandibular distraction osteogenesis in Pierre Robin sequence patients, J. Clin. Sleep Med. JCSM Off. Publ. Am. Acad. Sleep Med. 18 (2022) 1749–1755. https://doi.org/10.5664/jcsm.9960. C. Hansen, L. Sonnesen, M. Bakke, A. Markström, Prevalence of sleep-disordered breathing in children and adolescents with large horizontal maxillary overjet due to mandibular retrognathia: a case-control study, J. Clin. Sleep Med. JCSM Off. Publ. Am. Acad. Sleep Med. 20 (2024) 1871–1878. https://doi.org/10.5664/jcsm.11248. M. Shimatsu, S. Kawashima, M. Suzuki, Evaluating Craniofacial Morphology Ratios as Predictors of Obstructive Sleep Apnea Severity in Non-Obese Adult Males, Dent. J. 12 (2024) 374. https://doi.org/10.3390/dj12120374. K. Chen, X. Sun, L. Wang, S. Ni, Z. Wu, Y. Gao, X. Wang, K. Xie, S. Liu, G. Wu, A Modified Cosmetic Genioplasty Can Affect Airway Space Positively in Skeletal Class II Patients: Studying Alterations of Hyoid Bone Position and Posterior Airway Space, Aesthetic Plast. Surg. 44 (2020) 1639–1655. https://doi.org/10.1007/s00266-020-01790-3. A. Valls-Ontañón, S. Arjooca, O. Saavedra, M. Giralt-Hernando, J. Asensio-Salazar, O.L. Haas Jr., F. Hernández-Alfaro, Does Aesthetic Osseous Genioplasty Impact Upper Airway Volume?, Aesthet. Surg. J. 44 (2024) 354–362. https://doi.org/10.1093/asj/sjad341. H.-N. Cho, E. Gwon, K.-A. Kim, S.-H. Baek, N. Kim, S.-J. Kim, Accuracy of convolutional neural networks-based automatic segmentation of pharyngeal airway sections according to craniofacial skeletal pattern, Am. J. Orthod. Dentofacial Orthop. 162 (2022) e53–e62. https://doi.org/10.1016/j.ajodo.2022.01.011. Y. Süküt, E. Yurdakurban, G.S. Duran, Accuracy of deep learning-based upper airway segmentation, J. Stomatol. Oral Maxillofac. Surg. 126 (2025) 102048. https://doi.org/10.1016/j.jormas.2024.102048. D.-Y. Kim, S. Woo, J.-Y. Roh, J.-Y. Choi, K.-A. Kim, J.-Y. Cha, N. Kim, S.-J. Kim, Subregional pharyngeal changes after orthognathic surgery in skeletal Class III patients analyzed by convolutional neural networks-based segmentation, J. Dent. 135 (2023) 104565. https://doi.org/10.1016/j.jdent.2023.104565. T. Yang, G. Zhu, L. Cai, J.H. Yeo, Y. Mao, J. Yang, A benchmark study of convolutional neural networks in fully automatic segmentation of aortic root, Front. Bioeng. Biotechnol. 11 (2023) 1171868. https://doi.org/10.3389/fbioe.2023.1171868. A. Cariola, E. Sibilano, A. Guerriero, V. Bevilacqua, A. Brunetti, Deep learning strategies for semantic segmentation of pediatric brain tumors in multiparametric MRI, Sci. Rep. 15 (2025) 22595. https://doi.org/10.1038/s41598-025-07257-2. A. Diaz-Pinto, S. Alle, V. Nath, Y. Tang, A. Ihsani, M. Asad, F. Pérez-García, P. Mehta, W. Li, M. Flores, H.R. Roth, T. Vercauteren, D. Xu, P. Dogra, S. Ourselin, A. Feng, M.J. Cardoso, MONAI Label: A framework for AI-assisted interactive labeling of 3D medical images, Med. Image Anal. 95 (2024) 103207. https://doi.org/10.1016/j.media.2024.103207. W. Huang, H. Lai, Effect of exercise versus mandibular advancement device in moderate obstructive sleep apnea patients with mandibular retrognathia: A randomized clinical trial, Cranio J. Craniomandib. Pract. 43 (2025) 948–957. https://doi.org/10.1080/08869634.2024.2401643. G.E. Manrikyan, I.F. Vardanyan, M.M. Markaryan, M.E. Manrikyan, E.H. Badeyan, A.H. Manukyan, M.A. Gevorgyan, S.G. Khachatryan, Association between the Obstructive Sleep Apnea and Cephalometric Parameters in Teenagers, J. Clin. Med. 12 (2023) 6851. https://doi.org/10.3390/jcm12216851. I. Kobayashi, A. Perry, J. Rhymer, B. Wuyam, P. Hughes, K. Murphy, J.A. Innes, J. McIvor, A.D. Cheesman, A. Guz, Inspiratory coactivation of the genioglossus enlarges retroglossal space in laryngectomized humans, J. Appl. Physiol. Bethesda Md 1985 80 (1996) 1595–1604. https://doi.org/10.1152/jappl.1996.80.5.1595. A. Malhotra, J. Trinder, R. Fogel, M. Stanchina, S.R. Patel, K. Schory, D. Kleverlaan, D.P. White, Postural Effects on Pharyngeal Protective Reflex Mechanisms, Sleep 27 (2004) 1105–1112. https://doi.org/10.1093/sleep/27.6.1105. W.S. Mezzanotte, D.J. Tangel, D.P. White, Waking genioglossal electromyogram in sleep apnea patients versus normal controls (a neuromuscular compensatory mechanism), J. Clin. Invest. 89 (1992) 1571–1579. https://doi.org/10.1172/JCI115751. B.K. Choi, I.S. Yun, Y.S. Kim, T.S. Roh, S.E. Park, J.Y. Bae, B.K. Jung, Effects of Hat-Shaped Mortised Genioplasty with Genioglossus Muscle Advancement on Retrogenia and Snoring: Assessment of Esthetic, Functional, and Psychosocial Results, Aesthetic Plast. Surg. 43 (2019) 412–419. https://doi.org/10.1007/s00266-018-1290-z. M. Kauke-Navarro, L. Knoedler, L. Stucki, F.J. Klimitz, O. Allam, S. Knoedler, M.C. Navarro, P. Flores-Pérez, A.L. Rancu, M. Heiland, M. Alperovich, A.-F. Safi, Osseous Genioplasty and Its Impact on Airway Volume—What is the Evidence?, Aesthetic Surg. J. Open Forum 7 (2025) ojaf059. https://doi.org/10.1093/asjof/ojaf059. S. Jin, H. Han, Z. Huang, Y. Xiang, M. Du, F. Hua, X. Guan, J. Liu, F. Chen, H. He, Automatic three-dimensional nasal and pharyngeal airway subregions identification via Vision Transformer, J. Dent. 136 (2023) 104595. https://doi.org/10.1016/j.jdent.2023.104595. T. Pham, C.-K. Lin, D. Leek, R. Chandrashekhar, T.M. Annaswamy, Obstructive sleep Apnea’s association with the cervical spine abnormalities, posture, and pain: a systematic review, Sleep Med. 75 (2020) 468–476. https://doi.org/10.1016/j.sleep.2020.09.008. B. Solow, S. Skov, J. Ovesen, P.W. Norup, G. Wildschiødtz, Airway dimensions and head posture in obstructive sleep apnoea, Eur. J. Orthod. 18 (1996) 571–579. https://doi.org/10.1093/ejo/18.6.571. S. Cepeda, R. Romero, L. Luque, D. García-Pérez, G. Blasco, L.T. Luppino, S. Kuttner, O. Esteban-Sinovas, I. Arrese, O. Solheim, L. Eikenes, A. Karlberg, Á. Pérez-Núñez, O. Zanier, C. Serra, V.E. Staartjes, A. Bianconi, L.F. Rossi, D. Garbossa, T. Escudero, R. Hornero, R. Sarabia, Deep learning-based postoperative glioblastoma segmentation and extent of resection evaluation: Development, external validation, and model comparison, Neuro-Oncol. Adv. 6 (2024) vdae199. https://doi.org/10.1093/noajnl/vdae199. J. Shapey, A. Kujawa, R. Dorent, G. Wang, A. Dimitriadis, D. Grishchuk, I. Paddick, N. Kitchen, R. Bradford, S.R. Saeed, S. Bisdas, S. Ourselin, T. Vercauteren, Segmentation of vestibular schwannoma from MRI, an open annotated dataset and baseline algorithm, Sci. Data 8 (2021) 286. https://doi.org/10.1038/s41597-021-01064-w. E. Yurdakurban, Y. Süküt, G.S. Duran, Assessment of deep learning technique for fully automated mandibular segmentation, Am. J. Orthod. Dentofac. Orthop. Off. Publ. Am. Assoc. Orthod. Its Const. Soc. Am. Board Orthod. 167 (2025) 242–249. https://doi.org/10.1016/j.ajodo.2024.09.006. F. Nogueira-Reis, N. Morgan, I.R. Suryani, C.P.M. Tabchoury, R. Jacobs, Full virtual patient generated by artificial intelligence-driven integrated segmentation of craniomaxillofacial structures from CBCT images, J. Dent. 141 (2024) 104829. https://doi.org/10.1016/j.jdent.2023.104829. K. Orhan, M. Shamshiev, M. Ezhov, A. Plaksin, A. Kurbanova, G. Ünsal, M. Gusarev, M. Golitsyna, S. Aksoy, M. Mısırlı, F. Rasmussen, E. Shumilov, A. Sanders, AI-based automatic segmentation of craniomaxillofacial anatomy from CBCT scans for automatic detection of pharyngeal airway evaluations in OSA patients, Sci. Rep. 12 (2022) 11863. https://doi.org/10.1038/s41598-022-15920-1. S. Navasumrit, Y.-A. Chen, Y.-J. Hsieh, C.-F. Yao, C.-S. Chang, N.-H. Chen, Y.-F. Liao, Y.-R. Chen, Skeletal and upper airway stability following modified maxillomandibular advancement for treatment of obstructive sleep apnea in skeletal class I or II deformity, Clin. Oral Investig. 26 (2022) 3239–3250. https://doi.org/10.1007/s00784-021-04306-8. T. Jiang, Y. Qi, Y. Zhang, Y. Du, Q. Wu, H. Xiao, K. He, J. Zheng, Z. Jin, F. Li, Changes in upper airway morphology and respiratory function of adolescent patients with mandibular retrognathism treated with clear aligner mandibular advancement: a prospective study, BMC Oral Health 25 (2025) 975. https://doi.org/10.1186/s12903-025-06284-9. I.-C. Kuo, L.-J. Hsin, L.-A. Lee, T.-J. Fang, M.-S. Tsai, Y.-C. Lee, S.-C. Shen, H.-Y. Li, Prediction of Epiglottic Collapse in Obstructive Sleep Apnea Patients: Epiglottic Length, Nat. Sci. Sleep 13 (2021) 1985–1992. https://doi.org/10.2147/NSS.S336019. X. Wang, H. Chen, L. Jia, X. Xu, J. Guo, The relationship between three-dimensional craniofacial and upper airway anatomical variables and severity of obstructive sleep apnoea in adults, Eur. J. Orthod. 44 (2022) 78–85. https://doi.org/10.1093/ejo/cjab014. J.F. dos Santos Junior, M. Abrahão, L.C. Gregório, A.I. Zonato, E.H. Gumieiro, Genioplasty for genioglossus muscle advancement in patients with obstructive sleep apnea-hypopnea syndrome and mandibular retrognathia, Braz. J. Otorhinolaryngol. 73 (2015) 480–486. https://doi.org/10.1016/S1808-8694(15)30099-9. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8567392","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":585291243,"identity":"8de291c2-7dd8-4e04-baa7-4617c7ab6243","order_by":0,"name":"Jiaqi Zheng","email":"","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Jiaqi","middleName":"","lastName":"Zheng","suffix":""},{"id":585291244,"identity":"ddd0c9d9-b513-448e-8956-b5f87acdfc97","order_by":1,"name":"Yupeng Ruan","email":"","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Yupeng","middleName":"","lastName":"Ruan","suffix":""},{"id":585291245,"identity":"32df71c9-e1d3-4567-9f91-9686e1779b06","order_by":2,"name":"Xirui Wang","email":"","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Xirui","middleName":"","lastName":"Wang","suffix":""},{"id":585291246,"identity":"19f13865-484e-4dd4-8059-1595f5ae5f72","order_by":3,"name":"Jianhua Liu","email":"","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Jianhua","middleName":"","lastName":"Liu","suffix":""},{"id":585291247,"identity":"d02888f0-8ae8-467b-a5f5-8f6cdbef98f4","order_by":4,"name":"Tingwei Bao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIie3RsQrCMBCA4StCp2jXE6F9hYijPkyLYKeCkzgmFOoiuEYcfAbfoOXAqTg7OHRz7dhF0ergmLgJ5t8O7uMCAbDZfjDXk7IKl+gD5O1oQHpIKa/Lycic+BBn/W02i8T7qMnDoBAD5lK8lyWHekHg7YSGdKQYMUaJFCV31IkAL7n+ypQhJSmUvNPNCDiGOhIJYpxityU3QyKlCmcha4ljRLBIoc4nQwXHebE+xQzPGhJsVtcmumMQKDpUzWLse0pDPmH++kxmuv/ME18s22w221/1ALYYRdx8t24gAAAAAElFTkSuQmCC","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":true,"prefix":"","firstName":"Tingwei","middleName":"","lastName":"Bao","suffix":""}],"badges":[],"createdAt":"2026-01-10 09:54:00","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8567392/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8567392/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102209527,"identity":"e28d1891-0cc2-4e6f-a1d2-d61d21f95155","added_by":"auto","created_at":"2026-02-09 12:13:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":315265,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ea.\u003c/em\u003e \u003cem\u003eCBCT image illustrating the delineation of upper airway boundaries for defining ROIs: VP: velopharynx; OP: oropharynx.\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eb.\u003c/em\u003e \u003cem\u003eNine landmarks and six reference planes were constructed to manual segment 2 upper airway ROIs: N: nasion; PNS: posterior nasal spine; S: sella; OrL, OrR: orbitales in both side; PR: porion; Pog: pogonion; SPt: tip of the soft palate; Eb: base of the epiglottis; FH plane: Frankfort horizontal plane; MSP: Midsagittal plane; CP: Crown plane; PNS plane: plane passing through PNS and parallel to FH plane; SPt plane: plane passing through SPt and parallel to FH plane; Eb plane: plane passing through Eb and parallel to FH plane\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8567392/v1/46cbbe458e2e79a51b58d4fe.png"},{"id":102209511,"identity":"9560bdc5-a829-48e2-931b-76ceb125563e","added_by":"auto","created_at":"2026-02-09 12:13:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":105526,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eThe architecture of the 3D SegResNet\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8567392/v1/44b4b9ac039035b72a1b9615.png"},{"id":102209362,"identity":"812d14ac-f1a0-43da-a099-506dddaa8616","added_by":"auto","created_at":"2026-02-09 12:12:59","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":489848,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003ea.\u003c/em\u003e \u003cem\u003eground truth constructed by semi-automatic method\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eb.\u003c/em\u003e \u003cem\u003etest set constructed by AI-based method\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ec.\u003c/em\u003e \u003cem\u003ecomparison between test set and ground truth\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8567392/v1/909508658aa5ab8701845a59.png"},{"id":102209512,"identity":"bbe0f005-8fbb-497f-8efd-64506d432fc4","added_by":"auto","created_at":"2026-02-09 12:13:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":24343,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eVP: velopharynx; OP: oropharynx\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eComparison of the volume changes (ΔV) between the model performance and ground truth.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8567392/v1/d5f61db53195ef4f97a5ae86.png"},{"id":104231076,"identity":"88a2f137-505f-4b9d-9382-79799520294c","added_by":"auto","created_at":"2026-03-09 12:13:12","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1782685,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8567392/v1/ce81e097-899d-46bd-a4c8-428484fcec85.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial intelligence-based upper airway segmentation for evaluating volume changes following genioplasty in patients with obstructive sleep apnea","fulltext":[{"header":"Introduction","content":"\u003cp\u003eObstructive sleep apnea (OSA) is a sleep-disordered breathing condition characterized by intermittent obstruction of the upper airway. Anatomical and pathological variations in the upper airway region, which are largely influenced by the surrounding skeletal structures, are closely associated with the development of OSA.[1] Several studies have demonstrated the correlation between mandibular retrognathia and upper airway narrowing, indicating that mandibular retrognathia is an independent risk factor for the development of OSA.[2\u0026ndash;4] Genioplasty has become a widely adopted surgical option for mandibular retrognathia, particularly for patients who seek aesthetic improvement. With ongoing refinements in osteotomy design, the procedure has encompassed a range of functional applications. Increasing the upper airway volume is reported to be one of the major functions of genioplasty.[5]\u003c/p\u003e \u003cp\u003eCone beam computed tomography (CBCT) scans have been shown to be a reliable method of examining the morphology of the upper airway. The three-dimensional(3D) reconstruction from CBCT scans enables detailed monitoring of the upper airway by separating surrounding anatomical structures. Software provides semi-automatic image segmentation and 3D reconstruction capabilities, which are still too technique-dependent and time-consuming to be routinized in daily practice. Furthermore, the anatomical boundaries that are used for subdividing the upper airway are subject to individual variations and surgical influences, which can complicate the precise delineation of reference points within the pharynx.[6,7]\u003c/p\u003e \u003cp\u003eConvolutional neural networks (CNNs), a prominent branch of deep learning, are increasingly applied to analyze medical imaging data. These automatic methods enable rapid, precise segmentation of craniofacial anatomical structures, thereby reducing observer-related variability and improving result reproducibility. The reliability of the effectiveness of automatic segmentation models has been validated in studies of the analysis of orthodontic treatment and maxillofacial surgery.[8,9]\u003c/p\u003e \u003cp\u003eIn the present study, we aimed to evaluate the accuracy of fully AI-based segmentation of upper airway from CBCT images before and after treatment with genioplasty in OSA patients using a MONAI Label framework, which is a medical open network for artificial intelligence.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003eThis retrospective study received ethical approval from the Clinical Research Ethics Committee of The First Affiliated Hospital, Zhejiang University School of Medicine. Informed consent was obtained from patient. The study was conducted in accordance with the ethical standards of the responsible committee on human experimentation and the Declaration of Helsinki. Data from patients underwent genioplasty between January 2023 and September 2025, were collected. Furthermore, all CBCT images were screened to exclude those with significant artifacts or distortions that could compromise analytical quality. Based on a prior power analysis, a sample of 170 scans was randomly selected from the 230 eligible datasets that met the inclusion criteria for final analysis.\u003c/p\u003e\n\u003cp\u003eThe inclusion criteria were as follows:\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e1. Images containing the epiglottis base and hyoid bone.\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e2. Images captured in accordance with a standardized natural head and neck posture (maintaining a natural head position, eyes looking straight ahead, the jaw in a resting position, and the tongue at rest);\u003c/p\u003e\n\u003c/span\u003e\u003cspan\u003e\n \u003cp\u003e3. Images devoid of any discernible artefacts;\u003c/p\u003e\n\u003c/span\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe exclusion criteria were as follows:\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e1. Images showing severe distortion of the upper airway morphology due to inability to identify the epiglottis or abnormal tongue posture;\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e2. Images taken under different magnification ratios;\u003c/p\u003e\n\u003c/span\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset was partitioned into a training/validation set (n\u0026thinsp;=\u0026thinsp;110) and an independent test set (n\u0026thinsp;=\u0026thinsp;60). The test set comprised matched preoperative and postoperative CBCT scans from 30 OSA patients presenting with concomitant microgenia. In comparison, the training/validation set included preoperative and postoperative images from patients exhibiting a diverse range of dentomaxillofacial deformities.\u003c/p\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eData acquisition\u003c/h2\u003e\n \u003cp\u003eAll patients underwent a standardized imaging protocol for genioplasty assessment, which included photographic documentation and CBCT scans at two time points: preoperative (T0) and 6 months postoperatively (T1). CBCT imaging was acquired using a NewTom VGi scanner (Verona, Italy) with a consistent scanning protocol. The parameters were set as follows: tube voltage 110 kV, tube current 3.5 mA, exposure time 3.6 s, and voxel size 0.3 mm. The field of view extended from the superior orbital rim to the inferior border of the mandible. All CBCT datasets were archived in Digital Imaging and Communications in Medicine (DICOM) format.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eSemi-automatic segmentation\u003c/h3\u003e\n\u003cp\u003eCBCT datas were imported to the Mimics 21.0 software (Materialises Interactive Medical Image Control System, Leuven, Belgium). The regions of interest (ROIs) of upper airway were segmented based on a predefined Hounsfield unit (HU) window (-1000 to -500). The resulting binary volume mask was then generated and converted into Standard Triangle Language (STL) format for three-dimensional representation. All digitized landmarks were first transformed into three-dimensional coordinates using 3-Matic software (Version 12.0, Materialises NV, Leuven, Belgium). Subsequently, nine skeletal and soft tissue landmarks, along with six reference planes, were constructed based on this coordinate system (Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). Briefly, bilateral orbitales (OrL, OrR) and right porion (PR) was selected to generate Frankfort horizontal plane (FH plane), and the midsagittal plane (MSP) was defined as a plane passing through Nasion(N) and Sella(S) perpendicular to FH plane. Crown plane (CP) was then established at N, perpendicular to the FH plane and the MSP. Posterior nasal spine (PNS), Soft palate tip (SPt), Epiglottis base (Eb) and pogonion (Pog) were defined in MSP. Velopharynx (VP) and oropharynx (OP) were defined by the PNS-plane, SPt-plane, and Eb-plane which were parallel to FH plane (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e). To ensure reliability, the process was repeated twice at an interval of two weeks, and the ground truth was derived from the average of the two repeated measurements.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eskeletal and upper airway landmarks and reference planes\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLandmarks:\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNasion: the most anterior point of the nasofrontal suture\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSella: the center of the sella image\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOrL、OrR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOrbitales in both side: The lowest point of the left and right inferior orbital margin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePorion: The uppermost point of the right external ear canal\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePNS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePosterior nasal spine on the midsagittal sectional image\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSPt\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTip of the soft palate on the midsagittal sectional image\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBase of the epiglottis on the midsagittal sectional image\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePog\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePogonion: The anterior point of the chin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eReference planes:\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFH plane\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFrankfort horizontal plane\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMSP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMidsagittal plane: plane passing through N and S perpendicular to FH plane\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCrown plane: plane passing through N and perpendicular to FH and MSP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePNS plane\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlane passing through PNS and parallel to FH plane\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSPt plane\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlane passing through SPt and parallel to FH plane\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEb plane\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePlane passing through Eb and parallel to FH plane\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003ch3\u003eAI-based segmentation\u003c/h3\u003e\n\u003cp\u003eThe 3D SegResNet architecture was shown as Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e. The model employs a 3D encoder-decoder architecture with residual connections. The encoder is structured around four successive downsampling stages, each comprising ResNet-style residual blocks. Each block contains two 3\u0026times;3\u0026times;3 convolutional layers, each followed by batch normalization and a ReLU activation. A max-pooling operation (stride\u0026thinsp;=\u0026thinsp;2) reduces spatial dimensions between stages, while the number of feature channels doubles sequentially from an initial size of 32. Dropout (p\u0026thinsp;=\u0026thinsp;0.2) is applied within residual blocks to mitigate overfitting. The decoder mirrors the encoder through three upsampling stages, where feature channels are halved and spatial resolution is recovered. Skip connections concatenate corresponding encoder and decoder feature maps to preserve fine-grained anatomical details. The final layer utilizes a 1\u0026times;1\u0026times;1 convolution to map features to two output classes, followed by a softmax function to generate voxel-wise segmentation probabilities.[10,11]\u003c/p\u003e\n\u003cp\u003eMONAI Label, an open-source framework for medical image annotation and deep learning, was utilized to train the AI-based segmentation model via its active learning extension.[12] The 110 CBCT datasets comprising the training/validation set were imported into the 3D Slicer software platform (version 5.8.1, Harvard, USA). Three-dimensional models of the upper airway subregions, which were generated via a semi-automatic segmentation process, were imported into the software as labeled \u0026quot;segments\u0026quot;. An AI-based segmentation model was thereafter developed and trained on these annotations within the MONAI Label Radiology application (version 0.8.5, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://monai.io/\u003c/span\u003e\u003c/span\u003e). During the training phase, the algorithm randomly selected 20% of the 110 available CBCT scans to constitute an independent validation set, which was utilized for the ongoing assessment of model performance. The model training required a total of 19.59 minutes. Training was terminated after 137 epochs using an early stopping criterion, as the validation loss plateaued and showed no further convergence.\u003c/p\u003e\n\u003cp\u003eUpon completion of the model training, the 60 CBCT scans comprising the test set were processed in 3D Slicer, where they underwent AI-based segmentation by the trained model. The output 3D models were then saved in STL format.(Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e)\u003c/p\u003e\n\u003ch3\u003eEvaluation metrics\u003c/h3\u003e\n\u003cp\u003eThe performance of the AI-based segmentation was quantitatively evaluated against the semi-automatic ground truth by first deriving the four fundamental cardinalities: true positive (TP), false positive (FP), true negative (TN), and false negative (FN) voxels. These cardinalities served as the basis for calculating a comprehensive set of three evaluation metrics: the Dice Similarity Coefficient (DSC), volume similarity (VS), the 95 percentile Hausdorff Distance (95% HD). These metrics were selected to provide complementary assessments: the DSC evaluates overall spatial overlap, the VS refers to the similarity of segmented masque volumes between the two segmentation methods and the 95% HD measures the maximum boundary discrepancy while robustly excluding outliers.\u003c/p\u003e\n\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:DSC=\\frac{2TP}{2TP+FP+FN}$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e$$\\:VS=1-\\frac{|FN-FP|}{2TP+FP+FN}$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eThe volumes of upper airway subregions were obtained through automatic measurement. The magnitude of chin advancement was quantified as the linear difference in the distance from the Pog to CP between preoperative and postoperative states.\u003c/p\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003eStatistical analysis\u003c/h2\u003e\n \u003cp\u003eThe mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD was calculated for data with a normal distribution; the median, interquartile range, minimum and maximum were calculated for data that did not have a normal distribution. The intraclass correlation coefficient (ICC) was calculated to provide intra-examiner reliability between repeated measurements. According to the results of the Kolmogorov\u0026ndash;Smirnov test performed to assess the normal distribution of the datasets, the t-test and ANOVA were used to compare data between and within groups, respectively while non-parametric tests were used to analyze data that did not meet the conditions required for other analyses. Values of p\u0026lt;0.05 were considered statistically significant.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Result","content":"\u003cp\u003eThe test set comprised 30 patients, including 25 females and 5 males, with an age range of 22\u0026ndash;42 years and a mean age of 26.93 years. The mean body mass index (BMI) was 19.81\u0026thinsp;\u0026plusmn;\u0026thinsp;2.04 kg/m\u0026sup2;. The mean follow-up period was 14.13\u0026thinsp;\u0026plusmn;\u0026thinsp;5.58 months, ranging from 6 to 22 months. All patients underwent advancement sliding genioplasty, leading to a chin advancement of 6.26\u0026thinsp;\u0026plusmn;\u0026thinsp;1.66 mm. Seventeen patients (56.67%) experienced numbness, but the symptom resolved spontaneously within 9 months postoperatively.\u003c/p\u003e\n\u003cp\u003eThe study demonstrated excellent agreement with the manually performed ground truth, with mean ICCs of 0.954 preoperatively and 0.968 postoperatively. The average duration for subregional segmentation with the model was 17.87\u0026thinsp;\u0026plusmn;\u0026thinsp;2.86 s, while semi-automatic segmentation took 22.31\u0026thinsp;\u0026plusmn;\u0026thinsp;6.29 min. As detailed in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, the model attained mean DSC scores of 0.900\u0026thinsp;\u0026plusmn;\u0026thinsp;0.029 at T0 and 0.907\u0026thinsp;\u0026plusmn;\u0026thinsp;0.021 at T1. No statistically significant differences in performance were found between the T0 and T1 scans for any airway subregion. Analysis of individual airway subregions revealed consistent regional differences, with significantly higher agreement observed in the velopharyngeal region compared to the oropharyngeal region at both T0 and T1 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). VS showed no statistically significant differences between the two ROIs in either preoperative or postoperative comparisons. Furthermore, the 95% HD remained below 2 mm for all subregions at both T0 and T1 (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eModel performance of the upper airway subsections.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" rowspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eDSC\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eVS\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003e95%HD\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT0\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT0\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT0\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.907\u0026thinsp;\u0026plusmn;\u0026thinsp;0.027\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.911\u0026thinsp;\u0026plusmn;\u0026thinsp;0.020\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.484\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.949\u0026thinsp;\u0026plusmn;\u0026thinsp;0.037\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.948\u0026thinsp;\u0026plusmn;\u0026thinsp;0.035\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.875\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.574\u0026thinsp;\u0026plusmn;\u0026thinsp;0.701\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.537\u0026thinsp;\u0026plusmn;\u0026thinsp;0.679\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.779\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.888\u0026thinsp;\u0026plusmn;\u0026thinsp;0.026\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.896\u0026thinsp;\u0026plusmn;\u0026thinsp;0.026\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.308\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.948\u0026thinsp;\u0026plusmn;\u0026thinsp;0.034\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.953\u0026thinsp;\u0026plusmn;\u0026thinsp;0.034\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.607\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.688\u0026thinsp;\u0026plusmn;\u0026thinsp;0.999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.412\u0026thinsp;\u0026plusmn;\u0026thinsp;0.691\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.13\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.900\u0026thinsp;\u0026plusmn;\u0026thinsp;0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.907\u0026thinsp;\u0026plusmn;\u0026thinsp;0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.196\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.949\u0026thinsp;\u0026plusmn;\u0026thinsp;0.030\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.950\u0026thinsp;\u0026plusmn;\u0026thinsp;0.025\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.920\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.588\u0026thinsp;\u0026plusmn;\u0026thinsp;0.599\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.485\u0026thinsp;\u0026plusmn;\u0026thinsp;0.641\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.267\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.020*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.048*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.975\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.847\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.828\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMultiple comparison\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP\u0026thinsp;\u0026gt;\u0026thinsp;OP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP\u0026thinsp;\u0026gt;\u0026thinsp;OP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\"\u003e\u003cem\u003e*: p\u0026thinsp;\u0026lt;\u0026thinsp;0.05\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\"\u003e\u003cem\u003eDSC: Dice Similarity Coefficient ; VS: volume similarity; 95% HD: 95 percentile Hausdorff Distance; VP: velopharynx; OP: oropharynx\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eIn volumetric comparisons (Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e), the VP region demonstrated a significant volume increase of 1965.87\u0026thinsp;\u0026plusmn;\u0026thinsp;2118.15 mm\u0026sup3;, while the OP region increased by 1760.79\u0026thinsp;\u0026plusmn;\u0026thinsp;2638.38 mm\u0026sup3;. The calculated volume changes (\u0026Delta;V) for each ROI showed no statistically significant differences between the semi-automatic ground truth and the model-derived measurements (\u0026Delta;VP: 1965.87\u0026thinsp;\u0026plusmn;\u0026thinsp;2118.15 mm\u003csup\u003e3\u003c/sup\u003e vs. 2134.64\u0026thinsp;\u0026plusmn;\u0026thinsp;2128.84 mm\u003csup\u003e3\u003c/sup\u003e, p\u0026thinsp;=\u0026thinsp;0.446; \u0026Delta;OP: 1760.79\u0026thinsp;\u0026plusmn;\u0026thinsp;2638.38 mm\u003csup\u003e3\u003c/sup\u003e vs. 1878.31\u0026thinsp;\u0026plusmn;\u0026thinsp;2981.84 mm\u003csup\u003e3\u003c/sup\u003e, p\u0026thinsp;=\u0026thinsp;0.057) (Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eComparison of the volume between AI-based segmented subregions preoperatively and postoperatively\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eVolume(mm\u003csup\u003e3\u003c/sup\u003e)\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT0\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eT1\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8888.19\u0026thinsp;\u0026plusmn;\u0026thinsp;3106.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10615.96\u0026thinsp;\u0026plusmn;\u0026thinsp;3501.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001***\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6330.92\u0026thinsp;\u0026plusmn;\u0026thinsp;3218.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e7905.11\u0026thinsp;\u0026plusmn;\u0026thinsp;4413.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.0028**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTotal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e15219.11\u0026thinsp;\u0026plusmn;\u0026thinsp;5900.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18521.07\u0026thinsp;\u0026plusmn;\u0026thinsp;7529.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e\u0026lt;\u0026thinsp;0.001***\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003e*: p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; **: p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; ***: p\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003eVP: velopharynx; OP: oropharynx\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eCorrelations between clinical indicators, model performance, and volume changes are shown in Table \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e. BMI was not correlated with model performance. Similarly, chin advancement exhibited no significant association with model performance on postoperative images. The DSC of the VP region at T0 showed a positive correlation with patient age. Furthermore, a positive correlation was observed between the magnitude of chin advancement and the volumetric change of the OP region.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003ctable id=\"Tab4\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003ecorrelation between clinical indicators, volume changes and model performance\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eChin advancement\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP DSC(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.400*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.246\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.143\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP DSC(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.218\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.275\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.196\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP 95%HD(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.346\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.235\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.257\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP 95%HD(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.028\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.172\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.231\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP VS(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.264\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.124\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP VS(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.248\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.015\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.499**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP DSC(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.292\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.055\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.052\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP DSC(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.072\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.075\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.214\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP 95%HD(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.176\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.210\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.099\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eOP 95%HD(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.149\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.164\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.007\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVP VS(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.091\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.105\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.226\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026Delta;VP volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.112\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.136\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026Delta;OP volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.214\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.279\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.480**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026Delta;Total volume\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.074\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.215\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.350\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003e*: p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; **: p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; ***: p\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e\u003cem\u003eDSC: Dice Similarity Coefficient ; VS: volume similarity; 95% HD: 95 percentile Hausdorff Distance; VP: velopharynx; OP: oropharynx\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eMandibular retrognathia constitutes a significant risk factor for airway obstruction in non-obese OSA patients, often necessitating surgical intervention.[13,14] OSA is driven by two synergistic mechanisms: sleep-related reductions in upper airway muscle tone and inherent anatomical narrowing of the upper airway. The pharyngeal dilator muscles, particularly the genioglossus and palatopharyngeal muscles, are essential for maintaining airway patency. Kobayashi et al.[15] reported that the genioglossus counteracts pharyngeal collapse by producing anterior tongue displacement upon its retraction. Malhotra et al.[16] demonstrated that genioglossus responsiveness to negative pressure is enhanced during supine sleep compared to wakefulness but diminishes during lateral sleep, indicating a posture-specific compensatory mechanism against airway collapse. Altered upper airway muscle tone is another key pathological feature. Mezzanotte et al.[17] observed increased baseline submental muscle activity during wakefulness in OSA patients, suggesting a compensatory neuromuscular mechanism that fails during sleep, resulting in airway collapse. Advancement genioplasty is considered as a treatment option for OSA due to its ability to promote anterior displacement of the genioglossus muscle. The most common technique, sliding genioplasty, was the main focus of this study. To effectively advance the genioglossus muscle, the osteotomy must be planned to be located above the mental spine, which increased the risk of intraoperative traction on the mental nerve, resulting in an elevated incidence of temporary paresthesia of the lower lip and chin. [18]\u003c/p\u003e \u003cp\u003eWe developed a robust CNN model based on SegResNet using the MONAI Label framework for the AI-based segmentation of the upper airway before and after genioplasty in patients with OSA. Existing evidence indicates that genioplasty predominantly influences the glossopharyngeal and hypopharyngeal regions of the upper airway.[6,19] Consequently, the nasopharynx was excluded from the ROIs to avoid computational complexity and potential errors associated with segmenting the intricate anatomical structures of the nasal cavity.[20] To further enhance robustness against postural variations that disproportionately affect small volumetric spaces, the glossopharynx and hypopharynx were merged into a single segment for analysis.[21,22] Compared to manual segmentation, CNN-based methods for medical image analysis offer advantages in terms of efficiency and consistency, yet require specialized technical expertise to develop. MONAI Label addresses this barrier by integrating directly with well-established clinical platforms like 3D Slicer, which is widely used in stomatology. This allows clinicians to apply advanced segmentation without programming skills. As an open-source, plug-and-play tool, it reduces both financial costs and technical complexity, facilitating the translation of research into clinical practice. And it has been widely applied in imaging assessments across neurosurgery, orthopedics, and dentistry.[23\u0026ndash;25]\u003c/p\u003e \u003cp\u003eThe application of CNN-based models for automatic upper airway segmentation has facilitated a more efficient delineation workflow. Consequently, a growing body of research has been dedicated to exploring and validating these methodologies within craniofacial imaging. Fernanda Nogueira-Reis et al[26] developed an automatic framework integrating six 3D U-Nets to segment craniofacial structures, including the pharyngeal airway. Their model achieved superior performance to the semi-automatic approach, with a higher mean DSC (0.996 vs. 0.883) and markedly faster processing (1.1s vs. 48.4 min). Kaan Orhan et al.[27] developed a model for automatic upper airway segmentation, which was applicable to both OSA and non-OSA patients. The AI-based results showed high agreement with manual segmentations across key metrics, including the airway's narrowest point, cross-sectional area, and volume. Dong-Yul Kim et al.[9] employed a 3D U-Net for fully automatic pharyngeal segmentation, assessing subregional volume changes following orthognathic surgery in Class III patients. The model achieved a mean DSC of 0.89, demonstrating strong concordance with manual semi-automatic segmentations on both preoperative and postoperative CBCT images. Additionally, their comparative performance analysis revealed that segmentation accuracy was higher for preoperative nasopharyngeal volumes than that in their postoperative counterparts. In this study, the model achieved a mean DSC of 0.900, a mean VS of 0.949, and a mean 95% HD below 2 mm, reflecting robust segmentation accuracy. Concurrently, the automatic segmentation approach demonstrated superior efficiency in processing time. Notably, segmentation performance remained consistent between preoperative and postoperative scans, likely due to the incorporation of both image types during training. Although the minimum DSC for oropharyngeal segmentation exceeded the accepted threshold for radiographic analysis (\u0026gt;\u0026thinsp;0.88), performance in this region was significantly lower than in the velopharynx. The observed reduction in segmentation accuracy may stem from the complex craniofacial anatomy characteristic of OSA patients with concurrent microgenia. In these patients. These individuals often present with an elongated soft palate that blends radiographically with the base of the tongue, obscuring the SPt landmark.[28,29] Additional anatomical variability is introduced by common variations in epiglottic morphology and position, which contribute to structural heterogeneity around the Eb landmark.[30,31]\u003c/p\u003e \u003cp\u003eSimilar to the report in previous literature, we found significant increases in velopharynx volume and oropharynx volume. A positive correlation was observed between the degree of chin advancement and the increase in oropharyngeal volume, which can be attributed to the anterior displacement of the genioglossus and the movement of the hyoid bone. Dos Santos Junior et al.[32] demonstrated that sliding genioplasty for genioglossus advancement significantly increased the posterior airway space and reduced the AHI from 12.4 to 4.4 in patients with mandibular retrognathia. Chen et al.[5] found that a modified advancement genioplasty significantly increased total airway volume by 1,528\u0026thinsp;\u0026plusmn;\u0026thinsp;638 mm\u0026sup3;, with the hypopharynx being the primary site of expansion, and observed concomitant anterior-superior displacement of the hyoid bone. Valls-Onta\u0026ntilde;\u0026oacute;n et al.[6] reported that aesthetic osseous genioplasty increased total upper airway volume by 9.89% (2817\u0026thinsp;\u0026plusmn;\u0026thinsp;7256 mm\u0026sup3;) through the upward and forward movements of the hyoid bone, with the most significant gain occurring at the oropharyngeal level by 16.8%. Kauke-Navarro et al.[19], in their systematic review, synthesized evidence indicating that isolated genioplasty leads to an average total airway volume increase of approximately 8.5% (1511 mm\u0026sup3;) and a mean posterior airway space gain of 2.9 mm, correlating with improvements in AHI and snoring severity.\u003c/p\u003e \u003cp\u003eThis study has some limitations. First, this study was limited by its reliance on CBCT scans acquired from a single center and a single scanner model under relatively standardized acquisition protocols. Consequently, the generalizability of the developed model to heterogeneous multi-center datasets obtained from different CBCT devices with varying imaging parameters requires further validation. Expanding the sample size to include more diverse imaging sources would strengthen the model's robustness. Second, deep learning models function as \"black boxes\", which lake intuitive visualization of the decision-making process underlying segmentation\u0026mdash;for instance, the specific image features that determine airway boundary delineation. Further work should integrate explainable AI techniques, such as gradient-weighted class activation mapping, to improve model interpretability and enhance greater clinical trust. Additionally, the anatomical landmarks used to define the upper airway can vary depending on patient-specific craniofacial morphology and the site of obstruction, particularly in individuals with microgenia. Since the quality of ground truth in the training dataset is inherently dependent on clinician expertise, assessing inter-examiner reliability among multiple experienced raters would help quantify annotation consistency and further enhance model reliability.[20]\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrated the potential of a SegResNet-based CNN model for quantitatively assessing subregional upper airway changes following genioplasty, with robust segmentation of the velopharynx and oropharynx in CBCT images of OSA patients with microgenia supporting this conclusion. The results revealed significant increases in both velopharyngeal and oropharyngeal volumes after advancement sliding genioplasty. Future research should focus on expanding the sample size by including data from multiple centers and different CBCT devices, validating the model's generalizability with the aim of developing a more efficient analytical tool.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe appreciate the editor and reviewers for their valuable feedback, which has contributed to improving this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Basic Public Welfare Research Project of Zhejiang Province, China (grant no. LGF20H140004).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors consent to publicate the manuscript in COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCRediT authorship contribution statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJiaqi Zheng: Writing\u0026nbsp;–\u0026nbsp;original draft, Visualization, Software, Formal analysis, Data curation. Yupeng Ruan: Visualization, Investigation, Validation, Data curation. Xirui Wang: Visualization, Investigation, Validation. Jianhua Liu: Supervision, Resources, Project administration. Tingwei Bao: Writing\u0026nbsp;–\u0026nbsp;review \u0026amp; editing, Supervision, Resources, Project administration, Methodology, Funding acquisition.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analysed during the current study are not publicly available due [The clinical data of all participants in this study belongs to The First Affiliated Hospital, Zhejiang University School of Medicine and we need to obtain the approval of the hospital's medical department when obtaining it] but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study was approved by the Ethical Committee of The First Affiliated Hospital, Zhejiang University School of Medicine (Approve number: [2025B] IIT Ethics Approval No.1101. Approve date: 2025.09.23). All methods were carried out in accordance with relevant guidelines and regulations. Written informed consent was obtained from all the patients above 18 years old.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWritten informed consent for publication of identifying images or other personal or clinical details was obtained from all the patients above 18 years old.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eR. Rojas, R. Chateau, C. Gaete, C. Mu\u0026ntilde;oz, Genioglossus muscle advancement and simultaneous sliding genioplasty in the management of sleep apnoea, Int. J. Oral Maxillofac. Surg. 47 (2018) 638\u0026ndash;641. https://doi.org/10.1016/j.ijom.2017.10.008.\u003c/li\u003e\n \u003cli\u003eR. Kochhar, V. Modi, N. de Silva, S. Gueye-Ndiaye, J.M. Neugarten, M.J. Ward, L.M. Gerber, A. An, E. Mauer, H. Veler, Polysomnography-guided mandibular distraction osteogenesis in Pierre Robin sequence patients, J. Clin. Sleep Med. JCSM Off. Publ. Am. Acad. Sleep Med. 18 (2022) 1749\u0026ndash;1755. https://doi.org/10.5664/jcsm.9960.\u003c/li\u003e\n \u003cli\u003eC. Hansen, L. Sonnesen, M. Bakke, A. Markstr\u0026ouml;m, Prevalence of sleep-disordered breathing in children and adolescents with large horizontal maxillary overjet due to mandibular retrognathia: a case-control study, J. Clin. Sleep Med. JCSM Off. Publ. Am. Acad. Sleep Med. 20 (2024) 1871\u0026ndash;1878. https://doi.org/10.5664/jcsm.11248.\u003c/li\u003e\n \u003cli\u003eM. Shimatsu, S. Kawashima, M. Suzuki, Evaluating Craniofacial Morphology Ratios as Predictors of Obstructive Sleep Apnea Severity in Non-Obese Adult Males, Dent. J. 12 (2024) 374. https://doi.org/10.3390/dj12120374.\u003c/li\u003e\n \u003cli\u003eK. Chen, X. Sun, L. Wang, S. Ni, Z. Wu, Y. Gao, X. Wang, K. Xie, S. Liu, G. Wu, A Modified Cosmetic Genioplasty Can Affect Airway Space Positively in Skeletal Class II Patients: Studying Alterations of Hyoid Bone Position and Posterior Airway Space, Aesthetic Plast. Surg. 44 (2020) 1639\u0026ndash;1655. https://doi.org/10.1007/s00266-020-01790-3.\u003c/li\u003e\n \u003cli\u003eA. Valls-Onta\u0026ntilde;\u0026oacute;n, S. Arjooca, O. Saavedra, M. Giralt-Hernando, J. Asensio-Salazar, O.L. Haas Jr., F. Hern\u0026aacute;ndez-Alfaro, Does Aesthetic Osseous Genioplasty Impact Upper Airway Volume?, Aesthet. Surg. J. 44 (2024) 354\u0026ndash;362. https://doi.org/10.1093/asj/sjad341.\u003c/li\u003e\n \u003cli\u003eH.-N. Cho, E. Gwon, K.-A. Kim, S.-H. Baek, N. Kim, S.-J. Kim, Accuracy of convolutional neural networks-based automatic segmentation of pharyngeal airway sections according to craniofacial skeletal pattern, Am. J. Orthod. Dentofacial Orthop. 162 (2022) e53\u0026ndash;e62. https://doi.org/10.1016/j.ajodo.2022.01.011.\u003c/li\u003e\n \u003cli\u003eY. S\u0026uuml;k\u0026uuml;t, E. Yurdakurban, G.S. Duran, Accuracy of deep learning-based upper airway segmentation, J. Stomatol. Oral Maxillofac. Surg. 126 (2025) 102048. https://doi.org/10.1016/j.jormas.2024.102048.\u003c/li\u003e\n \u003cli\u003eD.-Y. Kim, S. Woo, J.-Y. Roh, J.-Y. Choi, K.-A. Kim, J.-Y. Cha, N. Kim, S.-J. Kim, Subregional pharyngeal changes after orthognathic surgery in skeletal Class III patients analyzed by convolutional neural networks-based segmentation, J. Dent. 135 (2023) 104565. https://doi.org/10.1016/j.jdent.2023.104565.\u003c/li\u003e\n \u003cli\u003eT. Yang, G. Zhu, L. Cai, J.H. Yeo, Y. Mao, J. Yang, A benchmark study of convolutional neural networks in fully automatic segmentation of aortic root, Front. Bioeng. Biotechnol. 11 (2023) 1171868. https://doi.org/10.3389/fbioe.2023.1171868.\u003c/li\u003e\n \u003cli\u003eA. Cariola, E. Sibilano, A. Guerriero, V. Bevilacqua, A. Brunetti, Deep learning strategies for semantic segmentation of pediatric brain tumors in multiparametric MRI, Sci. Rep. 15 (2025) 22595. https://doi.org/10.1038/s41598-025-07257-2.\u003c/li\u003e\n \u003cli\u003eA. Diaz-Pinto, S. Alle, V. Nath, Y. Tang, A. Ihsani, M. Asad, F. P\u0026eacute;rez-Garc\u0026iacute;a, P. Mehta, W. Li, M. Flores, H.R. Roth, T. Vercauteren, D. Xu, P. Dogra, S. Ourselin, A. Feng, M.J. Cardoso, MONAI Label: A framework for AI-assisted interactive labeling of 3D medical images, Med. Image Anal. 95 (2024) 103207. https://doi.org/10.1016/j.media.2024.103207.\u003c/li\u003e\n \u003cli\u003eW. Huang, H. Lai, Effect of exercise versus mandibular advancement device in moderate obstructive sleep apnea patients with mandibular retrognathia: A randomized clinical trial, Cranio J. Craniomandib. Pract. 43 (2025) 948\u0026ndash;957. https://doi.org/10.1080/08869634.2024.2401643.\u003c/li\u003e\n \u003cli\u003eG.E. Manrikyan, I.F. Vardanyan, M.M. Markaryan, M.E. Manrikyan, E.H. Badeyan, A.H. Manukyan, M.A. Gevorgyan, S.G. Khachatryan, Association between the Obstructive Sleep Apnea and Cephalometric Parameters in Teenagers, J. Clin. Med. 12 (2023) 6851. https://doi.org/10.3390/jcm12216851.\u003c/li\u003e\n \u003cli\u003eI. Kobayashi, A. Perry, J. Rhymer, B. Wuyam, P. Hughes, K. Murphy, J.A. Innes, J. McIvor, A.D. Cheesman, A. Guz, Inspiratory coactivation of the genioglossus enlarges retroglossal space in laryngectomized humans, J. Appl. Physiol. Bethesda Md 1985 80 (1996) 1595\u0026ndash;1604. https://doi.org/10.1152/jappl.1996.80.5.1595.\u003c/li\u003e\n \u003cli\u003eA. Malhotra, J. Trinder, R. Fogel, M. Stanchina, S.R. Patel, K. Schory, D. Kleverlaan, D.P. White, Postural Effects on Pharyngeal Protective Reflex Mechanisms, Sleep 27 (2004) 1105\u0026ndash;1112. https://doi.org/10.1093/sleep/27.6.1105.\u003c/li\u003e\n \u003cli\u003eW.S. Mezzanotte, D.J. Tangel, D.P. White, Waking genioglossal electromyogram in sleep apnea patients versus normal controls (a neuromuscular compensatory mechanism), J. Clin. Invest. 89 (1992) 1571\u0026ndash;1579. https://doi.org/10.1172/JCI115751.\u003c/li\u003e\n \u003cli\u003eB.K. Choi, I.S. Yun, Y.S. Kim, T.S. Roh, S.E. Park, J.Y. Bae, B.K. Jung, Effects of Hat-Shaped Mortised Genioplasty with Genioglossus Muscle Advancement on Retrogenia and Snoring: Assessment of Esthetic, Functional, and Psychosocial Results, Aesthetic Plast. Surg. 43 (2019) 412\u0026ndash;419. https://doi.org/10.1007/s00266-018-1290-z.\u003c/li\u003e\n \u003cli\u003eM. Kauke-Navarro, L. Knoedler, L. Stucki, F.J. Klimitz, O. Allam, S. Knoedler, M.C. Navarro, P. Flores-P\u0026eacute;rez, A.L. Rancu, M. Heiland, M. Alperovich, A.-F. Safi, Osseous Genioplasty and Its Impact on Airway Volume\u0026mdash;What is the Evidence?, Aesthetic Surg. J. Open Forum 7 (2025) ojaf059. https://doi.org/10.1093/asjof/ojaf059.\u003c/li\u003e\n \u003cli\u003eS. Jin, H. Han, Z. Huang, Y. Xiang, M. Du, F. Hua, X. Guan, J. Liu, F. Chen, H. He, Automatic three-dimensional nasal and pharyngeal airway subregions identification via Vision Transformer, J. Dent. 136 (2023) 104595. https://doi.org/10.1016/j.jdent.2023.104595.\u003c/li\u003e\n \u003cli\u003eT. Pham, C.-K. Lin, D. Leek, R. Chandrashekhar, T.M. Annaswamy, Obstructive sleep Apnea\u0026rsquo;s association with the cervical spine abnormalities, posture, and pain: a systematic review, Sleep Med. 75 (2020) 468\u0026ndash;476. https://doi.org/10.1016/j.sleep.2020.09.008.\u003c/li\u003e\n \u003cli\u003eB. Solow, S. Skov, J. Ovesen, P.W. Norup, G. Wildschi\u0026oslash;dtz, Airway dimensions and head posture in obstructive sleep apnoea, Eur. J. Orthod. 18 (1996) 571\u0026ndash;579. https://doi.org/10.1093/ejo/18.6.571.\u003c/li\u003e\n \u003cli\u003eS. Cepeda, R. Romero, L. Luque, D. Garc\u0026iacute;a-P\u0026eacute;rez, G. Blasco, L.T. Luppino, S. Kuttner, O. Esteban-Sinovas, I. Arrese, O. Solheim, L. Eikenes, A. Karlberg, \u0026Aacute;. P\u0026eacute;rez-N\u0026uacute;\u0026ntilde;ez, O. Zanier, C. Serra, V.E. Staartjes, A. Bianconi, L.F. Rossi, D. Garbossa, T. Escudero, R. Hornero, R. Sarabia, Deep learning-based postoperative glioblastoma segmentation and extent of resection evaluation: Development, external validation, and model comparison, Neuro-Oncol. Adv. 6 (2024) vdae199. https://doi.org/10.1093/noajnl/vdae199.\u003c/li\u003e\n \u003cli\u003eJ. Shapey, A. Kujawa, R. Dorent, G. Wang, A. Dimitriadis, D. Grishchuk, I. Paddick, N. Kitchen, R. Bradford, S.R. Saeed, S. Bisdas, S. Ourselin, T. Vercauteren, Segmentation of vestibular schwannoma from MRI, an open annotated dataset and baseline algorithm, Sci. Data 8 (2021) 286. https://doi.org/10.1038/s41597-021-01064-w.\u003c/li\u003e\n \u003cli\u003eE. Yurdakurban, Y. S\u0026uuml;k\u0026uuml;t, G.S. Duran, Assessment of deep learning technique for fully automated mandibular segmentation, Am. J. Orthod. Dentofac. Orthop. Off. Publ. Am. Assoc. Orthod. Its Const. Soc. Am. Board Orthod. 167 (2025) 242\u0026ndash;249. https://doi.org/10.1016/j.ajodo.2024.09.006.\u003c/li\u003e\n \u003cli\u003eF. Nogueira-Reis, N. Morgan, I.R. Suryani, C.P.M. Tabchoury, R. Jacobs, Full virtual patient generated by artificial intelligence-driven integrated segmentation of craniomaxillofacial structures from CBCT images, J. Dent. 141 (2024) 104829. https://doi.org/10.1016/j.jdent.2023.104829.\u003c/li\u003e\n \u003cli\u003eK. Orhan, M. Shamshiev, M. Ezhov, A. Plaksin, A. Kurbanova, G. \u0026Uuml;nsal, M. Gusarev, M. Golitsyna, S. Aksoy, M. Mısırlı, F. Rasmussen, E. Shumilov, A. Sanders, AI-based automatic segmentation of craniomaxillofacial anatomy from CBCT scans for automatic detection of pharyngeal airway evaluations in OSA patients, Sci. Rep. 12 (2022) 11863. https://doi.org/10.1038/s41598-022-15920-1.\u003c/li\u003e\n \u003cli\u003eS. Navasumrit, Y.-A. Chen, Y.-J. Hsieh, C.-F. Yao, C.-S. Chang, N.-H. Chen, Y.-F. Liao, Y.-R. Chen, Skeletal and upper airway stability following modified maxillomandibular advancement for treatment of obstructive sleep apnea in skeletal class I or II deformity, Clin. Oral Investig. 26 (2022) 3239\u0026ndash;3250. https://doi.org/10.1007/s00784-021-04306-8.\u003c/li\u003e\n \u003cli\u003eT. Jiang, Y. Qi, Y. Zhang, Y. Du, Q. Wu, H. Xiao, K. He, J. Zheng, Z. Jin, F. Li, Changes in upper airway morphology and respiratory function of adolescent patients with mandibular retrognathism treated with clear aligner mandibular advancement: a prospective study, BMC Oral Health 25 (2025) 975. https://doi.org/10.1186/s12903-025-06284-9.\u003c/li\u003e\n \u003cli\u003eI.-C. Kuo, L.-J. Hsin, L.-A. Lee, T.-J. Fang, M.-S. Tsai, Y.-C. Lee, S.-C. Shen, H.-Y. Li, Prediction of Epiglottic Collapse in Obstructive Sleep Apnea Patients: Epiglottic Length, Nat. Sci. Sleep 13 (2021) 1985\u0026ndash;1992. https://doi.org/10.2147/NSS.S336019.\u003c/li\u003e\n \u003cli\u003eX. Wang, H. Chen, L. Jia, X. Xu, J. Guo, The relationship between three-dimensional craniofacial and upper airway anatomical variables and severity of obstructive sleep apnoea in adults, Eur. J. Orthod. 44 (2022) 78\u0026ndash;85. https://doi.org/10.1093/ejo/cjab014.\u003c/li\u003e\n \u003cli\u003eJ.F. dos Santos Junior, M. Abrah\u0026atilde;o, L.C. Greg\u0026oacute;rio, A.I. Zonato, E.H. Gumieiro, Genioplasty for genioglossus muscle advancement in patients with obstructive sleep apnea-hypopnea syndrome and mandibular retrognathia, Braz. J. Otorhinolaryngol. 73 (2015) 480\u0026ndash;486. https://doi.org/10.1016/S1808-8694(15)30099-9.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Convolutional neural network (CNN), Cone-beam computed tomography (CBCT), Obstructive sleep apnea (OSA), Microgenia, Genioplasty, Upper airway","lastPublishedDoi":"10.21203/rs.3.rs-8567392/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8567392/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e: Obstructive sleep apnea (OSA) is strongly linked to upper airway anatomical compromise, with mandibular retrognathia being a risk factor. Genioplasty is commonly performed for both aesthetic and functional enhancement. Convolutional neural networks (CNNs) enable reliable segmentation of cone-beam computed tomography (CBCT) images. This study aimed to evaluate AI-based upper airway segmentation from CBCT in OSA patients who underwent genioplasty.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: A total of 170 CBCT images were utilized, divided into a training/validation set (n=110) and a test set (n=60). The test set consisted of 30 matched preoperative(T0) and postoperative(T1) image pairs from OSA patients with microgenia who underwent advancement sliding genioplasty. A SegResNet CNN model was employed for fully AI-based segmentation of subregional upper airway volumes, with performance assessed via dice similarity coefficient (DSC), volume similarity (VS), and 95 percentile Hausdorff Distance (95% HD). Correlations between clinical indicators, volume changes, and model metrics were analyzed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: The model exhibited a mean DSC value of 0.900-0.907, a mean VS value of 0.949-0.950 and a mean 95%HD of 1.485-1.588. Postoperatively, both subregions showed significant volume increases (velopharynx: 8888.19 ± 3106.34 vs. 10615.96 ± 3501.67; oropharynx: 6330.92 ±3218.49 vs. 7905.11 ± 4413.17, p\u0026lt;0.05), and oropharyngeal expansion weakly correlated with chin advancement magnitude.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions\u003c/strong\u003e: The present SegResNet-based model achieved fast and accurate upper airway segmentation from pre- and postoperative CBCT scans of OSA patients underwent genioplasty, establishing a basis for developing efficient analytical models to predict surgical outcomes for OSA patients.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical trial number\u003c/strong\u003e: not applicable.\u003c/p\u003e","manuscriptTitle":"Artificial intelligence-based upper airway segmentation for evaluating volume changes following genioplasty in patients with obstructive sleep apnea","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-09 12:10:49","doi":"10.21203/rs.3.rs-8567392/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a7ebd82a-0601-48ce-a679-f1f196e15e69","owner":[],"postedDate":"February 9th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-09T12:12:24+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-09 12:10:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8567392","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8567392","identity":"rs-8567392","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}