Introduction
Predicting soft tissue changes in the maxillofacial region after orthognathic surgery still remains a challenge due to the complexity of soft tissue dynamics. This study aimed to develop a deep learning (DL) model using CBCT images of ske...
Introduction
Predicting soft tissue changes in the maxillofacial region after orthognathic surgery still remains a challenge due to the complexity of soft tissue dynamics. This study aimed to develop a deep learning (DL) model using CBCT images of skeletal Class III malocclusion patients to predict three-dimensional (3D) changes in facial soft tissue following orthognathic surgery.
Materials and Methods
A total of 70 skeletal Class III malocclusion patients without significant asymmetry were included in this study. Using 3D Slicer, 3D facial soft tissue meshes were reconstructed from CBCT images obtained preoperatively and 1 year postoperatively. Soft tissue curvatures were simplified to generate 3D coordinate data, which, combined with the orthognathic surgery plan, served as input for the DL model. Of the total cohort, 64 patients were randomly assigned to the training group, where the DL model was designed and trained. The remaining 6 patients were allocated to the test group to assess the model’s prediction performance.
Results
A rendering engine was used to superimpose 3D facial meshes for visual validation. The comparison between preoperative and estimated meshes revealed similar changes, particularly in mandibular setback, despite slight differences in displacement, while postoperative and estimated meshes demonstrated high similarity. Analysis of 3D coordinate changes at facial soft tissue landmarks exhibited the highest similarity in vector changes at and Pog' and Me’, and lower similarity at Ala_L, Sn, and Sto. Vector distance analysis showed significant differences at specific landmarks at Sl. Overall, the DL model produced results closely matching actual outcomes, particularly in vertical facial height, nasal structure, and lateral aesthetic assessments based on anthropometric evaluation.
Conclusion
This study demonstrated the feasibility of incorporating a DL model with CBCT images in oral and maxillofacial surgery to predict three-dimensional facial soft tissue changes during orthognathic surgery, guided by the surgical plan. While the model achieved satisfactory overall accuracy, certain limitations were identified in regions characterized by detailed curvatures. Nevertheless, the model exhibited significant potential for visualizing surgical outcomes and supporting the development of patient-centered treatment. Continuous refinement of the model and the expansion of training data are anticipated to further enhance its performance and clinical applicability.