Feasibility of CBCT-based deep learning for predicting 3D soft tissue changes after orthognathic surgery in skeletal...

This study aimed to evaluate the feasibility of a cone-beam computed tomography (CBCT)-based deep learning (DL) model for predicting three-dimensional (3D) soft tissue changes following orthognathic surgery in skeletal Class III patients. 3D facial soft tissue meshes of 70 patients were reconstructed from preoperative and 1-year postoperative CBCT data by segmenting the facial region and applying a hollowing process. Soft tissue surface curvatures were simplified to generate 3D coordinate data, which were combined with surgical parameters as inputs for the DL model. A total of 64 patients were used for training, and the remaining six patients were reserved an independent test set. The model demonstrated overall agreement between estimated and postoperative meshes, with relatively better performance in the mandibular setback region, particularly at Pog' and Me', and reduced accuracy in anatomically complex areas including Ala_L, Sn, and Sto. Vector distance analysis revealed region-dependent discrepancies, indicating that estimation precision varied according to local anatomical complexity. Despite the small, single-center cohort, this study supports the feasibility of CBCT-based DL approach for predicting postoperative 3D soft tissue changes. The proposed framework may facilitate visualization of patient-specific facial soft tissue changes based on planned skeletal movements, pending further validation in larger and more diverse populations.