Deep learning-assisted cone-beam computed tomographic analysis of condylar changes after mandibular setback surgery.

This study aimed to assess condylar changes using a fully automated deep learning-based cone-beam computed tomography (CBCT) workflow. Preoperative and postoperative CBCT scans of 50 skeletal Class III patients (100 condyles) were analysed using a fully automated pipeline integrating nnU-Net-based segmentation, rigid surface registration, and standardised surface cropping. Condylar changes were quantified using volumetric and linear measurements and surface-based metrics. Segmentation accuracy was high (Dice: mandible 0.98, condyle 0.99). Mean (SD) condylar volume changes ranged from -12.3 (6.2) to -0.03 (11.1) mm3 on the left and from -11.3 (10.7) to -0.95 (12.6) mm3 on the right. Significant differences in inter-side volume were observed in left and right rotation groups (p = 0.003), but not in the non-rotation group (p = 0.442). Direction of mandibular rotation significantly affected change in condylar volume bilaterally (p = 0.039). Surface-based metrics differed significantly among rotation groups (p = 0.036). Change in condylar volume showed a negative correlation with preoperative volume (r = -0.44 to -0.77, p < 0.001). Condylar remodelling after mandibular setback surgery is rotation-dependent and regionally heterogeneous. The proposed automated CBCT-based workflow enables reproducible, operator-independent quantification of condylar changes, and provides a standardised framework for postoperative assessment.