Evaluation of the accuracy of surgical planning and development of a geometric-statistical model for the prediction of soft tissue in maxillofacial surgery

Craniofacial deformities, such as retrognathism and prognathism, can affect oral and respiratory function and negatively impact facial aesthetics. Orthognathic surgery represents the primary therapeutic solution because it allows for the three-dimensional repositioning of bone structures. In recent years, virtual surgical planning has significantly improved the precision and reproducibility of procedures, but discrepancies remain between preoperative predictions and postoperative results, especially in the vertical and anteroposterior directions. Furthermore, predicting soft tissue behavior remains a critical aspect for the aesthetic evaluation of surgical outcomes. This study, conducted in collaboration with the Department of Maxillofacial Surgery at San Giovanni Battista Le Molinette Hospital in Turin, has two main objectives: 1. To quantitatively analyze the discrepancies between planned and actual postoperative skeletal movements, assessing the accuracy and clinical significance of the planning; 2. To study soft tissue displacement following hard tissue displacement, developing and validating a geometric-statistical model based on Partial Least Squares (PLS) regression for the three-dimensional prediction of soft tissue displacement. For Aim 1, CT data were analyzed from 13 patients undergoing orthognathic surgery for retrognathic or prognathic malocclusions. Pre- and postoperative 3D models were reconstructed from DICOM images using 3DSlicer and MeshLab software and registered using rigid landmarks and alignment algorithms. Discrepancies between planning and actual results were calculated for each landmark in the three spatial directions (LR, AP, UD) and statistically analyzed using the t-test/Wilcoxon test, Bland-Altman analysis, and Cohen's effect size. The results indicate high planning accuracy in the latero-lateral and antero-posterior directions, with more significant errors in the vertical components, which are clinically relevant in specific anatomical areas. For Aim 2, a predictive workflow was developed based on morphometric similarity between patients and PLS regression. Similarity features were defined using standardized geometric measurements of preoperative soft tissue, and the model's response was represented by the actual postoperative displacements of landmarks and facial patches. Leave-One-Out Cross Validation demonstrated the model's good ability to estimate the direction and magnitude of soft tissue displacements, with average errors below 3 mm in key facial regions. Overall, the work proposes an integrated methodology for the quantitative and predictive analysis of orthognathic surgical outcomes and provides an innovative contribution towards more precise, reproducible and clinically reliable planning.