Construction of a Patient-Specific 3D Symmetry Facial Mask for Planning and Assessment in Maxillofacial Surgery

Facial asymmetry poses a complex challenge in maxillofacial surgery, where aesthetics, function, and individual anatomy must be reconciled through precise planning and outcome evaluation. While most current methods for symmetry analysis have leveraged machine learning techniques, these approaches rely on large annotated datasets and often lack anatomical interpretability, making them unsuitable for patient-specific clinical use or in case of significant asymmetries. In this context, the presented work addresses this gap by introducing a computational mesh-based pipeline to construct a patient-specific symmetric facial mask, enabling both the prediction of symmetric outcomes and both the qualitative and quantitative assessment of surgical corrections, applicable even with limited data. The dataset comprises 15 patients affected by facial asymmetry, grouped according to data availability: (1) pre and postoperative CT scans (with surgical planning available in some cases), (2) only preoperative CT, and (3) only postoperative CT. From each CT, soft and hard tissue segmentations were obtained, and outer shells were extracted as triangulated 3D meshes. The core of the work focuses on the generation of a symmetric soft tissue (ST) mask from the preoperative mesh, in case of group (1), or from postoperative mesh, in case of group (3). This is achieved by reflecting the original mesh across an external vertical plane (defined from the most lateral vertex), then aligning the mirrored mesh to the original using stable frontal landmarks. A local averaging strategy is implemented using three anatomical centroids (nasal, oral, and maxillofacial): for each vertex, a ray is projected from the corresponding centroid, and its intersection with the mirrored mesh is used to compute the midpoint, which defines the displacement to be applied to the vertices itself. When surgical planning is available, rigid specific regional transformations are applied to the mask to approximate planned bone displacements. The hard tissue (HT) mask is constructed by selecting and projecting fiducial points, including landmarks and generic points, radially onto the ST mesh to extract soft tissue thickness (STT) values, which are then re-applied to the symmetric soft mask to reconstruct a corresponding hard tissue shape. To assess the validity of the symmetric mask, asymmetry is quantitatively evaluated using the Asymmetry Index (AI) metric implemented in our research group, based on Swennen landmarks distances from orthogonal anatomical planes: AI is compared across preoperative, mask, and postoperative surfaces, providing a symmetry benchmark grounded in patient-specific geometry. To date, existing methods for predicting or evaluating facial symmetry outcomes in orthognathic surgery are predominantly based on machine learning or deep learning models, which require large annotated datasets and offer limited interpretability. This approach introduces a modular framework that operates directly on individual patient meshes, adapts to data availability, is suitable for small datasets and provides clinically meaningful outputs, supporting both pre-surgical simulation and post-surgical evaluation. Principal limitations include local geometric noise in complex regions such as the nose and mouth, the assumption of stable BMI for STT calculation and the use of rigid mask deformation; future work will focus on integrating elastic tissue models to enhance the soft-tissue outcome prediction following bone movements.