Three-dimensional quantification and assessment of facial asymmetry in patients undertaking maxillofacial surgeries

Facial asymmetry is defined as a clinically significant variation between the two halves of the face, perceived as an imbalance relative to a central midline axis. While perfect bilateral symmetry is rarely achieved, mild asymmetry is common. However, pronounced asymmetry can significantly affect facial and smile esthetics, leading to complications such as malocclusion, altered temporomandibular joint function, and psychosocial well-being. Therefore, thorough evaluation and precise treatment planning are crucial for optimal functional and aesthetic outcomes. In literature, various 2D and 3D methodologies are employed for facial asymmetry analysis and quantification. Various 2D imaging techniques, including posterior-anterior cephalograms, panoramic radiographs, and digital photography, have historically been used to quantify facial asymmetry through metrics like the Asymmetry Index (AI), Z-score, Facial Asymmetry (FA), along with angular and area assessments. Three-dimensional techniques, such as Cone-Beam Computed Tomography (CBCT), stereophotogrammetry and 3D scanners offer enhanced accuracy. These 3D methods utilize landmark-based approaches such as Asymmetry Index (AI), Angular Asymmetry (AIang3D) and Euclidean Distance Matrix Analysis (EDMA), or surface-based methods like original-mirror alignment algorithms (e.g., Procrustes Analysis, Iterative Closest Point) which quantify asymmetry via indicators like Root Mean Square (RMS) and Mean Absolute Deviation (MAD). Spatially Dense Anthropometric Masks, also provide a template-mapping strategy for detailed analysis. Recent advancements in Artificial Intelligence, particularly Deep Convolutional Neural Networks (CNNs), are increasingly applied for automated facial asymmetry evaluation and landmark detection. This work presents an asymmetry analysis based on CT scans from a database of 15 patients who underwent asymmetry correction via maxillofacial surgery at San Giovanni Battista Hospital in Turin. A preliminary image pre-processing pipeline was implemented, encompassing: segmentation of hard and soft tissues, mesh refinement, alignment of pre and post-operative meshes and manual selection of soft and hard tissue landmarks, based on Swennen’s definitions. Upon obtaining meshes with their respective landmarks, three orthogonal planes—mid-sagittal, coronal, and horizontal—were identified. The Asymmetry Index (AI) was then defined by calculating the difference of the distances of each bilateral landmark from these planes, while for central landmarks, the distance from the mid-sagittal plane was used. For comprehensive clinical insight, both a global asymmetry value (obtained by averaging individual AI values) and local AI values for specific soft tissue (upper, middle, lower, chin) and hard tissue (upper, maxilla, mandible, symphysis) regions were provided for each patient. From these, three indices were calculated to evaluate surgical efficacy: Relative Real Operation Improvement (ReOI), Relative Ideal Operation Improvement (IdOI), and Percentage Deviation (PDev), indicating actual improvement, ideal potential improvement, and percentage deviation of real from ideal correction, respectively. Furthermore, the ideal displacement (delta in mm) for each landmark, necessary to achieve perfect symmetry, was determined. Finally, to complete the quantification and classify asymmetries, Reyneke’s methodology was followed, determining facial inclination relative to the mid-sagittal plane through the calculation of specific angles.