P-67686931-185 - A craniofacial statistical shape model for the virtual reconstruction of bilateral maxillary defects.

Bilateral maxillary defects (whether traumatic, surgical, or congenital in origin) challenge fibula free flap (FFF) reconstruction surgery due to limitations in virtual surgery planning (VSP) workflows. While segmented 3D meshes of unilateral maxillary defects can be mirrored to virtually reconstruct missing anatomy, bilateral defects (Brown class c and d) lack a contralateral reference and associated anatomical landmarks. This often results in poor placement of osteotomized fibula segments. The current study improves the VSP workflow for FFF reconstructions using statistical shape modelling (SSM) – a form of unsupervised machine learning – to virtually reconstruct premorbid anatomy in an automated, reproducible, and patient-specific manner. The SSM training set was sourced from the New Mexico Decedent Image Database via stratified random sampling to ensure even age and sex distributions. Skulls from 104 computed tomography scans were segmented and rigidly aligned, after which principal component (PC) analysis was applied to construct a discrete Gaussian process (GP). Defect reconstruction was accomplished on a validation set of 8 skulls via an iterative closest point algorithm with GP regression. Preliminary analysis shows that Brown class 2c maxillectomy defects can be virtually reconstructed with promising accuracy [Hausdorff distance = 7.89±1.63 mm, volumetric Dice coefficient = 99.1±0.01%, compactness = 7.22×105 mm2 (over the first 97 PC), specificity = 1.18 mm, and generality = 1.06×10−5 mm]. SSM-guided VSP will allow surgeons to create patient-centric treatment plans, increasing reconstruction accuracy and reducing the risk of complications. This is expected to improve post-operative outcomes. [ABSTRACT FROM AUTHOR]