Predicting the Premorbid Shape of a Diseased Mandible.

Objectives/hypothesis: Virtual surgical planning (VSP) for reconstructions of advanced mandibular neoplasms that have distorted the contour of the mandible is challenging, as the premorbid shape of the mandible is unknown. We introduce a novel modeling technique, based on a statistical shape model (SSM), that has learned the shape of a normal mandible from a set of 84 mandibles, such that given a diseased mandible, the model can determine its premorbid shape.

Methods: Eighty-four control mandibles were used to generate an SSM. Various mandibular defects were created, and the SSM was applied to predict the shape of the original mandible. The predicted and original shape of the defect were compared for accuracy using volumetric overlap and Hausdorff distance. All mandibular VSP cases in the past 2 years were reviewed to identify those that required virtual preprocessing due to significantly distorted mandibular contours. The SSM was compared to those cases requiring preprocessing and highlighted in one prospective VSP.

Results: The average volumetric overlap and Hausdorff distance between the defect replacement and the defect are 73.9% ± 13.3% and 4.51 mm ± 2.65 mm, respectively. The SSM is more accurate for smaller defects, and those not including the condyle. Ten out of 40 VSP cases required preprocessing using four different techniques. Qualitatively, the SSM outperformed those preprocessing techniques applied in the retrospective cases.

Conclusions: The SSM can accurately predict the premorbid shape of a distorted mandible and is superior to current preprocessing techniques. The SSM was successfully applied to a retrospective series and one prospective index case.

Level Of Evidence: 4 Laryngoscope, 131:E781-E786, 2021.