Shape Statistical Model-Based Regression for Predicting Anatomical Landmarks

Determining appropriate shapes and establishing corresponding relationships based on 3D models remains a significant challenge. This research employs shape statistical models and direct correspondence between two morphologies via support vector regression. The displacement of sampled points to the standard model, calculated through non-rigid registration, represents the shape state. Constructing shape statistical models based on the deformation fields of sampled points from different models and subsequently mapping high-dimensional cranial features using support vector regression allows for establishing relationships between different models. The study used 77 mandibles and maxillas for the validation. Predicting the mandible using the maxilla resulted in a 2.43mm average error, while predicting the maxilla using the mandible yielded a 1.95mm average error. It offers potential avenues for further research and applications in craniofacial reconstruction.