Application of orthogonal neural network to craniomaxillary reconstruction

Abstract

Traditionally, the orthopaedist, according to their past experience, reconstructs damaged area while the operation is in progress. This may prolong the operation and cause the wound to become infected. Most importantly it is difficult to precisely match the skeletal defect. A well-disciplined network of prediction re-fabricates the damaged area through automation. This research is based on the CT image file, which is the product of X-ray computed tomography (CT), and computes the skeletal positions around the damaged area through image processing and boundary detection. The skeletal positions are inputted into the orthogonal neural network and discipline the network so that it possesses the scattering characteristic of bone. The network then calculates skeletal positions in the damaged area and revises the former CT image file to rebuild a 3D model. Accordingly, in comparison with a manual sketch, the orthogonal neural network forecast is more geometrically precise. Moreover, the forecast satisfies the second order derivative, which is a continuous function, and the edge of the fabricated bone is therefore kept smoother.