Surface Modeling of Multiple Bone Objects By Staged Self-Organizing Map Neural Network

Abstract

In this paper the surface modeling of complex (ill-posed) bone objects by the self-organizing map (SOM) artificial neural network is introduced for the purpose of future 3-D surgical planning and 3-D/2-D registration in intra-operative fluoroscopic image guidance and monitoring of orthopedic surgery. Self-organizing map, an unsupervised neural network is initialized with the three-dimensional globe of grids. The 3-D point-clouds used by SOM network learning are obtained by delineating the interested bone outlines on each slice of MRI or CT images. Depending on the complexity of bone structure, each bone segment can be modeled by either one step or two step unsupervised neural network learning. The transformation of constructed 3-D bone models can be performed in 6 degree of freedom (DOF) plus scaling. Thus it is possible that the 3-D surgical planning can be executed in OR through 3-D/2-D registration by surgery monitoring and guidance.