Global registration of multiple bone fragments using statistical atlas models: feasibility experiments.

We propose a new technique to automatically register multiple bone fragments of a fracture using a global registration method guided by a statistical anatomical atlas model. The atlas is created by using the principal component analysis of three-dimensional meshes generated from the Computed Tomography (CT) data of five left femur bone cadavers. The fracture fragments are initially aligned to the generated atlas by using a local point descriptor which can robustly identify a set of potential corresponding points among fragments and the mean atlas model. A global registration algorithm is then utilized to fine-tune the alignment among the fractures and the mean atlas model. Performed experimental simulations on the human femur bone cadavers verify the feasibility of the proposed registration algorithm.