The reconstruction of malformed or missing bony structures often requires a template for harvesting autologous bone, or for the design of prosthetic implants. In cases where healthy contralateral structures exist, these structures can be mirrored from tomographic data. However, in severe cases of hemifacial microsomia or craniostosis, for example, no foundation for surgical reconstruction is available. Hence, the aim of our work is to provide statistical 3D shape models of normally developed human anatomy for surgical planning. The models will be derived from a large set of anatomic structures, reconstructed from tomographic data. Statistical shape models are representing a mean shape in addition with all variations within a normal range being contained in the training set. Thus, they can serve as a basis for surgical reconstruction of distinct deformities.
In a first investigation we created a statistical shape model of a human mandible out of 25 CT data sets, as well as a shape model for normally developed infant skulls out of 23 MRI data sets. A model for the human orbits is going to be obtained as well. Via Principle Component Analysis (PCA), and by using a common surface parametrization, the major modes of variation are determined.
These eigenmodes together with the averaged surface model of the corresponding anatomy enable us to analyze characteristic shapes, and to reconstruct new shape occurences as well as all data from the training set by linear interpolation of all or just a few selected eigenmodes. Thus, a 3D shape model can serve as a template for surgery planning, by finding an optimal fit from any of its variations to a given malformed structure. Furthermore, it is investigated, whether or not 3D shape models are suited for automatic segmentation.
A statistical 3D shape model of human anatomy seems to be a valuable planning aid for surgical reconstruction of bone defects. This is especially true for severe cases of hemifacial microsomia, and for premature ossified cranial sutures of infants (craniosynostis), that may lead to skull deformities during the growth process. The surgical orbita reconstruction also takes profit of 3D shape models, since the reconstruction of thin structures from tomographic data is difficult. With a best matching candidate of the shape model, regarding size and shape of available bone, a surgeon gets a good mental perception of the reconstruction.
This paper won the “Best Podium Presentation” award at the CAS-Head International Conference in June 2005 in Berlin.