Improved Shape Modeling of Tubular Objects Using Cylindrical Parameterization

Statistical shape modeling is widely used for medical image segmentation and interpretation. The main problem in building a shape model is the construction of a pointwise correspondence between the training objects. Manually corresponding objects is a subjective and time consuming task. Fortunately, surface parameterization can be automated and it has been successfully used. Mostly, the objects are of spherical nature such that spherical parameterization can be employed. However, for tubular objects, this method falls short. In this paper, a cylindrical parameterization technique is proposed and compared to spherical parameterization. As an application, both methods are applied to establish correspondences for a set of tympani scali of human cochleas and the quality of the models built from these correspondences is assessed.