Template-based surface reconstruction from cross-sections

In this work we explore a new approach to solving the problem of surface reconstruction from cross-section curves. Existing reconstruction methods focus on producing smooth interpolations and may require a large number of cross-sections to recreate feature-rich objects. In the context of medical image segmentation, we make the observation that anatomical structures usually share a common overall shape across subjects. This observation motivates taking a template-based approach that can better capture geometric features, instead of solving for the surface from the cross-sectional curves alone. We deform an existing template, whose shape represents the structure of interest, to fit a set of target curves. We describe our algorithm with the main focus on finding a correspondence between a mesh and a cross-section curve network. We show our results with real medical data and compare to current reconstruction methods.