Seeing structure: using knowledge to reconstruct and illustrate anatomy

Abstract

Current medical imaging technology makes it possible to gather remarkably detailed three-dimensional data about an individual's anatomy. In domains ranging from education to clinical medicine, a common desire is the ability to examine selected structures from such volume datasets. This dissertation describes tools for performing the two key tasks in that process: reconstructing (or segmenting) specific structures from volume data and illustrating them in meaningful ways.On the reconstruction side, this work offers new, in-depth analysis of two previously proposed methods for using shape knowledge to guide image segmentation. The ideas are generalized to create a 3D shape model, which is used as part of a novel algorithm for semi-automatic segmentation of the brain. Unlike other methods, this approach offers intuitive user controls and explicitly addresses the removal of the skull and other surrounding structures. This method is incorporated into a working, interactive system for recording and studying functional data from the human brain. On the illustration side, several realworld situations are used to demonstrate how non-standard rendering methods can enhance the clarity of anatomical illustrations. The lessons learned from these examples lead to requirements for and a prototype of a medical illustration system.