Modeling the human aorta for MR-driven real-time virtual endoscopy.

As interventional magnetic resonance imaging (iMRI) is getting closer to clinical practice, new means of visualization and navigation are required. We present an approach to create a virtual endoscopic view inside the human aorta in real-time. In our approach, defined cross-sectional slices are acquired and segmented in a highly optimized fashion. A geometric shape model is fit to the segmentation points and continuously updated during the intervention. The physician can then view and navigate inside the structure to plan the intervention and get immediate feedback about the procedure. As a component of this system, this work focuses on the segmentation of the cross-sectional images and the fitting of the shape model. We present a real-time 2D segmentation implementation for this application domain and a model fitting scheme for a generalized cylinder (GC) model. For the latter we employ a new scheme for choosing the local reference frame.