Automatic 3D Geometry Reconstruction from Image Data
Dagmar Kainmueller, Stefan Zachow, Hans-Christian Hege (Zuse Institute Berlin)

In the Medical Planning group at Zuse Institute Berlin, we developed a general framework for accurate, fully automatic segmentation of anatomical objects from tomographic image data.  Firstly, a global initialization procedure roughly places a Statistical Shape Model of the respective object in the image data. After initialization, the Statistical Shape Model is adapted to the image data. Finally, a free form adaptation is performed to achieve fine grain segmentations that are not restricted by the shape space of the Statistical Shape Model. Model adaptation is guided by heuristic intensity models.

We will demonstrate the performance of our fully automatic segmentation framework for various tissues and imaging modalities, including the liver in contrast enhanced CT, the pelvic bones in conventional CT, and the mandibular bone and nerves in Cone-Beam CT. Particularly, we will present a fully automatic method for accurate fine grain segmentation of bones in joint regions, which is a prerequisite for generating patient specific biomechanical models, e.g. of the human lower limb.