A reliable segmentation of the vertebral column is essential for numerous orthopedic, neurological, and oncological applications. In the context of computer aided spinal surgery, exact knowledge about the shape of individual vertebrae is of great importance, e.g., for spinal biopsies, implants, or the insertion of pedicle screw. The project was done in cooperation with Philips Research Europe - Hamburg.
The aim of this project is to generate a comprehensive model of the human spine. The prior knowledge covered in the model can then be used to provide an accurate segmentation of the vertebral column in CT images. From a technical point of view, the modeling of the vertebral column is of special interest as it extends the modeling of individual organs to the modeling of articulated objects. Although bony structures show high contrast in CT a reliable segmentation of the spine including a labeling of the individual vertebrae is challenging due to the similarity of neighbour structures and unclear object boundaries.
The generated model of the spine covers not only intensive shape information of the individual vertebrae but also statistical information of relative pose. Furthermore, appearance and gradient models provide an identification of vertebrae.