Reconstruction of the 3D Structure of Cerebral Vessel Trees from few DSA Projections

A new model based approach is presented which permits the 3D reconstruction of blood vessels with multiple branches from few DSA projections by iteratively projecting single truncated cones, local optimization of an appropriate cost function and with the usage of anatomical constraints. In contrast to voxel based methods [5] our algorithm produces and works on a symbolic object representation. Simulations on synthetic data sets and real projections of a MRA/DSA flux phantom confirmed the feasibility of our approach. As a simulation result it has turned out that usually not more than six accurate projections are necessary to reconstruct complex synthetic vessel structures comparable to real cerebral vessel subtrees. Due to inadequate DSA imaging devices difficulties arise from inaccuracies of the camera parameters and asynchronously taken projections. With the use of a fiducial marker system for the reconstruction of the camera parameters and image preprocessing, i.e. distortion correction and segmentation of the DSA images, these problems can be overcome.