3-D image reconstruction of brain vessels from angiograms

This paper presents a reconstruction process to build the three-dimensional cerebral vessel tree from a pair of DSAs (digital subtraction angiograms). To segment the blood vessels from the background, the local thresholding and global thresholding was used to bilevel the input image. Thinning is applied to obtain the skeleton of the vessels. The B-spline curve is used to smooth the skeleton. The skeleton is then tracked to find the branch points of the skeleton and used as the feature points for matching. The radius of the vessel was detected by the matched filter concept. For each feature point in the first image, we use the epipolar constraint and the property of feature points to find its correspondence on the other image. From the DSA machine structure, the data provided by the machine and the correspondence information, the least squares estimate of the real 3D skeleton position is obtained. By using the Gouraud shading method and the radius information, we display the reconstructed 3D shading images for various viewing angles and light source directions.