A new method of reconstructing three-dimensional (3D) tubular structures from cone-beam projections is proposed in this paper. Minimal path method is applied to detect 2D tubular structures in the cone-beam projections. The extracted 2D information are fused to design a weight function over the 3D volume so that minimal path method can be applied again to extract centerlines of 3D tubular structures. The validity of the proposed weight function relies on the sparsity of 3D images containing tubular structures. The procedure of estimating the radius of 3D tube from 2D projections is also described. The proposed method requires neither cone-beam reconstruction, nor tree matching, one of which is usually necessary in the existing reconstruction methods. The method is robust to small motions between different projections.
Jia Li, Laurent D. Cohen