3D Vessel Tree Reconstruction from Rotational C-arm Projections by Multi-view Stereo Reconstruction

Abstract

In biplane angiography, a three-dimensional (3D) vessel tree can be reconstructed from two-view stereo reconstruction. During an intervention procedure, the rotational C-arm scanning produces a sequence of cone-beam projections (angiograms) that can be used for vessel tracking and vessel tree reconstruction. Since any two projection images can be considered as a two-view pair as captured by a biplane system, a 3D vessel tree can be reconstructed by searching the corresponding points in two projection images followed by two-view stereo reconstruction. The geometry of the 3D tree can be measured in the calibrated Euclidean space. During C-arm angiographic imaging, the contrast bolus flowing through blood vessels may sustain a bolus/blood equilibrium state in the scan field of view, which can be considered as quasi-static or stationary state, the C-arm images can be used for vessel tree reconstruction. In this paper, we propose a vessel tree reconstruction method from C-arm images: 1) add epipolar constraint on the search for corresponding points (feature points); 2) apply depth-first search strategy for tree construction from node points and point correspondence; 3) find fundamental and projection matrices for projective reconstruction; 4) reconstruct the 3D tree in Euclidean space with a cube phantom calibration; 5) perform 3D tree measurements (tree segment lengths). Computer simulations and tree phantom experiments under a rotational C-arm system are reported, which justify the method in this paper.