Exhaustive graph cut-based vasculature reconstruction

We propose an exhaustive extension to graph cut-based coronary artery reconstruction from multiple views of a rotational angiography sequence. The reconstruction is formulated as an energy minimization problem that is solved using the a-expansion algorithm. We enforce reprojection-based data consistency and completeness conditions on the reconstructed centerline. The proposed strategy omits the need for selection and manual refinement of a reference view. A phantom study is used to assess the performance in 2D and 3D. The average reprojection error decreased from 1.32 ± 0.99mm to 0.54 ± 0.02 mm. Moreover, an increase in Dice score from 0.50 ± 0.04 mm to 0.59 ± 0.00 mm was observed, indicating superior volumetric reconstruction quality. The results suggest that the proposed extension removes the susceptibility to reference frame selection and manual interaction, while increasing reconstruction quality.