Interventional 2D-3D Registration in the Presence of Occlusion

Abstract

It is crucial that 2D-3D medical image registration algorithms meet special requirements in terms of accuracy and robustness in order to be applied during clinical interventions. Existing algorithms may be affected by image dissimilarities introduced by medical instruments visible only in the interventional 2D image. Based on our previous results, we present a fully automatic framework for disocclusion-based 2D-3D registration technique that detects the occlusion and performs a matching on the reconstructed 2D image. Compared to earlier approaches, the proposed algorithm is fully automatic and therefore highly appropriate for clinical application. Our technique is validated on synthetic and real interventional data and compared with conventional methods. Results prove that disocclusion-based registration yield higher accuracy and robustness and outperforms existing approaches in terms of speed.