Robustness and Accuracy of Feature-Based Single Image 2-D–3-D Registration Without Correspondences for Image-Guided Intervention

摘要: 2-D-to-3-D registration is critical and fundamental in image-guided interventions. It could be achieved from single image using paired point correspondences between the object image. The common assumption that such can readily established does not necessarily hold for guided Intraoperative clutter an imperfect feature extraction method may introduce false detection and, due to physics of X-ray imaging, 2-D features indistinguishable each other and/or obscured by anatomy causing features. These create difficulties establishing 3-D data points. In this paper, we propose accurate, robust, fast accomplish 2-D-3-D a without need presence detection. We formulate as maximum likelihood estimation problem, which then solved coupling expectation maximization with particle swarm optimization. proposed was evaluated phantom cadaver study. study, it subdegree rotation errors submillimeter in-plane ( X- Y plane) translation errors. both studies, outperformed state-of-the-art methods do use same accuracy global optimal uses correct correspondences.