An Iterative Framework for Improving the Accuracy of Intraoperative Intensity-Based 2D/3D Registration for Image-Guided Orthopedic Surgery

摘要

We propose an iterative refinement framework that improves the accuracy of intraoperative intensity-based 2D/3D registration. The method optimizes both the extrinsic camera parameters and the object pose. The algorithm estimates the transformation between the fiducials and the patient intraoperatively using a small number of X-ray images. The proposed algorithm was validated in an experiment using a cadaveric phantom, in which the true registration was acquired from CT data. The results of 50 registration trials with randomized initial conditions on a pair of X-ray C-arm images taken at 32° angular separation showed that the iterative refinement process improved the translational error by 0.32 mm and the rotational error by 0.61 degrees when compared to the 2D/3D registration without iteration. This tool has the potential to allow routine use of image guided therapy by computing registration parameters using only two X-ray images.