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

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 degrees 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.