Effective intensity-based 2D/3D rigid registration between fluoroscopic X-ray and CT

We present an effective intensity-based method for rigid registration of a patient preoperative CT to its intraoperative situation with a few fluoroscopic X-ray images obtained with a tracked C-arm. It improves upon existing methods and overcomes most of their intrinsic speed, robustness, and accuracy problems. For speed, we generate Digitally Reconstructed Radiographs on small, dynamically selected regions of interest from precomputed ray gray levels in expected viewing directions, and use a multiresolution hierarchy of fluoroscopic X-ray images. For robustness and accuracy, we use a two-step comparison measure: Normalized Cross Correlation followed by Variance Weighted Sum of Local Normalized Correlation. To avoid local minima, we use a genetic search method. Our experiments on simulated, in-vitro, and cadaver data show an overall mean Target Registration Error of 0.8mm (max=1.5mm), 95% of the time, computed in 20-100 seconds under realistic conditions.