A statistical framework for the registration of 3D knee implant components to single-plane X-ray images

Abstract

Registration of 3D knee implant components to single-plane X-ray image sequences provides insight into implanted knee kinematics. In this paper a maximum likelihood approach is proposed to align the pose-related occluding contour of an object with edge segments extracted from a single-plane X-ray image. This leads to an expectation maximization algorithm which simultaneously determines the objectpsilas pose, estimates point correspondences and rejects outlier points from the registration process. Considering (nearly) planar-symmetrical objects, the method is extended in order to simultaneously estimate two symmetrical object poses which both align the corresponding occluding contours with 2D edge information. The algorithmpsilas capacity to generate accurate pose estimates and the necessity of determining both symmetrical poses when aligning (nearly) planar-symmetrical objects will be demonstrated in the context of automated registration of knee implant components to simulated and real single-plane X-ray images.