A 2D-3D Image Registration Algorithm Using Log-Polar Transforms for Knee Kinematic Analysis

In medical imaging, 2D-3D registration approaches have been used for image guided surgery (IGS) and pre-, post-operative measurement of joint kinematics. The application addressed in this paper is the registration of single-plane 2D fluoroscopy images to 3D Computed Tomography (CT) data. A limitation of previously proposed 2D-3D registration algorithms has been an inability to register images with large initial displacements. The new algorithm proposed in this paper uses log-polar transforms in the frequency domain to extend the range of initial displacements that can be registered automatically. Our experiment results show that the proposed approach can increase the range of initial displacements up to ±30 mm and ±90º. This range of motions is significantly greater than any previously proposed multi-modal 2D-3D registration algorithm.