Toward Accurate and Robust 2-D/3-D Registration of Implant Models to Single-Plane Fluoroscopy

Abstract

A fully automatic and highly accurate 2-D/3-D registration technique with an extended capture range is proposed for registering implant models to single-plane fluoroscopy. The proposed method utilizes library-based registration for pose initialization, followed by an intensity-based iterative registration for pose fine tuning. The algorithm matches the 2-D silhouette of the implant extracted from the fluoroscopy with a pre-computed library of 2-D silhouettes of the implant model to estimate the initial pose. Each library entry represents a combination of out-of-plane rotation parameters. Library matching is performed by computing Shape Context (SC) of the extracted 2-D silhouette and minimizing the Jensen-Shannon Divergence of SCs from the fluoroscopy and the library entry. After pose initialization, iterative optimization is performed to fine tune the registration by maximizing the intensity-based similarity measure between the fluoroscopic image and the simulated X-ray image. In the iterative registration, we use a novel two-layer hierarchical optimization strategy to achieve a high accuracy in depth estimation, to which the projection image is very insensitive. The proposed approach is validated on both computer simulated images and real X-ray images. Validation results show significant improvements over conventional methods in terms of robustness and accuracy.