Particle filter for implanted knee kinematic analysis using dynamic radiograph video

Implanted knee kinematic analysis plays one of important role in clinical and research fields of total knee arthroplasty. Although there are some studies to analyze X-ray images for estimating 3-D knee kinematics, most of them cannot analyze dynamic video because they strongly depend on manual interaction of giving initial pose/position. This paper utilizes particle filter for analyzing dynamic radiograph video of implanted knee. By using particle filter, the proposed method does not require not only user interaction but also computational iteration of parameter optimization. As the result, we shorten the computation time and improved the estimation accuracy. The estimation error was lower than 0.7 mm for rotation, and 0.5 mm for translation including out-plane, and the computation time was 1.27 sec per frame using a cluster computer.