A Fast and Robust Framework for 3D/2D Model to Multi-Frame Fluoroscopy Registration

Abstract

Three-dimensional (3D) kinematic analysis plays an important role in improving diagnosis and in the evaluation of treatments and surgical procedures. For example, measuring the 3D kinematics of knee joints is essential for understanding their normal function and diagnosing any pathology, such as ligament injury and osteoarthritis. Image registration is a method which can be used to compute kinematic measurements without involving the introduction of instruments into the body. However, in these techniques, the trade-off between accuracy and computation time is still a challenging problem which needs to be addressed. In this paper, a fast and robust registration method is proposed for the measurement of post-operative knee joint kinematics. Using this method, after total knee arthroplasty (TKA) surgery, a 3D knee implant model can be registered with a number of single-plane fluoroscopy frames of the patients’ knee. Generally, when the number of fluoroscopy frames is quite high, the computation cost for the registration between the frames and a 3D model is expensive. Therefore, in order to speed up the registration process, we apply an interpolation-based prediction method, to initialize and estimate the 3D positions of the 3D model in each fluoroscopy frame. The estimated 3D positions are then fine-tuned. The experimental results, which were performed on the knee joints of 18 patients post-surgery, show that the computational time required to register each frame for each bone using our proposed method is only 67 seconds, which is much faster (almost 6.5 times faster) than the best existing registration method (a registration method based on sum of conditional variance (SCV) similarity measure) while maintaining almost the same accuracy. The average of the mean difference ± standard deviation of the proposed method for femoral and tibial bones for translation and rotation parameters are 0.0603 ± 0.2966 (mm) and −0.0069 ± 0.2922 (degree) respectively.

Keywords

• Cubic spline interpolation
• edge position difference
• image registration
• model to multi-frame fluoroscopy registration
• similarity measure