Abstract
Purpose
For 3D kinematic analysis of total knee arthroplasty (TKA), 2D/3D registration techniques which use X-ray fluoroscopic images and computer-aided design model of the knee implants, have been applied to clinical cases. These techniques are highly valuable for dynamic 3D kinematic analysis, but have needed time-consuming and labor-intensive manual operations in some process. In previous study, we reported a robust method to reduce manual operations to remove spurious edges and noises in edge detection process of X-ray images. In this study, we address another manual operations problem occurred when setting initial pose of TKA implants model for 2D/3D registration. To set appropriate initial pose of the model with manual operations for each X-ray image is important to obtain the good registration results. However, the number of X-ray images for a knee performance is very large, and thus to set initial pose with manual operations is very time-consuming and a problem for practical clinical applications. Therefore, this study proposes an initial pose estimation method for automated 3D kinematic analysis of TKA.
Methods
3D pose of an implant model is estimated using a 2D/3D registration technique based on a robust feature-based algorithm.
To reduce labor-intensive manual operations of initial pose setting for large number of X-ray images, we utilize an interpolation technique with an approximate function. First, for some X-ray images (key frames), initial poses are manually adjusted to be as close as possible, and 3D poses of the model are accurately estimated for each key frame. These key frames were appropriately selected from the 2D feature point of knee motion in the X-ray images. Next, the 3D pose data estimated for each key frame are interpolated with an approximate function. In this study, we employed a multilevel B-spline function. Thus, we semi-automatically estimate the initial 3D pose of the implant model in X-ray images except for key frames. Fig. 1 shows the algorithm of initial pose estimation, and Fig. 2 shows the scheme of the data interpolation with an approximate function.
Experimental results
To validate the feasibility of the proposed initial pose estimation method, experiments using X-ray fluoroscopic images of 8 TKA patients during knee motions were performed. For the experiments, we prepared two sorts of contour images, and applied the proposed method to the one image contained spurious edges and noises. The other image which spurious edges and noises didn't exist was used for determination of correct poses (reference data) using 2D/3D registration. In order to assess the performance of the proposed method, automation rate was calculated, and the rate was defined as the X-ray frame number of satisfying clinical required accuracy (error within 1 mm, 1 degree) relative to all X-ray frame number.
As results of the experiments, the automation rate of the femoral and tibial component were about 79 % and 73 %, respectively.
Conclusions
This study presented an initial pose estimation method for automated 3D kinematic analysis of TKA using X-ray fluoroscopic images. The method without labor-intensive operations is thought to be very useful for practical clinical applications.
