Accuracy evaluation of a shape-based registration method for a computer navigation system for total knee arthroplasty.

This study evaluated the effect of computed tomography (CT) slice thickness, reconstruction pitch, intraoperative data sampling area, and data sampling volume on the accuracy of registration and determined a clinically acceptable trade-off between accuracy and surgical invasiveness. One cadaveric femur and one cadaveric tibia were used. Computed tomography of the femur and tibia were obtained using a helical scanner. Three sets of slice thickness and slice pitch were chosen for data acquisition, and two additional sets of reconstructed data were made. Bone contours were extracted by removing surrounding substrate. Surface models of bones were made from the resulting data. Registration of surface models to real objects was performed by measuring the position of various surface points on various areas of each object using an OPTOTRAK pen-probe (Northern Digital Inc, Ontario, Canada). The following trade-off is proposed as clinically optimal: perform CT with 3-mm slice thickness and 1-mm reconstruction pitch, and sample a periarticular area of 30 sampling points. The accuracy of registration in terms of position and angle was 0.8 mm and 0.6 degrees of bias with 0.2 mm and 0.3 degrees of root-mean-square in the femur, and 0.5 mm and 0.4 degrees of bias with 0.2 mm and 0.3 degrees of root-mean-square in the tibia.