Dependence of model-based RSA accuracy on higher and lower implant surface model quality

BACKGROUND: Model-based Roentgen Stereophotogrammetric Analysis (MBRSA) allows the accurate in vivo measurement of the relative motion between an implant and the surrounding bone (migration), using pose-estimation algorithms and three dimensional geometric surface models of the implant. The goal of this study was thus to investigate the effect of surface model resolution on the accuracy of the MBRSA method. METHODS: Four different implant geometries (knee femoral and tibial components, and two different hip stems) were investigated, for each of which two reversed engineering (RE) models of differing spatial digitizing resolution were generated. Accuracy of implant migration measurement using MBRSA was assessed in dependence on surface model resolution using an experimental phantom-model set up. RESULTS: When using the lower quality RE models, the worst bias observed ranged from -0.048 to 0.037 mm, and -0.057 to 0.078 deg for translation and rotation respectively. For higher quality reverse engineering models, bias ranged from -0.042 to 0.048 mm, and -0.449 to 0.029 deg. The pair-wise comparisons of digitizing resolution (higher vs. lower quality) within the different implant type revealed significant differences only for the hip stems (p < 0.001). CONCLUSION: The data suggest that the application of lower resolution RE models for MBRSA is a viable alternative method for the in vivo measurement of implant migration, in particular for implants with non symmetrical geometries (total knee arthroplasty). Implants with larger length to width aspect ratio (total hip arthroplasty) may require high resolution RE models in order to achieve acceptable accuracy. Conversely, for some axis the bias for translation are clearly worse for translation, and are marginally better for rotations using the lower resolution RE models instead of the higher ones. However, performed box plots ranges were well within what has been reported in the literature. The observed lower accuracy and precision of the measurements for hip stem components for rotations about the superior-inferior direction is presumably the result of the nature of the MBRSA method. This well known effect within MBRSA for rotations about the axis of symmetry of axially-symmetric objects do not change the contour of the projected image to as large a degree as motion about a non-symmetric axes. It is not possible to detected this small motion as accurately using pose-estimation methods. This may affect the “higher” accuracy for the applied lower resolution RE models.