Inter-rater accuracy and precision of markerless quantification of 3d tibiofemoral displacement using dual fluoroscopy

Abstract

Purpose: Early changes in osteoarthritic (OA) knee cartilage include tissue swelling and softening, which may predispose cartilage to mechanical damage during activities of daily living. Current clinical OA diagnosis relies on radiographic bone morphology change and pain, which can be insensitive to early change. Developing a marker that specifically targets early mechanical tissue changes may advance the current state of OA assessment. Potential candidates for such markers are in-vivo magnitude and rate of tibiofemoral (TF) soft tissue deformation during weight bearing (compression) and unloading (recovery). Respective joint “signatures” for healthy and OA groups may then allow for early OA identification. Dual Fluoroscopy (DF) is an emerging tool with the potential to provide the temporal and spatial resolutions necessary for such assessment. Tissue deformation mechanics may be derived using translation and rotation changes of the femur with respect to the tibia over time. Using a bead tracking technique, similar to radiostereometric analysis (RSA), translation changes of 0.05 mm (1/ 10th expected maximal cartilage deformation) have previously been accurately detected. It is however unknown if a non-invasive markerless DF approach is sufficiently accurate and precise for quantifying the small changes during weight bearing. The purpose of this study was to establish the Minimum Detectable Displacement (MDD) of 3D femoral translations with respect to the tibia in a custom built DF system using a markerless shape matching image registration approach. Methods: The DF system consisted of two X-ray tubes (Varian, USA) and generators (EMD, Canada), quad mode image intensifiers (Toshiba, Japan) and high speed digital video cameras (PCO, Germany). DF images were acquired with an entrance field diameter of 228 mm and 0.155 x 0.155 mm2 pixel resolution. Dry tibia and femur bones were fixed to a plexiglass frame, where the tibia remained stationary and the femur was translated using a digital micrometer (Mitutoyo, Japan). The frame was placed within the DF field-of-view and did not move during data capture. Ten displacement intervals (0.01-0.1 mm)were imagedwith 10 repetitions of each interval. DF calibration was performed using a calibration cube and modified direct linear transform. Images were distortion corrected using a perforated steel grid and XrayProject (Brown University, USA). 3D femur and tibia bone models were generated using segmentation (Amira, USA) of Computed Tomography image sequences (0.21 x 0.21 x 0.31 mm). 3D bone orientations were obtained by three raters using AutoScoper (Brown University, USA) and a markerless shape matching approach. Bone-specific models were positioned to align their visible features to those of the underlying 2D X-ray images. Relative femur and tibia displacements were obtained using the Euclidean distance between bone models. Independent samples t-tests were used to detect MDDs of the moving femur relative to the stationary tibia at each displacement interval. Accuracy and precision were computed as means and standard deviations of the absolute differences of the femur translations compared to the micrometer displacements. Inter-rater reliability values of the femur were computed using intra-class correlation coefficient (ICC). All statistical analyses were performed in SPSS (IBM, USA) and p 0.05. Results: Femur displacement measurements were significantly different from tibia (p<0.045) and had greater than 95% confidence in accuracy at micrometer displacements 0.:08mm (accuracy precision: 0.022 0.019) (Figure 1). Between raters femur displacement measurements were not significant for micrometer displacements 0.:05mm and inter-rater reliability across all displacements was 92%. Conclusions: These findings indicate that 3D bone translations within relevant in-vivo ranges can be observed accurately and precisely using a markerless DF shape matching approach. Femur displacements of 0.:08mm were accurately and reliably detected with respect to set micrometer displacements. The data further indicated good between rater agreement when quantifying femur displacements. Given the above system specifications, the markerless DF approach provides a means for non-invasive in-vivo quantification of TF tissue response to loading and unloading stimuli. Using this approach, further workwill be performed to establish the mechanical behaviour of TF soft tissues of healthy and ligament deficient and OA knees. While significant user input is required, compared to RSA type techniques, this non-invasive approach may be more readily applied in a future clinical setting.