Abstract
Retrieval analysis has been valuable in the assessment of in-vivo surface damage of orthopedic devices. Historically, subjective techniques were used to grade damage on the implant's surface. Microscopy improved the ability to localize and quantify damage, but cannot measure volumetric wear due to this damage. Laser scanning provides volumetric wear, but lacks image data. Recent techniques superimpose image data on laser scan data (photorendering) and combine the strengths of both methods. Our goal is to use such methods to improve our damage assessment and potentially correlate this assessment to volumetric wear.
This project focused on two areas: image-stitching and photorendering. Image-stitching registers multiple images into large-scale high-resolution composites. Six total disc replacement components were imaged with a digital microscope (Moticam 2, Motic). Three sets were taken of each component: a single template at 10x zoom (1×1), a 4-image composite at 18x zoom (2×2), and a 9-image composite at 18x zoom (3×3). The 2×2 and 3×3 sets were image-stitched to resemble their template counterpart. Measurement error was defined using common pixels identified between the composite and template images for comparison with a semi-automated feature detection algorithm (Figure 1).
For photorendering, a pilot study was performed on a single retrieved tibial bearing. The component was imaged with a digital microscope (VHX-2000, Keyence) under a 3D image-stitching setting, providing a high-resolution photo embedded with height values. MATLAB was used to convert the image into a photo-rendered point cloud approximating the surfaces. The component was then laser scanned, creating a 3D point cloud with resolution 0.127 mm. The photo-rendered point cloud data was registered to the laser scan data using an iterative closest point algorithm (Geomagic Studio, Geomagic).
An analysis of all composite images showed a mean error of 0.221 mm. Figure 2 compares regions of images for the template, 2×2, and 3×3 composites. Zooming in shows the effect of the increased resolution contained in the composite. The 2×2 and 3×3 composites had mean errors of 0.231 mm and 0.209 mm, respectively; these were not significantly different. Comparisons among image types showed that components with less features exhibited larger errors during image-stitching. Figure 3 shows images resulting from each step of the photorendering process. The final image of the figure shows a qualitative result of our ability to photorender the tibial bearing surface of the component.
While combining microscopy and laser scan data works anecdotally, further analyses must be performed to assure the robustness of the technique. The digital microscope's embedded image-stitching software is limited in its maximum field of view; we look to extend this by taking multiple scans and using in-house software to generate a composite of a whole implant. The improved resolution provided by microscopy offer an opportunity to automate damage assessment, yielding damage mapped images which can also be overlaid on laser scan data. This may provide a means to better quantify observed damage and yield meaningful correlations with volumetric loss due to wear.
