Longitudinal Synthesis of CT Data of Bony Implants
Figure 1: Early time point in vivo x-ray CT scan of bony implants and spinal column of rabbit.
For each subject, each vertebra was isolated at the earliest available time point, where the bone growth in the implants was not yet sufficient to corrupt the local bone signal. These vertebrae were then rigidly co-registered by subject to all subsequent time points. The entirety of each vertebral region was then cut from the image, allowing for a straightforward segmentation of each implant. This not only increased the ease of segmentation, but also the accuracy since the native bone was unable to corrupt higher level analysis.
Figure 1: For later time points, the implants can be difficult to isolate from the bone.
Figure 2: The vertebrae from earlier time points are registered and removed.
Figure 3: The implants are easily segmentable. The final segmentation of the implants is robust, and the image is prepared for higher level analysis.
The ex vivo samples that were measured using high resolution x-ray CT were then analyzed for their structural integrity. In principle it is possible to construct a reliable finite elements model to predict strength under loads. However, these models require a prohibitive amount of training and computational footprint for high resolution samples. Instead, a structural parameter based on the electrical analogue of conductivity was used to analyze the samples for structure. First, the samples were segmented from the background using a two-call, one threshold entropy minimization histogram method. Due to biological variation, each sample was a slightly different length. In order to accurately compare samples to each other, it was also necessary to isolate a uniform length. Finally, an adaptive, dual-direction morphological connectivity scheme was used to isolate the subsections of the implant which were structurally relevant given a simulated end load.
The longitudinal segmentation scheme provided notable decreases in processing time and increases in accuracy with respect to hand segmentation. This is one example of many of the development and use of nonstandard but creative segmentation and quantification methods at Invicro.