In this study, image registration was successfully used to align multiple stacks in HR-pQCT images of the 2nd and 3rd MCP joints, reducing the impact of stack shift artifact to improve the number of usable scans and increase the number of erosions that could be analyzed longitudinally. Consistent with our current study, where 23% of erosions could not have been quantified without stack registration, we observed in a different cohort of patients that 10/48 (21%) of images had to be excluded from erosion analysis due to stack shift artifact that crossed through the middle of an erosion (25). Further, a 3D longitudinal image registration and bone remodeling analysis method was applied to these MCP scans. It was observed that over the course of 6 months, there was a net loss of bone in patients classified in clinical remission. Both of these image registration techniques could be implemented in future HR-pQCT studies to quantify bone changes over time in patients with RA.
While multi-stack registration reduces the impact of the stack artifact and improves longitudinal image registration to assess bone remodeling, motion artifact still has an impact on the quantitative analysis of bone remodeling. Although it was not significant, we observed that low quality images of the MCP joints with higher motion grades led to a trend towards higher bone formation and resorption volumes compared with images with a higher quality and lower motion grades. This trend was also observed in a previous study using a similar bone remodeling analysis method on the radius and tibia on the first generation HR-pQCT scanner (18). This suggests that even if scans meet the motion scoring criteria to be included in analysis, motion could still have an impact on bone remodeling analysis. Despite the improvements in the number of erosions available for analysis after multi-stack registration, there were several stacks with a high motion grade that had to be excluded from the study. Further, visual inspection was used to confirm registration alignment. Using rigid registration, we presumed that no significant changes in shape occurred over the 6-month period. In particular, our visual inspection focused primarily on the region around the erosion as this was our primary region of interest. Slight errors in registration and/or motion artifact may lead to inaccuracies in the global bone remodeling analysis. Previous studies using microCT in animal models have demonstrated that longitudinal analyses of bone remodeling produces better accuracy and precision when performed on a local region of interest rather than the whole bone (26–28). It is likely that future studies using this bone remodeling analysis in the context of rheumatoid arthritis will be primarily concerned with bone changes around the region of an erosion, rather than the global changes throughout the bone. Therefore, addition, further methods to reduce motion artifact and improve image registration, including focus on a local region of interest should be investigated. In addition, future studies in rheumatoid arthritis should evaluate the utility of bone remodeling analysis in the peri-erosion region.
We also explored whether applying interpolation to both fixed and moving images decreased the interpolation error to an extent that it impacted bone remodeling volume fraction. The application of a transformation matrix to the moving image results in errors caused by image interpolation which may have an impact on longitudinal bone analysis (24). However, by comparing bone remodeling fractions between scan-rescan registrations where a cubic interpolation was applied to either just the moving image or both images, we found that the difference in bone remodeling fractions due to interpolation was not significant. These results are consistent with previous work comparing the effects of various image interpolation methods on bone morphometric analysis from micro-CT scans using an animal model (24). The current study demonstrated this finding in a human population using HR-pQCT scans.
Scan-rescan analysis was also completed to act as a negative control. There was a significant difference in the bone formation and resorption fractions between the 6-month longitudinal and negative control, indicating that the changes observed in formation and resorption were due to actual bone changes, rather than image noise and patient repositioning. Part of the challenge for scan-rescan analysis of bone remodeling may be due to the absence of scans with a motion score of 1 (i.e., no motion). As a result, the scan-rescan bone remodeling volume fractions may have been overestimated when compared to the longitudinal analysis based on the lack of any extremely high-quality scans. While variations in x-ray tube or detector functioning can affect the density values used to perform bone remodelling analysis, the fluctuations recorded on the daily quality control scans were very small relative to the differences detected over the 6-month follow-up period
There are some limitations associated with the multi-stack image registration technique. First, the acquisition of 3 separate stacks can be challenging for the operator and participant. As an overlapping acquisition protocol has not been implemented by the manufacturer, each stack must be acquired individually rather than contiguously. This results in a longer acquisition time, as well as the scanner stopping and starting between stacks, which could potentially lead to even more movement. This may also explain why there was such a large proportion of scans that would have had to be excluded without the registration of stacks as the noticeable breaks between stack acquisition are not present in the standard protocol. Further, this technique does not eliminate motion artifact, as the reconstruction of each stack takes into account the average position of each stack. When there is motion within a stack, this can impact the scan in a way that cannot be corrected by this image registration technique. While the current study improved the number of scans that could be analyzed, only 9 out of 20 joints (45%) met the criteria to be included in the longitudinal analysis. There were 4 joints that could not be analyzed due to acquisition error or motion that prevented successful multi-stack registration, 1 joint that failed stack registration despite appropriate motion scores, and 6 joints that successfully had the stacks aligned, but were excluded since they did not meet our motion criteria. Despite the benefits of the multi-stack registration, motion artifact still presents a major limitation when analyzing HR-pQCT scans of the MCP joints. Future investigations should build upon this work to reduce the impact motion artifact has on these scans.
There are also limitations associated with the bone remodeling analysis method used in this study. First, the threshold value of 225 mgHA/cm3 presented in previous work (17,18) was not used in this study. After several tests assessing the accuracy of the results with visual inspection, it was observed that obvious areas of bone changes on the grey-scale images were not detected using this threshold for the RA patients in this cohort. Therefore, a new threshold was selected (125 mgHA/cm3) after several tests that appeared to give the best depiction of bone change. This may be because the previous work was done on the radius and tibia, on the first-generation HR-pQCT scanner with lower spatial resolution, which may have different thresholds of change than the MCP joints analyzed in the current study. This new threshold should be validated using cadaveric samples to ensure that the change being visualized is due to real bone change and not noise.
The multi-stack image registration technique proposed in this study could be a valuable tool to improve the utilization of HR-pQCT scans for the research of bone changes in RA. Improving the number of useable scans will be instrumental in furthering the use of HR-pQCT to study bone changes in RA. Further, 3D longitudinal image registration can be applied to second generation HR-pQCT scans to visualize and quantify bone changes at a higher resolution, over a relatively short time period (6-months) without the need for a bone biopsy.