Automated bone volume and thickness measurements in small animal whole-body MicroCT data

Quantification of osteolysis is crucial for monitoring treatment effects in preclinical research and should be based on MicroCT data rather than conventional 2D radiographs to obtain optimal accuracy. However, data assessment is greatly complicated in the case of 3D data. This paper presents an automated method to follow osteolytic lesions quantitatively and visually over time in whole-body MicroCT data of mice.

This novel approach is based on a previously published approach to coarsely locate user-defined structures of interest in the data and present them in a standardized manner (Baiker et al., Med Image Anal 14:723-737, 2010; Kok et al., IEEE Trand Vis Comput Graph 16:1396-1404, 2010). Here, we extend this framework by presenting a highly accurate way to automatically measure the volumes of individual bones and demonstrate the technique by following the effect of osteolysis in the tibia of a mouse over time. Besides presenting quantitative results, we also give a visualization of the measured volume to be able to investigate the performance of the method qualitatively. In addition, we describe an approach to measure and visualize cortical bone thickness, which allows assessing local effects of osteolysis and bone remodeling. The presented techniques are fully automated and therefore allow obtaining objective results, which are independent of human observer performance variations. In addition, the time typically required to analyze whole-body data is greatly reduced.

Evaluation of the approaches was performed using MicroCT follow-up datasets of 15 mice (n = 15), with induced bone metastases in the right tibia. All animals were scanned three times: at baseline, after 3 and 7 weeks. For each dataset, our method was used to locate the tibia and measure the bone volume. To assess the performance of the automated method, bone volume measurements were also done by two human experts. A quantitative comparison of the results of the automated method with the human observers showed that there is a high correlation between the observers (r = 0.9996), between the first observer and the presented method (r = 0.9939), and also between the second observer and the presented method (r = 0.9937). In addition, Bland-Altman plots revealed excellent agreement between the observers and the automated method (interobserver bone volume variability, 0.59 ± 0.64%; Obs1 vs. Auto, 0.26 ± 2.53% and Obs2 vs. Auto, -0.33 ± 2.61%). Statistical analysis yielded no significant difference (p = .10) between the manual and the automated bone measurements and thus the method yields optimum results. This could also be confirmed visually, based on the graphical representations of the bone volumes. The performance of the bone thickness measurements was assessed qualitatively.

We come to the conclusion that the presented method allows to measure and visualize local bone volume and thickness in longitudinal data in an accurate and robust manner, proving that the automated tool is a fast and user friendly alternative to manual analysis.