Reproducibility of direct quantitative measures of cortical bone microarchitecture of the distal radius and tibia by HR-pQCT

Abstract

Quantitative cortical microarchitectural end points are important for understanding structure-function relations in the context of fracture risk and therapeutic efficacy. This technique study details new image-processing methods to automatically segment and directly quantify cortical density, geometry, and microarchitecture from HR-pQCT images of the distal radius and tibia. An automated segmentation technique was developed to identify the periosteal and endosteal margins of the distal radius and tibia and detect intracortical pore space morphologically consistent with Haversian canals. The reproducibility of direct quantitative cortical bone indices based on this method was assessed in a pooled data set of 56 subjects with two repeat acquisitions for each site. The in vivo precision error was characterized using root mean square coefficient of variation (RMSCV%) from which the least significant change (LSC) was calculated. Bland-Altman plots were used to characterize bias in the precision estimates. The reproducibility of cortical density and cross-sectional area measures was high (RMSCV <1% and <1.5%, respectively) with good agreement between young and elder medians. The LSC for cortical porosity (Ct.Po) was somewhat smaller in the radius (0.58%) compared with the distal tibia (0.84%) and significantly different between young and elder medians in the distal tibia (LSC: 0.75% vs. 0.92%, p<0.001). The LSC for pore diameter and distribution (Po.Dm and Po.Dm.SD) ranged between 15 and 23 microm. Bland-Altman analysis revealed moderate bias for integral measures of area and volume but not for density or microarchitecture. This study indicates that HR-pQCT measures of cortical bone density and architecture can be measured in vivo with high reproducibility and limited bias across a biologically relevant range of values. The results of this study provide informative data for the design of future clinical studies of bone quality.

Figure 1

Diagram of the image processing algorithm for the cortical compartment segmentation. In the first step, the periosteal surface identified (1A). Next the endosteal boundary is detected which defines the trabecular compartment (2B). In the final step the periosteal and endosteal regions are digitally subtracted to define the apparent cortical compartment (3C), which is used to mask the bone structure to yield the mineralized cortical bone segmentation (3D).

Figure 2

Diagram of the image processing algorithm for the intra-cortical porosity segmentation. Pores are first extracted using a 2D connectivity filter to find pores that are fully surrounded by bone (1E). A second pore estimate (2F) is obtained by region growing the initial porosity along the axis of the bone, which will extract any connected pores that did not meet the criteria of being fully surrounded by bone. The binary cortex image is then filled with all found pores, and a final pass is made using 2D connectivity to extract remaining pores. All extracted pores are combined into a final image (3H) for assessment of Po.Dm and Po.Dm.SD.

Figure 3

Diagram of the image processing procedure to refine the cortical region of interest using the combination of the results from the first two stages. The result of this stage (I) is used to calculate Ct.Th and Ct.Po using direct measurement techniques.

Figure 4

Representative 2D slices of a radius (A) and tibia (B) illustrating the cortical compartment (white), intra-cortical porosity (red), and trabecular compartment (gray) at the distal (left) and proximal (middle) extent of the scan region. 3D visualization of the segmented cortical bone (white, transparent) and intra-cortical porosity (red) is shown on the right.

Figure 5

Representative images from young (A) and elderly (B) subjects illustrating inadequate localization of the endosteal contour (green) on the left, and the operator corrected contour on the right.

Figure 6

Bland-Altman plots for the density, thickness and cross-sectional area parameters. The radius is shown in circles and the tibia in squares. Young and elder groups were not significantly different, except for cortical and trabecular area, where there was a slight positive bias for the repeated measurement in the radius of the elder individuals. Errors for all other parameters were normally distributed.

Figure 7

Bland-Altman plots for porosity measurements. The radius is shown in circles and the tibia in squares. All parameters showed normal error distributions.