Advanced Knee Structure Analysis (AKSA): a comparison of bone mineral density and trabecular texture measurements using computed tomography and high-resolution peripheral quantitative computed tomography of human knee cadavers

Abstract

Background: A change of loading conditions in the knee causes changes in the subchondral bone and may be a cause of osteoarthritis (OA). However, quantification of trabecular architecture in vivo is difficult due to the limiting spatial resolution of the imaging equipment; one approach is the use of texture parameters. In previous studies, we have used digital models to simulate changes of subchondral bone architecture under OA progression. One major result was that, using computed tomography (CT) images, subchondral bone mineral density (BMD) in combination with anisotropy and global homogeneity could characterize this progression. The primary goal of this study was a comparison of BMD, entropy, anisotropy, variogram slope, and local and global inhomogeneity measurements between high-resolution peripheral quantitative CT (HR-pQCT) and CT using human cadaveric knees. The secondary goal was the verification of the spatial resolution dependence of texture parameters observed in the earlier simulations, two important prerequisites for the interpretation of in vivo measurements in OA patients.

Method: The applicability of texture analysis to characterize bone architecture in clinical CT examinations was investigated and compared to results obtained from HR-pQCT. Fifty-seven human knee cadavers (OA status unknown) were examined with both imaging modalities. Three-dimensional (3D) segmentation and registration processes, together with automatic positioning of 3D analysis volumes of interest (VOIs), ensured the measurement of BMD and texture parameters at the same anatomical locations in CT and HR-pQCT datasets.

Results: According to the calculation of dice ratios (>0.978), the accuracy of VOI locations between methods was excellent. Entropy, anisotropy, and global inhomogeneity showed significant and high linear correlation between both methods (0.68 < R ² < 1.00). The resolution dependence of these parameters simulated earlier was confirmed by the in vitro measurements.

Conclusion: The high correlation of HR-pQCT- and CT-based measurements of entropy, global inhomogeneity, and anisotropy suggests interchangeability between devices regarding the quantification of texture. The agreement of the experimentally determined resolution dependence of global inhomogeneity and anisotropy with earlier simulations is an important milestone towards their use to quantify subchondral bone structure. However, an in vivo study is still required to establish their clinical relevance.

Keywords: Computed tomography; High-resolution peripheral quantitative computed tomography; Knee OA; Subchondral bone; Texture.

Fig. 1

Axial slice of one specimen obtained from clinical CT using a high-resolution kernel (a) and from HR-pQCT (b)

Fig. 2

Multi-planar reformations: transversal (left), coronal (center) and sagittal (right). Top CT dataset with segmented periosteal/articular surface (red) and analysis VOIs (blue); for the CT reconstruction, the high-resolution kernel U70u was applied. Bottom HR-pQCT dataset of the same knee (repositioned) with periosteal/articular surface registered (red) and analysis VOIs (blue) transferred from the CT dataset. The names of the analyses VOIs are only indicated in the femur (top, center) but apply to the tibia as well. For the purpose of illustration, the HR-pQCT was downsampled to the same size as the CT dataset. Each CT image has 512 × 512 pixels with a size of 254 × 254 μm² each, while the HR-pQCT image consists of 1352 × 1484 pixels with a size of 82 × 82 μm². Navigation lines were added to every image in order to indicate the relative positions of the reformed slices. cort cortical, mid-epi mid-epiphyseal, sub epi subchondral epiphyseal

Fig. 3

a Measured BMD across VOIs for CT and HR-pQCT in tibia and femur with error bars as standard deviations from 57 cadavers. b HR-pQCT results unchanged, CT results corrected by the equation obtained from linear regression in (c). c Linear regression analysis of BMD results. d Bland-Altman plots for corrected trabecular BMD. Upper (lower) LOA: 95% upper (lower) confidence limit (LOA = 1.96 × standard deviation of difference). %err = LOA divided by the mean BMD_HR-pQCT. med medial, lat lateral, LOA limit of agreement, S1 subchondral epiphyseal, S2 mid-epiphyseal, S3 juxtaphyseal

Fig. 4

Texture parameters measured with CT or HR-pQCT in the VOIs shown in Fig. 2

Fig. 5

Comparison (using Bland-Altman plots) of texture parameters measured with CT and HR-pQCT. MEAN mean of CT and HR-pQCT measurements, DIFFERENCE CT measurement in CT – HR-pQCT measurement

Fig. 6

Texture parameter ratios D between HR-pQCT and CT measurements. Bars are mean values for 40 digital models simulating a wide variety of trabecular architectures and mean values from twelve trabecular VOIs of 57 cadaveric datasets, respectively. Error bars represent the respective standard deviations. A value of 1 means that the texture parameter does not depend on spatial resolution within the investigated range from about 100 μm (HR-pQCT) to 400 μm (CT)