Automated simulation of areal bone mineral density assessment in the distal radius from high-resolution peripheral quantitative computed tomography

Abstract

Summary: An automated image processing method is presented for simulating areal bone mineral density measures using high-resolution peripheral quantitative computed tomography (HR-pQCT) in the ultra-distal radius. The accuracy of the method is validated against clinical dual X-ray absorptiometry (DXA). This technique represents a useful reference to gauge the utility of novel 3D quantification methods applied to HR-pQCT in multi-center clinical studies and potentially negates the need for separate forearm DXA measurements.

Introduction: Osteoporotic status is primarily assessed by measuring areal bone mineral density (aBMD) using 2D dual X-ray absorptiometry (DXA). However, this technique does not sufficiently explain bone strength and fracture risk. High-resolution peripheral quantitative computed tomography (HR-pQCT) has been introduced as a method to quantify 3D bone microstructure and biomechanics. In this study, an automated method is proposed to simulate aBMD measures from HR-pQCT distal radius images.

Methods: A total of 117 subject scans were retrospectively analyzed from two clinical bone quality studies. The distal radius was imaged by HR-pQCT and DXA on one of two devices (Hologic or Lunar). Areal BMD was calculated by simulation from HR-pQCT images (aBMD(sim)) and by standard DXA analysis (aBMD(dxa)).

Results: The reproducibility of the simulation technique was 1.1% (root mean-squared coefficient of variation). HR-pQCT-based aBMD(sim) correlated strongly to aBMD(dxa) (Hologic: R (2) = 0.82, Lunar: R (2) = 0.87), though aBMD(sim) underestimated aBMD(dxa) for both DXA devices (p < 0.0001). Finally, aBMD(sim) predicted aBMD at the proximal femur and lumbar spine with equal power compared to aBMD(dxa).

Conclusion: The results demonstrate that aBMD can be simulated from HR-pQCT images of the distal radius. This approach has the potential to serve as a surrogate forearm aBMD measure for clinical HR-pQCT studies when axial bone mineral density values are not required.

Fig. 1

Images indicating the standard ultra-distal ROI for each device; HR-pQCT scout scan (a), Hologic DXA (b), Lunar DXA (c)

Fig. 2

Schematic of the algorithm for simulating aBMD from 3D HR-pQCT image data

Fig. 3

Diagram of the common anatomic coordinate system the radius HR-pQCT image is aligned to. The transformation (θ) is applied about the midpoint (mp) of the line connecting the centroids of the radius (c _R) and ulna (c _U) in the central slice

Fig. 4

Representative simulated projection image of the UD radius

Fig. 5

Regression analysis (a) and Bland–Altman (b) plots comparing aBMD_sim against aBMD_dxa

Fig. 6

Regression analysis plots for aBMD_sim and aBMD_dxa at the UD radius against standard aBMD measurements at the proximal femur (a, b) and lumbar spine (c, d)