Purpose: Existing microstructure parameters are able to predict vertebral in vitro failure load, but for noisy in vivo data more complex algorithms are needed for a robust assessment.
Methods: A new algorithm is proposed for the microstructural analysis of trabecular bone under in vivo quantitative computed tomography (QCT). Five fractal parameters are computed: (1) the average local fractal dimension FD, (2) its standard deviation FD.SD, (3) the fractal rod volume ratio fRV/BV, (4) the average fractal trabecular thickness fTb.Th, and (5) its coefficient of variation fTb.Th.CV. The algorithm requires neither an explicit skeletonization of the trabecular bone, nor a well-defined transition between bone and marrow phases. Two experiments were conducted to compare the fractal with established microstructural parameters. In the first, 20 volumes-of-interest of embedded vertebrae phantoms were scanned five times under QCT and high-resolution (HR-)QCT and once under peripheral HRQCT (HRpQCT), to derive accuracy and precision. In the second experiment, correlations between in vitro HRQCT structural parameters were obtained from 76 human T11, T12, or L1 vertebrae. In vitro fracture data were available for a subset of 17 human T12 vertebrae so that linear regression models between failure load and microstructural HRQCT parameters could be analyzed.
Results: The results showed correlations of fTb.Th and fRV/BV with their nonfractal pendants trabecular thickness (Tb.Th) and respective structure model index (SMI) while higher precision and accuracy was observed on the fractal measures. Linear models of bone mineral density with two and three fractal microstructural HRQCT parameters explained 86% and 90% (adjusted R2) of the failure load and significantly improved the linear models based only on BMD and established standard microstructural parameters (68%-77% adjusted R2).
Conclusions: The application of fractal methods may grant further insight into the study of bone quality in vivo when image resolution and quality are less than optimal for current standard methods.