Bone Abstracts

Introduction: The quantification of cortical BMD and thickness in QCT images remains challenging due to the limited spatial resolution of CT scanners. We simulated the impact of longitudinal cortical BMD and thickness changes on accuracy of cortical measurements using three different segmentation algorithms.

Methods: A step function of varying width d (cortical thickness) and height (cortical BMD) and an additional step representing trabecular BMD was convoluted with a Gaussian function of varying full width at half maximum (FWHM) describing the CT scanner resolution and simulating the density distribution within a reconstructed CT image. Used segmentation algorithms: local adaptive 50% thresholds (LA), global thresholds (GT), Levenberg-Marquardt based optimization method (OM)¹. Accuracy errors of ΔBMD, Δd and ΔBMC measurements in the CT images were estimated by simulating a 2.5, 5 and 7.5% BMD increase at constant d and a 5, 10 and 20% increase of d at constant BMD.

Results: i) Simulated change in d: with LA and GT increasing accuracy errors in d occur for d<2FWHM and with OM for d<FWHM. All three algorithms resulted in false cortical BMD increases. ii) Simulated BMD change: all three algorithms showed accuracy errors in cortical BMD for d<2FWHM. LA showed no effects on d, GT overestimated d for d<2FWHM, while OM overestimated d for d<4FWHM. Added noise (20HU, obtained from standard QCT images of the spine) affected particularly OM if the cortex was thin (d<FWHM). In general, BMC errors were smaller than those for BMD.

Conclusion: The investigated algorithms show good results for d>2FWHM. For thinner cortices, each segmentation method affects cortical parameters differently. It is important to measure cortical BMC in addition to BMD.

Reference: 1. Treece et al. Medical Image Analysis 2010.