Improved accuracy in the assessment of vertebral cortical thickness by quantitative computed tomography using the Iterative Convolution OptimizatioN (ICON) method.

Abstract

Vertebral whole bone strength is substantially affected by cortical bone properties. Disease and therapy may affect cancellous and cortical bone differently. Unlike Dual X-ray Absorptiometry (DXA), Quantitative Computed Tomography (QCT) permits selective assessment of cortical and cancellous bone, but image quality limits the accuracy. We present an image processing method specifically adopted to thin cortices that substantially improves accuracy. Ten human vertebrae embedded in epoxy resin were imaged using clinical QCT and High-Resolution QCT (HR-QCT) protocols, both acquired on a clinical whole body CT scanner, whereas high resolution peripheral QCT (HR-pQCT) was used as gold standard. Microstructural variables and BMD were calculated using in-house software StructuralInsight for QCT and HR-QCT and the manufacturer s μCT evaluation software for HR-pQCT. An adjusted measure, a deconvolved cortical thickness (dcCt.Th), corrected for partial volume effects, was derived applying the new Iterative Convolution OptimizatioN (ICON) method. Direct measurements of cortical thickness (Ct.Th) showed substantial overestimation with mean\u202f±\u202fstandard deviation of 1.8\u202f±\u202f0.5\u202fmm for QCT and 1.5\u202f±\u202f0.3\u202fmm for HR-QCT compared to 0.37\u202f±\u202f0.07\u202fmm using HR-pQCT. Correlations of both QCT (r2\u202f=\u202f0.05, p\u202f>\u202f0.5.) and HR-QCT (r2\u202f=\u202f0.38, p\u202f=\u202f0.060) with the gold standard HR-pQCT were not significant. Also QCT-based BMD and BMC as well as HR-QCT-based BMD did not show a significant correlation with the gold standard approach. Only HR-QCT-based BMC showed a modest correlation (r2\u202f=\u202f0.59, p\u202f=\u202f0.01) After applying ICON corrections, dcCt.Th resulted in 0.52\u202f±\u202f0.09\u202fmm for QCT and 0.43\u202f±\u202f0.07\u202fmm for HR-QCT, both significantly correlated to HR-pQCT (r2\u202f=\u202f0.75, p\u202f=\u202f0.0012 and r2\u202f=\u202f0.93, p\u202f<\u202f0.0001, respectively). The average overestimation bias of Ct.Th was reduced from (402\u202f±\u202f157)% to (45\u202f±\u202f17)% for QCT and from (330\u202f±\u202f69)% to (19\u202f±\u202f8)% for HR-QCT. Due to inaccurate segmentation uncorrected QCT-based Ct.Th measures as well as BMD and BMC showed no correlation to HR-pQCT and thus such bias cortical data can be misleading. The application of ICON reduced random overestimation bias to about 50\u202fμm and 20\u202fμm for QCT and HR-QCT, respectively, leading to a recovery of a significant correlation with the reference data of HR-pQCT. This reveals the potential for fairly accurate assessment of cortical thickness, allowing to better characterize cortical mechanical competence. These results warrant testing of the performance in vivo.