ECTS Abstracts (2015) 1 P292

Improved Accuracy of Vertebral Cortical Thickness Based on CT Data of Different Quality (QCT, HR-QCT And HR-Pqct) by Means of Iterative Convolution Optimisation

Timo Damm, Pena Jaime, Bastgen Jan, Campbell Graeme, Barkmann Reinhard & Glüer Claus-Christian


Section Biomedical Imaging, Department of Radiology and Neuroradiology, University Hospital Schleswig-Holstein, Kiel, Germany.


Computed Tomography (CT) permits assessment of densitometric and structural bone characteristics. New therapies may not only improve the cancellous bone, but also strengthen the rather compact cortex, where this is achievable by endosteal or periosteal apposition, increasing mineralisation, or reduction in resorption space. In clinical CT data, limited spatial resolution and hence significant partial volume effects blur thin structures, especially the cortex, which is typically 150-350 μm thick (Ritzel 1997). We analysed CT data of 9 excised embedded vertebrae scanned using 3 different CT protocols: (I) HR-pQCT) (Scanco Medical Xtreme CT, voxel size (0,082*0,082*0,082)mm3, 60kV, 190mAs) used as gold standard, (II) HR-QCT (Siemens Sensation64, voxel size (0,156*0,156*0,300)mm3, 120kV, 360mAs) and (III) QCT (Siemens Sensation 64, voxel size (0,234*0,234*1,000)mm3, 120kVp, 100mAs). We developed a method for measuring radial BMD profiles perpendicular to the cortex in a laminar fashion based on an initial cortical segmentation done with StructuralInsight (in-house development). Using a priori knowledge of vertebral skeletal structure we can fit a BMD profile to the measured data, what results in a deconvolved cortical thickness measure (dcCt.Th). For comparison we also calculated a direct maximum-sphere based cortical thickness (Ct.Th) and, as a simple correction for partial volume effects, weighted cortical thickness (wCT.Th = cortical BMD x CT.Th). Cortical thickness by HR-pQCT was (370±70)μm. Compared with these results the table shows the mean offsets and the root mean square errors (RMSE) of QCT and HR-QCT based estimates. Here dcCt.Th shows very low random residual errors even in QCT data analysis. Our results document that HR-pQCT and QCT both overestimate the cortical thickness by 370% and 320%. This can be reduced for QCT to 40% (dcCt.Th =434±71)μm and for HR-QCT to 16% (dcCt.Th =522±93)μm. Variability of the accuracy error was 10% and 5% for QCT and HR-QCT.

Disclosure: The authors declared no competing interests.

Article tools

My recent searches

No recent searches.