Improved assessment of cortical thickness from clinical CT

A method for cortical thickness estimation has been developed using an analytical deconvolution approach to extract the thin cortex structure from standard clinical CT data. After estimating the point-spread-function (PSF) of a scanner/kernel combination, the associated line-spread-function (LSF) and edge-spread function (ESF) have been analytically derived.

Fig. 1: Segmentation of the bone (a) generated by semi-automatic 3D segmentation of StructuralInsight and 3D visualization of a bone with the resulting equidistant layers (b).

Using sophisticated 3D segmentation, a radial bone mineral density (BMD) distribution anchored at the cortex segmentation was calculated. By fitting a combined LSF/ESF forward model to these distributions assuming full mineralization of 1200mgHA/cc in a compact cortex subregion of interest, we directly get an averaged de-convoluted cortical thickness dcCt.Th for it.

Fig. 2: 3D visualization of the estimated cortical thickness. For comparision, a representative slice from the analysed CT volumes is shown.

Advantages:

• Deconvolution-based estimation of Cortical Thickness in HRQCT data reduces the accuracy error from over 400% to 28%.
• Using the layer-based approach with sufficient thick layers, noise is reduced and the use of the forward model further stabilizes the procedure.
• The Cortex itself is used as a test body to estimate the scanner- and kernel-dependent point spread function.
• After segmentation, which is semi-automatic, the determination of the deconvoluted cortical thickness runs without user interaction.