Densitometric and morphometric analysis of computed tomography (CT) data is an important measure of bone stability. StructuralInsight (SI) is our tool to access these parameters in an integrated and verified manner. It has been developed in our working group for over ten years and its capabilities are steadily extended.
StructuralInsight combines all mean quantitative CT data processing steps. It has been used in several medical trials and underwent extensive validation – these validation procedures are carried out regularly to guarantee consistency.
In a typical medical trail, the data has to be verified immediately upon reception by our team. For this purpose, a Quality Assurance module is part of SI. The overall scan quality and the scan parameters can be checked. The module can understand a variety of data structures, as they are generated by the different CT scanner manufacturer’s own steering software. It also transfers the data into an homogenous data structure and sorts it.
The Calibration is our initial step of actual data analysis. As our results are quantitative and in units of mg bone mass, we have to find the mathematical relation between bone mass and the CT scanner’s own unit Hounsfield Units (HU). This is done by evaluation volumes of known bone mass density, which are scanned together with the bone to be analyzed.
Segmentation is done to partition the bone into several volumes of interest (VOI). We have developed a standard vertebral segmentation consisting of 7 compartments, namely Spongiosa, Ellipse (within the Spongiosa), Vertical Cortex, Upper End Plate, Lower End Plate, Cut Pedicles and Foramen. This segmentation is created by aligning a triangulated network to the ridge of the bone by means of a semi-automatic 3D active snake algorithm or alternatively by a fully-automatic shape-driven algorithm. Moreover, geometric segmentation with arbitrary shapes and used-defined sites can be done.
To better access the cortex parameters, a layer-based segmentation has been implemented recently in combination with a convolution-based forward modelling of the Spongiosa-Cortex signal, which results in very realistic cortical thicknesses of around 0,35mm.
To align follow-up scans, mainly in the context of treatment effect monitoring, registration can be carried out. In that process, the follow-up scan is rotated and translated to coincide as good as possible with the baseline scan. Alternatively, the VOI segmentation of BL can be registered to the follow-up scan to avoid a resampling of the CT data.
Finally in the analysis, parameters like bone mineral density (BMD), bone mineral content (BMC, tissue mineral density (TMD) and tissue mineral content (TMC), bone volume to total volume (BV/TV), trabecular number (Tb.N), trabecular separation (Tb.Sp), trabecular thickness (Tb.Th), mean intercept length (MIL), cortical thickness (Ct.Th), vertebral height (Vt.Ht), cortical area (Ct.Ar) and several more can be computed for each VOI.
StructuralInsight is written in C++ using mainly Qt, ITK and VTK libraries amongst some others. It can be extended to specific needs on user request.
Figure 1: StructuralInsight graphical user interface. The vertebral bone segmentation is shown in red except for the Spongiosa area. An additional geometric segmentation is shown in green. Some quantitative results are shown in the Table of Results in the lower right corner. In the upper left corner, a layer-based assessment of bone mineral density (BMD) across the Cortex is shown.