The morphological escalator – a new technique for automatic trabecular-cortical separation

Bone morphometry in the mouse and rat preclinical models requires separation of the medullary trabecular bone from surrounding cortical bone at the endosteal boundary. This can be done by manual drawing of region of interest (ROI) shapes. However advanced image processing methods also allow this trabecular delineation to be automated. A method note already exists, MN008, entitled “Automated trabecular and cortical bone selection”, which shows how morphological (erode, dilate, open, close), despeckle and bitwise (Boolean) image operations can be combined to effectively separate trabecular from cortical bone at the rodent metaphyseal medulla.

Here we introduce a modified version of this method using a technique called the “morphological escalator”. This addresses a difficult challenge to automatic trabecular delineation where parts of the cortical wall are highly porous with only thin structures. Such regions include peripheral growth plate remnants that persist some distance downstream of the growth plate, especially in young fast-growing rodents. Such regions are shown in figure 1 at three locations in a mouse femur metaphysis cross-section.

Trabecular bone auto-separation methods are based on morphological operations: erode (remove a pixel layer from surface), dilate (add a pixel layer to surface), open (erode then dilate) and close (dilate then erode). The medullary space can be binarised, and gaps from trabecular structures closed out, to make a trabecular volume of interest (VOI). What complicates this, however, is overlap in thickness between trabecular and cortical bone. This means that a single combination of opening and closing sometimes does not cleanly separate trabecular bone, but creates artefacts. 

A solution to this is the “morphological escalator”. This is a series of open-close operations, starting with a voxel radius of 1, then repeating with a successively increasing voxel radius up to a value depending on the thickness of the trabecular structures (open 1 pixel; close 1 pixel; open 2 pixels; close 2 pixels, etc.). In figure 1 the effects of opening and closing are shown to a radius of 12 voxels, comparing a single open-close morphological operation (a) to the stepwise morphological escalator (b).

The morphological escalator appears in figure 1 to do quite a good job of excluding from the trabecular VOI peripheral remnants of growth plate – places in the cortical boundary which are fine textured and highly porous. This restricts the analysis to mature secondary trabecular bone. Other trabecular-cortical separation algorithms, while otherwise effective, do not always succeed in excluding these peripheral growth plate remnants. Figure 2 illustrates how the morphological escalator works. Regions become progressively solidified as white or black depending on the local percentage volume and relative thickness of bone or soft tissue spaces.

Image of the Month

Bruker microCT 2019 User Meeting