Segmentation Techniques

The followings are the internal segmentation techniques implemented by the toolkit.

Steps in Periosteal Surface Segmentation:

1. Denoise with a Gaussian smoothing filter.
2. Binarize the scan using global thresholds.
3. Label each separate bone structure differently using a connectivity component filter.
4. Extract one bone and paste it onto a sufficiently large blank image.
5. Inflate the bone, fill holes in the middle of that bone, and deflate the bone back to its original size. The inflation and deflation could be achieved by some morphological operations, but the Maurer Distance Map filter with a binary threshold is used instead to achieve the same effect. The Maurer Distance Map applies distance transformation and gives faster performance.
6. Paste the bone back to the image from Step 3.
7. Repeat Steps 4 to 6 for all other bone structures.

This contouring technique is based on an algorithm for segmenting cortical and trabecular compartments described elsewhere [1].

Steps in Cortical Break Detection

1. Denoise with a Gaussian filter and separate the bone from the background using global thresholds.
2. Obtain the periosteal surface.
3. The outer shell of the bone is depicted in red.
4. Obtain a cortical mask using morphological erosion. This shell has a constant thickness specified by the user (default for XtremeCT I is 4 voxels [2]).
5. This cortical mask is used to identify the cortex.
6. The bone within this shell is morphologically dilated by a user-specified distance (default for XtremeCT I is 1 voxel) to fill small cavities.
7. The image is inverted.
8. Only breaks that are connected to the periosteal and endosteal surfaces are selected.
9. The remaining cortical breaks are morphologically dilated to their original size.
10. The result is visualized by adding a transparent cortical mask.

This cortical break detection technique is based on an automatic algorithm by Peters et al [2].

To better visualize the cortical breaks, the underlying trabecular bone loss is segmented following these steps:

For XtremeCT images, voids inside the periosteal surface is selected using global thresholds; distance transformation keeps voids of large diameter; the voids are morphologically eroded; only voids connected to the cortical breaks from the previous steps are selected; the remaining voids are morphologically dilated to their original size and combined with the cortical breaks from the previous steps. This is based on the trabecular loss segmentation algorithm by Peters et al [3].

For cone beam CT images, a level set region growing operation is applied to the cortical breaks from the previous steps.

A list of seed points is generated using the resulting cortical break mask.

Steps in Erosion Segmentation:

1. Start with a greyscale image.
2. Obtain the periosteal surface.
3. Manually place a seed point in each erosion.
4. Select voids inside the periosteal surface using global thresholds.
5. Apply distance transformation to the voids and select voids of large diameter.
6. Morphologically erode the voids to remove connection.
7. Apply a connectivity filter to remove all voids not connected to the seed points.
8. Morphologically dilate the voids to their original size.
9. Apply a level set region growing operation to the voids.

This erosion segmentation technique is based on the Medical Image Analysis Framework (MIAF) [4] and a trabecular loss segmentation algorithm by Peters et al [3].

---

[1] Buie HR, Campbell GM, Klinck JA, MacNeil JA, Boyd SK. Automatic segmentation of cortical and trabecular compartments based on a dual threshold technique for in vivo micro-CT bone analysis. Bone (2007) 41:4. doi:10.1016/j.bone.2007.07.007

[2] Peters M, Scharmga A, de Jong J, van Tubergen A, Geusens P, Arts JJ, Loeffen D, Weijers R, van Rietbergen B, van den Bergh J. An automated algorithm for the detection of cortical interruptions on high resolution peripheral quantitative computed tomography images of finger joints. PLOS ONE (2017) 12:e0179138. doi:10.1371/journal.pone.0179138

[3] Peters M, de Jong J, Scharmga A, van Tubergen A, Geusens P, Loeffen D, Weijers R, Boyd SK, Barnabe C, Stok KS, et al. An automated algorithm for the detection of cortical interruptions and its underlying loss of trabecular bone; a reproducibility study. BMC Med Imaging (2018) 18:13. doi:10.1186/s12880-018-0255-7

[4] Toepfer D, Finzel S, Museyko O, Schett G, Engelke K. Segmentation and quantification of bone erosions in high-resolution peripheral quantitative computed tomography datasets of the metacarpophalangeal joints of patients with rheumatoid arthritis. Rheumatology Oxford (2014) 53:65-71. doi:10.1093/rheumatology/ket259