Information Technology Reference
In-Depth Information
Fig. 27 The shape embedding and final segmentation results are shown in 3D views. a A CT data
is shown in the sagittal axis (without the refinement). b The initial location of the shape models.
2D shape models are propagated in z-axis to form 3D models. The blue color (outer volume)
shows the variability region, whereas the yellow color (inner volume) represents the object region.
c The shape model after registration. d The final segmentation results using the three models
the region labels, Potts model is used. The spatial pairwise interactions between two
neighboring voxels, which de
ne the potentials of Potts model, are estimated using
our new analytical approach [
37
]. The last step is integrating these region and
boundary properties as well as the shape information using a new energy function,
which is globally minimized using s/t graph cuts to get the optimal segmentation.
In this method, only the VB separation process is used as a pre-processing step.
2.5.1 Shape Modeling
We create a 3D shape of vertebral body a subset of VB data sets. This is done as
follows: The VBs
volumes, where each VB consists of 8 CT slices, are manually
segmented by a medical expert. The segmented VB slices are binary images, as
shown in the Fig.
30
. These segmented images are aligned to the ESP, which is
'
