Information Technology Reference
In-Depth Information
before the registration. An example of our registration step is shown in
Fig.
35.
4. Compute the
final segmentation (f) using graph cuts: where the gray level
marginal densities of the VB and its background are approximated using
the proposed LCG model. Then we use a 3D graph where each vertex in
this graph represents a voxel in the VB volume. Then we de
ne the
weight of each edge as shown in Table
5
. After that, we get the optimal
segmentation surface between the VB and its background by
finding the
minimum cost cut on this graph.
2.5.5 Experimental Results
We test the proposed segmentation framework on CT of human lumbar and thoracic
spine data. The experiments are conducted on 30 data sets for which we have the
ground truths. The real data sets were scanned at 120 kVp and 2
:
5 mm slice
Fig. 35 The registration step. a The testing volume (or initially segmented volume) as shown with
the pink color. b The testing and 3D shape prior before the registration. The green and yellow
colors represent the object (
O
) and variability (
V
) regions, respectively. c The testing and 3D shape
prior after the registration. The new object and variability regions are named as
