Information Technology Reference
In-Depth Information
Fig. 38 Precision errors of BMD measurement of each method. The BMD measurement of our
method has the lowest error and standard deviation. VBs are subsequently segmented using the
graph cuts without shape constrained (A1), statistical level sets (A2) methods, and the b-spline
based interpolation (A3)
Bone Mineral Density Measurements
The main goal of this work is to accurately separate VB from around processes in
order to obtain the BMD measurements with high trueness and precision from
volumetric CT datasets. Thirty volmetric VBs from thoracic and lumbar spine are
used in our experiments. For comparison purposes, the BMD measurements for
each segmentation method are obtained. The relative errors in BMD measurements
for each method are shown in Fig.
38
. Our proposed method achieves an average
BMD precision error %3.03. This re
fl
ects how accurate the proposed segmentation
approach is.
2.6 Segmentation Using PCA-based Shape Model,
Gaussian-based Intensity Model, and Asymmetric Gibbs
potential-based Spatial Interaction Model
In this section, we present another idea to segment vertebral bodies. As pre-pro-
cessing steps, we use the Matched
filter to detect the region of interest and manual
VB separation. In the second phase, we obtain initial labeling (f*) using the graph
cuts which integrates the intensity and spatial interaction models. Finally, we
register the initial labeled image and the shape priors to obtain the optimum
labeling. To obtain the shape priors, we use the 2D-PCA on all training images.
Figure
39
summarizes the main components of our framework. The following
sections give more details about the shape model construction and the segmentation
method. This method generally works in two dimensional space.