Biomedical Engineering Reference
In-Depth Information
3.0 10
4
2.5 10
4
2.0 10
4
1.5 10
4
1.0 10
4
5.0 10
3
0
5.0 10
3
−
0
500
1000
1500
2000
2500
Intensity
Figure 13.2:
A typical histogram for an MR image.
background region first. Then all the background region voxels are put into one
bin, while the foreground region voxels are binned using a binning technique.
Two-level multiresolution registration method is applied next. For binning we
have experimented three techniques.
13.2.1.1
Region-growing
Region-growing is an approach to image segmentation that has received con-
siderable attention in the computer vision segment of the artificial intelligence
community [14]. The basic approach is to start with a set of “seed” points and
from these grow regions by appending to each seed those neighboring voxels
that have properties similar to the seed [15].
In our approach, a seed point is selected automatically near the left-upper
corner of the given CT and MR images. This is based on the observation that
there is always a large background area and that the object is always centered.
This seed point is used to begin region-growing for the background. There
are two criteria for a voxel to be annexed to the background: (1) the absolute
intensity difference between the voxel and the seed must be less than a threshold;
and (2) the voxel must be adjacent to at least one voxel in the background. The
threshold for the region is determined by the histogram, see Figs. 13.1 and 13.2.
In the figures the spike at the left delimits the background and so the valley to
