Biomedical Engineering Reference
In-Depth Information
0.03
0.08
0.025
0.07
0.06
0.02
0.05
0.015
0.04
0.01
0.03
0.02
0.005
0.01
0
−
2.5
−
2
−
1.5
−
1
−
0.5
0
0
10
4
×
800
1000
1200
1400
1600
1800
2000
2200
2400
(a)
(b)
Figure 5.5:
(a) Histogram from a 3DRA image of a brain aneurysm. (b) Histogram from
a CTA image of a brain aneurysm. In the first case, the distributions from background-bone
(intensity values from
20,000 to
10,000 approximately) and vessels (intensity values from
−
−
10,000 to 0) are well differentiated. In the second case, the limit of the distribution from
background (intensity values from 1000 to 1200 approximately) is not differentiated from the
distributions of vessel (intensity values from 1100 to 1600) and bone (intensity values from
1100 to 2400) that are totally overlapped.
−
5.1.4
Chapter Outlook
This chapter presents a method to address the problem of brain aneurysm seg-
mentation in CTA images. Due to the limited resolution of CTA brain images,
the front is usually not able to evolve through narrow and twisted objects as,
for example, the thinnest brain vessels of the Circle of Willis. For this reason, a
two-stage algorithm is devised. In the first stage, a fast and rough segmentation
of all the tissues present in the image is obtained using a Maximum a posteri-
ori (MAP) classifier. In the second stage, a GAR method is used to obtain the
final segmentation with sub-voxel accuracy. The novelty of the method consists
in the use of differential image descriptors of high order, and the inclusion of
non-parametric information in the GAR model. This is done using a k-Nearest
Neighbor (kNN) classifier to estimate the underlying probability density func-
tions of the main tissue types that are present in the CTA images. The result is an
algorithm that provides accurate segmentations with very little user intervention
in the selection of its parameters.
The method has been evaluated on a database of 39 brain aneurysms
placed
within
the
Circle
of
Willis.
The
technique
is
compared
against
manual
measurements
of
three
geometrical
descriptors
of
the
aneurysm
