Information Technology Reference
In-Depth Information
4 Image Registration
The next step of classification of perfusion map is describing (labeling) the visua-
lized brain tissues (in example detecting which pixels of image - if any - belongs
to anterior lobe). Because the fact that each perfusion map is generated from set of
“plain” CT images each CT image visualized same brain tissues as each perfusion
map. Knowing that it is possible to label tissues in perfusion map by labeling tis-
sues in CT image corresponding to it. Anatomic atlas (AA) approach for labeling
brain images is reliable and broadly used method [32]. The atlas is consisted of set
of brain CT slices templates, corresponding to them brain tissue labels and regis-
tration algorithm. The very important issue is to choose proper image registration
technique. The goal of registration algorithm is to minimize given error function
(maximize similarity function) between to images: the template image (also called
static image) and the newly acquired image (moving image). In case of brain atlas
brain CT slice template (moving image) is registrate on CT image that we want to
label (static image) generating transformation that “matches” those two images
(maximizing similarity function between them). The same transform is used to
deform brain labels. After applying transformation (which coefficients was deter-
mined during registration process) deformed labels describes the static image.
In case of registration of 2D brain images and with possibility of different gan-
try tilt between template and moving images it is not possible to use methods that
are based on detection some characteristic structural elements in brain images. To
overcome this author used intensity - based approach [32]. The correlation coeffi-
cient (CC) was used as function that measures similarity between images before
and after registration process.
n
m
(
)(
)

A
A
B
B
ij
ij
i
==
11
j
CC
(
A
,
B
)
=
(4.1)
(
)
(
)
n
m
n
m
2
2


A
A
B
B
ij
ij
i
==
11
j
i
==
11
j
A - average value of pixels colors.
n, m - size of the image (rows and columns).
Image registration of brain CT is a vital part of the DMD system. In the next
sections the author introduce and compare three algorithms: affine registration,
free form deformation (FFD) and Thirion's demons algorithm (TDA).
 
Search WWH ::




Custom Search