Biomedical Engineering Reference
In-Depth Information
Let us introduce some notation for the definition of the features:
M
(
i, j
) is the (
i, j
)
th
element of a normalized co-occurrence matrix:
M
x
(
i
)=
j
M
(
i, j
)
,
M
y
(
j
)=
i
M
(
i, j
)
,
=
i
i
j
M
(
i, j
)=
i
µ
x
iM
x
(
i
)=
E
{
i
}
,
=
j
j
i
M
(
i, j
)=
j
µ
y
jM
y
(
j
)=
E
{
j
}
.
With the above notation, the features can be written as follows:
=
i,j
M
(
i, j
)
2
,
Energy
Entropy
=
−
M
(
i, j
)
logM
(
i, j
)
,
i,j
=
i,j
1
1+(
i
IDM
M
(
i, j
)
,
−
j
)
2
=
i,j
µ
y
)
3
M
(
i, j
)
,
Shade
(
i
+
j
−
µ
x
−
=
i,j
(
i
−
j
)
2
M
(
i, j
)
,
Inertia
=
i,j
µ
y
)
4
M
(
i, j
)
.
Prominence
(
i
+
j
−
µ
x
−
In this way, for each measure
m
k
:
2
→
2
we obtain an image according to each
particular feature of the co-occurrence matrix. Therefore, if we are computing the
four orientations, given a single distance
D
we obtain 24 different filtered images.
Those images are the possible inputs to the STOP andGOactivemodel formulation.
Figure 5 shows different filter outputs using the co-occurrence measures in a
polar representation of the IVUS images. As we can observe, not all the figures
are suitable for use as inputs of the deformable model. Note that Figure 5b,c do
not follow the guidelines for the filter to be used, since the regions of interest (the
textured tissue) is not defined by high values. Note also that the rest of the images
comply with this requisite in different ways.
