Biomedical Engineering Reference
In-Depth Information
5.3. Strain Quantification Subsystem
After segmentation of the tumor contours from the continuous ultrasound
images, four features are defined for tumor labeling (benign or malignant): contour
difference (
C
d
), shift distance (
M
d
), area difference (
A
d
), and solidity (
S
o
). The
first three are related to the strain and the fourth to shape. It is important to
review data acquisition (see Section 3.2), because the neighboring images are very
similar and the variation in elasticity is very small, as only one of eight neighboring
images is used in this work. Therefore, only 8 of 60 total images are analyzed,
that is
I
s×
8+1
, where
s
= 0-7. We now discuss the concepts involved of the above
four features.
5.3.1. Contour difference (
C
d
)
The contour difference value is used to evaluate changes in lesion shape be-
tween two images. Because of compression, the contours between two neighboring
images
{
I
k
,I
j
}
will undergo some slight movement (see Leymarie [42]), where
{
, such that
s
= 0-6. Hence, these contours
between
I
k
and
I
j
should first be registered, so that the center point of the tumor
is used as the reference point. After registering the two images, the pixels in only
one contour of
I
k
or
I
j
are identified as the different pixels
P
d
.
The contour difference
C
d
(
I
k
,I
j
) between two neighboring images
I
k
,I
j
}
is the set
{
s
×
8+1
,s
×
8+9
}
I
k
,I
j
}
is defined as the number of different pixels divided by the number of pixels in the
image
I
k
:
{
N
P
d
(
I
k
,I
j
)
N
I
k
C
d
(
I
k
,I
j
)=
×
100%
,
(14)
where
N
P
d
is the pixel difference between the two registered contours, and
N
I
k
is the number of tumor pixels in the image
I
k
. Finally, for each breast case, the
average contour difference value is defined as
1
N
s
−
1
C
d
=
C
d
(
I
k
,I
j
)
,
(15)
(
k,j
)
where
N
s
is the number of image slices (
N
s
=8). For example, in Figure 14 the
tumor area in (a) is 12 pixels and the center pixel (black) is at position (4,4), while
in (b) the center is at position (5,5) after compression. After registering the two
tumor centers, the number of different pixels between (a) and (b) is 5 (slash pixels
in Figure 14c). Thus,
C
d
=5
/
12.
5.3.2. Shift distance (
M
d
)
We employ a motion estimation technique (see Richardson [43]) to find the
motion of two image contours. For example, in Figure 15, if there is a block
