268
F. Shi et al.
Table 10.8 The features extracted in the feature extraction stage
Feature
number
Feature
name
Details
1
Normalized
intensity
I normalized (i, j, k)
Ibilater_filtered(i, j,k)−I min
I max −I min
(i, j, k) represents the (i, j, k) th voxel in the VOIs.
2
Block mean M block (i, j, k)
1
125
i+2
li−2
j+2
m j−2
k+2
nk−2
I normalized (l, m, n)
3
Block
standard
deviation
ST D block (i, j, k)
i+2
li−2
j+2
m j−2
k+2
nk−2
(Inormalized(l,m,n)−Mblock(i, j,k)) 2
124
4, 5
Absolute
intensity
difference
in 13
directions
AI D(i, j, k)
13 directions
|I normalized (i, j, k) − I normalized (l, m, n)|
l i, i + step; m j, j + step; n k, k + step; step 1, 2
6–18
GLCM
based
contrast (in
13
directions)
Contrast direction m (i, j, k)
Ng
x1
Ng
y1
(x − y)
2 p(x, y)
m 1, 2, . . . , 13
p(x, y) represents the (x, y)th entry in the GLCM, N g represents the
number of distinct grey levels in the quantized image. Here, N g 8.
19–31
GLCM
based
correlation
(in 13
directions)
Correlation direction m (i, j, k)
Ng
x1
Ng
y1
(x − μ x )(y − μ y ) p(x, y)
σ x σ y
m 1, 2, . . . , 13
μ x , μ y , σ x and σ y are the means and standard deviations of the x row
and the y column, respectively.
32–44
GLCM
based
energy (in
13
directions)
Energy direction m (i, j, k)
Ng
x1
Ng
y1
p
2 (x, y)
m 1, 2, . . . , 13
45-57
GLCM
based
homogeneity (in 13
directions)
Homogeneity direction m (i, j, k)
Ng
x1
Ng
y1
p(x, y)
1 + |x − y|
m 1, 2, . . . , 13