Multifeature Quantification of Nuclear Properties from Images of H&E-Stained Biopsy Material for Investigating Changes in Nuclear Structure with Advancing CIN Grade
Table 2
(1) Histogram features
1
Mean value
, where is the pixel intensity in position and is the total number of pixels
2
Standard deviation
3
Skewness
4
Kurtosis
(2) Co-occurrence matrix-based features
5
Angular second moment
, where is the number of gray levels in the image, , and is the co-occurrence matrix. ASM describes image smoothness and takes minimum values for smooth-textured nuclei. was calculated using the MATLAB function graycomatrix
6
Contrast
7
Inverse different moment
8
Entropy
9
Correlation
, where ,,, and are the respective mean values and standard deviations of and , described below:
10
Sum of squares
11
Sum average
, where is
12
Sum entropy
13
Sum variance
14
Difference variance
15
Difference entropy
, where is
16
Information measure of correlation 1
17
Information measure of correlation 2
, where , ,
(3) Run-length matrix-based features
18
Short-run emphasis
, where is the run-length matrix, is the number of gray values in the image, is the largest possible run,
19
Long-run emphasis
20
Gray-level nonuniformity
21
Run-length nonuniformity
22
Run percentage
, where is the total number of pixels in the image
(4) Wavelet-based features
23
dwt2H Mean Value
MATLAB function: mean(W(:)), where is the 2nd level dwt in the horizontal direction
24
dwt2H Median Value
MATLAB function: median(W(:))
25
dwt2H Max Value
MATLAB function: max(W(:))
26
dwt2H Min Value
MATLAB function: min(W(:))
27
dwt2H Range of Values
MATLAB function: range(W(:))
28
dwt2H Standard Deviation
MATLAB function: std(W(:))
29
dwt2H Median Absolute Deviation
MATLAB function: mad(W(:),1)
30
dwt2H Mean Absolute Deviation
MATLAB function: mad(W(:),0)
31–38
same as 23–30, where is the 2nd level dwt in the diagonal direction (MATLAB function dwt2)
39–46
same as 23–30, where is the 2nd level dwt in the vertical direction
(5) Tamura-based features
47
Tamura coarseness 1
, where m and n are region dimensions and , in which is the best scaling for highest neighborhood average
48–50
Tamura coarseness 2–4
Values of the 3-bin histogram of
51
Tamura contrast
, where is the standard deviation and a4 is the kurtosis
52
Tamura roughness
(6) Local binary pattern-based features
53
LBP mean
Mean value of the LBP histogram: , where and are, respectively, gray-level values of the central pixel and surrounding pixels in the circle neighborhood of radius and
54
LBP standard deviation
Standard deviation of the LBP histogram
(7) Morphological-based features
55
Nucleus area
MATLAB function: regionprops(BW, properties), where BW is the binary image nucleus and properties = ‘Area’
56
Nucleus perimeter
Where properties = ‘Perimeter’
57
Nucleus equivalent diameter
Where properties = ‘EquivDiameter’
58
Nucleus convex area
Where properties = ‘ConvexArea’
59
Nucleus major axis length
Where properties = ‘MajorAxisLength’
60
Nucleus minor axis length
Where properties = ‘MinorAxisLength’
61
Nucleus eccentricity
Where properties = ‘Eccentricity’
62
Nucleus solidity
Where properties = ‘Solidity’
63
Nucleus extent
Where properties = ‘Extent’
The parameters used for calculation of the abovementioned features were the following: (1) histogram features: the number of grayscale values = 256. (2) Co-occurrence matrix-based features: directions (0°, 45°, 90°, and 135°), interpixel distance = 1, and the number of grayscale values = 16. (3) Run-length matrix-based features: directions (0°, 45°, 90°, and 135°) and the number of grayscale values = 16. (4) Wavelet-based features: MATLAB function dwt2, Daubechies 2 transform, 2nd level coefficient matrices along the horizontal, diagonal, and vertical directions, and the number of grayscale values = 256. (5) Tamura-based features: k = 0 : 5. (6) Local binary pattern-based features: and . (7) Morphological-based features: MATLAB function regionprops.