ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 Journal of Telecommunication, Electronic and Computer Engineering
6
image histogram. The maximum peak is located at 0.1. Ater applying Gamma-law
transformation algorithm, the histogram has been enhanced in which the peak is
located at 0.4 as shown in Fig. 6c. On the other hand, the maximum peak is located
at 0.2 for contrast stretching as shown Fig. 6d.
γ
γ vel, r
γ sed on
The
prove es. Unlike
to a xel, the
ation 5
=
6 r this
mage es will have
0.1 0.2
0.3 0.4
0.5 0.6
1000 2000
3000 4000
5000 6000
7000
a Image and histogram of intensity normalization
0.1 0.2
0.3 0.4
0.5 0.6
100 200
300 400
500
b Image and histogram of background removal
0.2 0.4
0.6 0.8
50 100
150 200
250
c Image and histogram of Gamma-law transformation
0.2 0.4
0.6 0.8
50 100
150 200
250
d Image and histogram of contrast stretching
Fig. 6 Pre-processing steps
Fig. 6 Pre-processing steps
v. segmentation proCess using threshoLDing
teChniQue
For the image segmentation process, irstly the entire image is divided by 8 x
8 regions where 256 x 256 pixels of the entire image is split to 16 x 16 pixels size
in each region. Fig. 7 shows the image
spliting block processing with 16 x 16 pixels size per segment.
1 2
3 4
5 6
7 8
9 10
11 12
13 14
15 16
17 18
19 20
21 22
23 24
25 26
27 28
29 30
31 32
33 34
35 36
37 38
39 40
41 42
43 44
45 46
47 48
49 50
51 52
53 54
55 56
57 58
59 60
61 62
63 64
Fig. 7 Image splitting 16 x 16 pixels size per segment
= ⎩⎨
⎧ ≥
=
Fig. 7 Image spliting 16 x 16 pixels size per segment
Region 46 which is indicated by red circle is the lesion, examine by neuroradiologist.
Next, histogram is calculated at each region as shown in Fig. 8. The red circle
shows the histogram distribution of lesion, whereas the others are histogram
of normal brain area.
0.5 1
1000 2000
1 0.5
1 1000
2000 2
0.5 1
1000 2000
3 0.5
1 1000
2000 4
0.5 1
1000 2000
5 0.5
1 1000
2000 6
0.5 1
1000 2000
7 0.5
1 1000
2000 8
0.5 1
1000 2000
9 0.5
1 1000
2000 10
0.5 1
500 1000
11 0.5
1 500
1000 12
0.5 1
500 1000
13 0.5
1 1000
2000 14
0.5 1
1000 2000
15 0.5
1 1000
2000 16
0.5 1
1000 2000
17 0.5
1 1000
2000 18
0.5 1
200 400
19 0.5
1 50
20 0.5
1 20
40 21
0.5 1
200 400
22 0.5
1 1000
2000 23
0.5 1
1000 2000
24 0.5
1 1000
2000 25
0.5 1
500 1000
26 0.5
1 20
40 27
0.5 1
20 40
28 0.5
1 20
40 29
0.5 1
50 100
30 0.5
1 500
1000 31
0.5 1
1000 2000
32 0.5
1 1000
2000 33
0.5 1
500 1000
34 0.5
1 20
40 35
0.5 1
50 36
0.5 1
50 100
37 0.5
1 20
40 38
0.5 1
500 1000
39 0.5
1 1000
2000 40
0.5 1
1000 2000
41 0.5
1 500
1000 42
0.5 1
50 100
43 0.5
1 50
100 44
0.5 1
20 40
45 0.5
1 20
40 46
0.5 1
500 1000
47 0.5
1 1000
2000 48
0.5 1
1000 2000
49 0.5
1 1000
2000 50
0.5 1
500 1000
51 0.5
1 200
400 52
0.5 1
200 400
53 0.5
1 500
1000 54
0.5 1
1000 2000
55 0.5
1 1000
2000 56
0.5 1
1000 2000
57 0.5
1 1000
2000 58
0.5 1
1000 2000
59 0.5
1 1000
2000 60
0.5 1
1000 2000
61 0.5
1 1000
2000 62
0.5 1
1000 2000
63 0.5
1 1000
2000 64
Fig. 8 Histogram distribution of each region
= ⎩⎨
⎧ ≥
=
Fig. 8 Histogram distribution of each region
Two histograms which are normal and abnormal lesion are then generated. All
normal and abnormal regions are overlap respectively to ind the maximum number
of pixels at each intensity level. The maximum number of pixels is calculated
by using the function shown in equation 7.
n ,
m R
: 1
R MaxPixels
Pixelsn Max
= Where n is the intensity at level n, R1 and Rm are
⎩⎨ ⎧
≥ =
= 7
are regions in ⎩⎨
⎧ ≥
=
Where n is the intensity at level n, R1 and Rm are regions in the block histogram
according to each intensity level. The purpose of overlapping the histogram
ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 Brain Lesion Segmentation from Difusion-weighted MRI based on Adaptive Thresholding and Gray Level
7
for each block is to enhance the lesion for comparison with normal. This will
produce a new histogram as shown in Fig. 9. The optimal threshold is determined
by the ROI indicator, which the intensity level of normal histogram is reached zero
pixels.
=
0.1 0.2
0.3 0.4
0.5 0.6
0.7 0.8
0.9 1
5 10
15 20
25
Fig 9 Optimal threshold with Gamma-law enhancement
⎩⎨ ⎧
≥ =
Optimal Threshold
Normal
Abnormal
Fig 9 Optimal threshold with Gamma-law enhancement
Fig. 10 shows the maximum block histogram, which is done by overlapping
the histogram in all blocks including both normal and abnormal. By using
the proposed technique, we can clearly characterize the lesion area because it has
been enhanced.
sion, d at
the are
=
0.1 0.2
0.3 0.4
0.5 0.6
0.7 0.8
0.9 1
5 10
15 20
25
Fig 10 Maximum block histogram
⎩⎨ ⎧
≥ =
Lesion
Fig 10 Maximum block histogram
The statistical features representing the hyperintense and hypointense regions
are then calculated according to equation 8, where T
optimal
is the threshold value to obtain the segmentation.
= elsewhere
, for
1 int
, ⎩⎨
⎧ ≥
= optimal
T y
x I
ense hyper
y x
I
= ⎩⎨
⎧ ≥
= 8
vi. segmentation proCess using gLCm
Fig. 11 shows an example of GLCM for image in Fig. 1c. The image has
hyperintense lesion due to acute infarction. The GLCM is computed for
Ng=128, d=1 and at average orientations, and is represented in a contour plot.
Colour intensity represents the co- occurrence frequencies, or the number of
repetitions between each pixel pair, u and v. It can be seen that hypointense region
and CSF occur at smaller co-occurrence
entry; normal brain tissue is located in the middle of the matrix, while hyperintense
region exists at higher entry. The red dash-line shows a cross-section at u=v. At
this line, the co-occurrence frequency is the highest. The plot also shows that the
co-occurrence frequencies are diagonally symmetry and the gray level resolutions
are brighter when the co-occurrence
transitions increase of-diagonally with the matrix.
Averaged GLCM contour plot
gray level,v g
ra y
l e
v e
l, u
20 40
60 80
100 20
30 40
50 60
70 80
90 100
500 1000
1500 2000
2500
Fig. 11 GLCM for DWI in Fig. 1c
⎩ ⎨
⎧ ≥
≥ =
⎩ ⎨
⎧ ≤
≤ =
≤ ≤ ≤ ≤
≤ ≤
∑∑
= =
⎩ ⎨
⎧ =
+ +
+ =
φ
= +
= =
Hypointense region
Normal brain
tissue Hyperintense
region
T T
T
1
T
1
T
2
T
2
1 2
3 1, 2
1, 3 3, 2
3, 1 2, 1
2, 3
Fig. 11 GLCM for DWI in Fig. 1c
a. minimum and maximum threshold
