a novel approach for self organisation based … · xiuming li et al.[8] states that the chan -vese...
TRANSCRIPT
A NOVEL APPROACH FOR SELF ORGANISATION
BASED SEGMENTATION OF MRI BRAIN IMAGES
E. SYNTHIYA JUDITH GNANASELVI1, Dr. M. MOHAMED SATHIK
2
1Research Scholar, Bharathiar University,Coimbatore, India 2Principal, Sadakathullah Appa College, Tirunelveli,India.
Abstract:
The detection of brain tumor in Magnetic Resonance Imaging (MRI) is very
significant but intricate task, in which different segmentation techniques get
evolved. In many of these techniques, segmentation is applied directly without any
boundary detection which takes long time, large number of iterations and yields
less accuracy result. This is a major drawback of existing segmentation
methodologies. The goal of this paper is to overcome this drawback and proposed a
novel approach for brain tumor segmentation. In this paper, there are three
instances. In the first instance, a boundary is detected from the given MRI image
and in the second instance, the segmentation techniques are applied within the
detected boundary for better accuracy in reduced time. For this purpose, SOM
method is used for predict a particular area of brain and then use geodesic, Chan
vese and region based ACM method for segmentation. In the third instance, the
morphological operation is carried out for removing the unwanted object and refills
the missing elements. The result is compared with the techniques such as geodesic,
Chan vese, region based ACM method, SOM & geodesic, SOM & Chan vese, SOM &
region based ACM method. The result shows that the proposed system works better
than existing techniques with the high accuracy, sensitivity, specificity, hammoude
distance, hausdorff distance and it has less border error and mean square error
than others.
Keywords: segmentation, Chan vese, SOM, region based ACM, Geodesic ACM.
II. Introduction:
International Journal of Pure and Applied MathematicsVolume 119 No. 12 2018, 2481-2500ISSN: 1314-3395 (on-line version)url: http://www.ijpam.euSpecial Issue ijpam.eu
2481
In current days, image processing is an fascinating research discipline and in
particular the medical image processing is more and more challenging field to deal
with diverse medical image types. In medical image analysis and interpretation,
automatic and exact detection and classification of tumors in brain MR images is
very significant. Tumors which are recognized and treated in the beginning time
gives preferred long haul survival over those identified of late. Prior days the tumor
was ascertained physically. Be that as it may, it will take long time and diminish
the precision. To overcome the drawback automatic brain segmentation was
implemented.
In image analysis phase, image segmentation is the initial and most decisive
task of image analysis. The objective of an image analysis is that of extracting
information from an image via image segmentation, object representation, and
feature measurement. The segmentation of medical image plays an significant task
in medical imaging applications. Segmentation can be applied in various areas of
medical field such as diagnosis, localization of pathobiology, examining anatomical
structure, treatment scheduling, and computer- incorporated surgery. However, the
unpredictability and the intricacy of the anatomical structures in the human body
have resulted a rigid setback in medical image segmentation[1].
There is lot of segmentation approaches followed. Many of these approaches
take large number of iterations, long time and thus reduce the accuracy of output
images. The goal of this paper is to overcome this drawback and proposed a novel
approach for brain tumor segmentation. The preeminent segmentation algorithm
alleviates to obtain suitable decision and endow with the finest treatment. This
paper proposes a novel approach in which SOM(Self Organising Map) method is
used for predict a particular area of brain and then use geodesic, Chan vese and
region based ACM method for accurate segmentation and classification of the brain
tumor from MR images.
Ayşe Demirhan et al. [2] developed an algorithm which is based on self-
organizing map (SOM) that is trained through unsupervised learning algorithm and
fine tuned by learning vector quantization (LVQ). In their work, they developed an
algorithm intended for clustering the SOM as an alternative of utilizing an
additional network. Input feature vector is constructed with the features acquired
from stationary wavelet transform (SWT) coefficients.
Vidhya et al. [3] offered a method which uses a SOM for segmentation that is
trained with unsupervised learning algorithm and fine-tuned with learning vector
quantization (LVQ). SOM method only considers the texture feature for
segmentation, it does not consider shape feature for increase accuracy. To resolve
this problem, the texture, shape based features are combined to classify the tumor,
edema, white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF).
Modified Fourier descriptor is used for extract the shape feature. The segmentation
accuracy of the proposed system is high compared to the existing system. Synthiya
International Journal of Pure and Applied Mathematics Special Issue
2482
Judith Gnanaselvi et al [4] compared the various segmentation approaches and
concluded that SOM method is better than other methods.
Mohammed M. Abdelsameaa et al. [5] concisely reviews a few region-based
ACMs, in which supervised and unsupervised image segmentation models are
distinguished. Mainly ACMs deal with shape priors in an unsupervised way owing
to the probable lack of a specific prior information on the shape of the objects to be
segmented because a smooth boundary is preferred in segmentation. This target is
accomplished by integrating a appropriate regularization term into the energy
functional.
Phyo Thant Thant Aung et al [6] depicted a segmentation method based on
region based active contour model in which the level set approach is utilized for
retrieving the region of interest (ROI) based image compression system. The intend
of this paper is to segment an image into region of interest, other than region of
interest and background for region based medical image compression system.
A. Shenbagarajan et al [7] proposed the region based Active Contour Method
(ACM) for segmentation method which offers high accuracy, and sensitivity,
specificity measures. Van-Truong Pham et al.[11] presented a method which
implants the Geodesic Active Contour (GAC) model into the region based method
because only the edge information is considered in the original Geodesic Active
Contour model. His paper demonstrate that the proposed region aided GAC method
is much more effective than the existing, especially when dealing with images with
holes, weak edges and noises.
Xiuming Li et al.[8] states that the Chan-Vese (CV) model, fails to segment
MR images with intensity inhomogeneity. So this paper proposed a new
segmentation method which overcomes the problems in traditional methods. Robert
Crandall [9] established the Chan-Vese algorithm for image segmentation which is
efficient on a variety of images. It is especially useful in cases where an edge-based
segmentation algorithm will not adequates, because it relies on global properties
such as region areas, graylevel intensities, contour lengths to a certain extent than
local properties such as gradients. This method is very useful for noisy and blurry
images. It is very slow for some applications depending on the type and size of the
image and the number of iterations needed, the segmentation can take several
seconds.
Mohammed M. Abdelsamea et al. [16] proposed a novel ACM which combines
the advantages of the Self-Organizing Map (SOM) within the level set framework of
a new unsupervised global ACM, the Chan–Vese (C–V) model. This model is
resilient to the extra noise. It can handle the images which include objects
portrayed by complex intensity distributions.Also it yields the high accuracy results.
The remainder of the manuscript is organized as follows. In section 3, we
present a system architecture for proposed method. In section 4, we present a quick
International Journal of Pure and Applied Mathematics Special Issue
2483
precis of a few strategies that already have been used for segmentation of brain
MRI images. In section 5, we proposed our novel approach for segmentation of
brain image that combines some existing methods with some new ideas which is
more vigorous than its individual components. In section 6, we present a
comparison of our consequences on a database of 25 brains, every of size
256×256×124 voxels, against segmentations generated by using experts. Finally we
conclude with a discussion of ability generalizations of our technique to phase other
systems for segmentation.
III. System Architecture:
The following system architecture is implemented in this proposed method.
Figure 1. Flow diagram of proposed method
IV. Existing Methods:
This section presents a few existing strategies that have been used for
segmentation of brain MR images.
IV.1. Geodesic ACM:
In geodesic ACM, the technique which is based on active
contours evolving in time according to intrinsic geometric measures of the image is
followed. The evolving contours naturally split and merge, allowing the
simultaneous detection of several objects and both interior and exterior boundaries.
The relation between active contours and the calculation of geodesics or minimal
distance curves is maintained. This geodesic approach for object segmentation
allows connecting classical “snakes” based on energy minimization and geometric
active contours based on the theory of curve evolution. Experimental results of this
I/P Image
Analysis
Boundary Detection Techniques
Segmentation Techniques
Preprocessing
International Journal of Pure and Applied Mathematics Special Issue
2484
geodesic ACM proves that it can be used for real images including objects with holes
and medical data imagery. Also the results may be elongated to 3D object
segmentation too.[11][12] Typical variational procedures for calculating geodesic
contours generate local minima of the energy which may be sensitive to
initialization. Highly desirable anisotropic formulations tend to be slower due to
increased computational burden. [13].
IV.2. Region Based ACM:
The region based ACM for segmentation seeks to drive the curves to
reach the boundaries of the input MRI brain images. This region-based ACM use
the region based descriptors to drive the curve which provide an efficient way for
segmentation in MRI brain image analysis.
This model works based on the hypothesis that the pixel regions of the image are
statistically homogenous. It deals fit with blur images, noisy images and images
with multiple holes, disconnected regions, etc. In MRI brain image analysis the
region based active contour model since considers global properties of images such
as contour lengths and MRI image pixel regions as against local properties such as
gradients. The energy minimizing function can be represented as:
ln P(Is | p) = ∫ ∫A0 Is(x,y)dA
Where Is (x, y) is the intensity at the pixel location (x, y) in the image, and the
integral gives the total area A enclosed by the curve p. As is evident, the region-
based information visually improved the segmentation quality compared to the one
using only gradient information.[14].
IV.3. Chan vese Model:
Chan-Vese model is a prevailing and flexible method which is able to segment
many types of images, including some that would be quite difficult to segment in
means of "classical" segmentation – i.e., using thresholding or gradient based
methods The model is based on an energy minimization problem, which can be
reformulated in the level set formulation, leading to an easier way to solve the
problem [15].
V. Proposed Method:
The proposed method consists of two steps. In the first step, a
boundary is detected from the given MRI image and in the second step,
the segmentation techniques are applied within the detected boundary.
V.1. Boundary Detection:
To detect the boundary in a brain image, the following algorithm is followed.
[10]
International Journal of Pure and Applied Mathematics Special Issue
2485
1. Input the MRI image in which consider image intensity value is neuron
2. Initialize the variables sigma, weight vector, winning neuron
Sigma=number of neighborhood pixels (8 or 24 or 48 or 80 or 120)
3. locate the neighborhood function using
Nf(j)=Img(j)-img(j+1)*sigma(j)
4. calculate the weight vector using
Wj(j+1)=wj+Nf(j)*img(j)-w(j)
5. compute the winning neuron using
Wn=max(wn,img(j)-w(img(j))
6. segmentation using the following:
Img(j)>=wn then
img(j)=1
Else
Img(j)=img(j)
V.2. Segmentation within the Boundary Area:
After defining the boundary, segmentation algorithms are applied within the
boundary area using the following algorithm[15].
Step 1: Give the input image.
Step 2: Create the customized boundary mask.
Step 3: Start the initial counter on the brain.
Step 4: Process the number of iterations on the infected brain in square and
elliptic form.
Step5: selecting the probable area of the brain.
Step6: Segmenting the probable area of the brain.
V.3. Refinement:
In the refinement stage, the morphological operation is performed for
eradicate the unnecessary object and replenishes the missing elements.
VI. Performance Analysis:
VI. 1. Experimental Dataset:
Experiments were carried out on a set of brain images to prove the
efficiency of the proposed scheme. For the experimental purpose, some standard 512
× 512 brain MRI images are chosen.
VI. 2. Performance Analysis:
To estimate the performance of all segmentation methods, several
performance metrics are available. This paper utilizes the performance metrics such
as Sensitivity, Accuracy, Border Error, Specificity, Hammoude Distance, Hausdorff
Distance, Mean Square Error and PSNR which were calculated using below
formulas:
International Journal of Pure and Applied Mathematics Special Issue
2486
• True Positive (TP): Abnormal brain correctly identified as abnormal.
• True Negative (TN): Normal brain correctly identified as normal.
• False Positive (FP): Normal brain incorrectly identified as abnormal.
• False Negative (FN): Abnormal brain incorrectly identified as normal.
Sensitivity:
Sensitivity also called the true positive rate or the recall rate in some fields
measures the proportion of actual positives.
Sensitivity = #(TP)
#(TP)+#(FN)
Accuracy:
The accuracy can be defined as the percentage of correctly classified
instances.
Accuracy = #(TP) + #(TN)
#(TP) + #(TN) + #(FP) + #(FN)
Border Error:
It is the ratio of the area covered by the XOR of segmented result (SR)
and ground truth (GT) images to the area covered by GT image.
Border Error = (FP)+(FN)
(TP)+ (FN)
Specificity:
Specificity measures the proportion of negatives which are correctly identified
such as the percentage.
Specificity = #(TN)
#(FP)+#(TN)
Hammoude Distance
It makes a pixel by pixel comparison enclosed by the two boundaries.
Hammoude Distance = #(FP)+#(FN)
#(TN)
Hausdorff Distance
The Hausdorff distance finds the largest distance between the boundary
points.
Hausdorff distance = max{maxid(gti, SR), maxid(sri, GT)}
Mean Square Error
International Journal of Pure and Applied Mathematics Special Issue
2487
The mean square error (MSE) is used to evaluate the difference between a 3D
image and the original 2D image. The MSE can be calculated by,
MSE = 1
n ∑(Yi − Yi)
2
n
i=1
where, Y is the 3D image and the Y is the original 2D image.
Peak Signal-to-Noise-Ratio
The peak signal-to-noise ratio (PSNR) is used to evaluate the quality between
the 3D image and the original 2D image. The PSNR formula is defined as follows:
PSNR = 10 × log 10255 × 255
1H × W
∑ ∑ [f(x, y) − g(x, y)]2W−1y=0
H−1x=0
dB
where H and W are the height and width of the image, respectively; and f(x,y) and
g(x,y) are the grey levels located at coordinate (x,y) of the original image and
attacked image, respectively.
VII. Results & Discussion:
We give comparison of the proposed algorithm to the other methods and
quantitative and qualitative experimental results obtained from the system in this
section. Segmentation algorithm developed in this study is compared to the well-
known algorithms like region based algorithm, geodesic algorithm, chan vese
algorithm.
Figure 2. Segmentation result of each processing step for a normal brain
MRI (a) Ground truth image.
International Journal of Pure and Applied Mathematics Special Issue
2488
(b to j) Segmented image using various algorithms (b) SOM, (c)
geodesic ACM, (d) Chan Vese,
(e) Region Based ACM, (f) SOM and geodesic ACM, (g) SOM and
Chan Vese, (h) SOM and
Region Based ACM, (j) proposed method.
Figure 3. Segmentation result of each processing step for a abnormal brain MRI (a)
Ground truth image,
(b to j) Segmented image using various algorithms (b) SOM, (c) geodesic
ACM, (d) Chan Vese,
(e) Region Based ACM, (f) SOM and geodesic ACM, (g) SOM and Chan
Vese, (h) SOM and Region
Based ACM, (j) proposed method.
In figure 2 and 3, the normalized image and smoothened image is shown.
They are displayed after the pre processing step. The pre processing is carried out to
remove the unnecessary noise and increase the contrast of the image by using
Gaussian filter.
The visual result of each processing step for the normal and abnormal MRI
brain image which shows the segmented area of the brain most clearly in figure 2
and figure 3 respectively. Figure 2(a) and 3(a) shows the ground truth image used
for segmentation. The segmented images using various algorithms such as Geodesic
ACM, Chan vese and Region based ACM, SOM & Geodesic ACM, SOM & Chan vese
and SOM & Region based ACM are shown from Figure 2(b) to Figure 2(h) and
Figure 3(b) to Figure 3(h). Figure 2(i) and figure 3(i) shows the segmented image
using proposed method.
International Journal of Pure and Applied Mathematics Special Issue
2489
Overall Performance
Algorithms Sensitivity Accuracy B_Error Specficity Hamm_D Haus_D MSE PSNR
SOM 38.638 91.063 79.952 97.764 10.336 4.995 0.089 58.697
Geodesic ACM 36.821 84.984 132.054 91.127 18.807 5.385 0.150 56.500
Chan vese 71.224 94.973 43.117 98.165 5.837 4.228 0.050 61.514
Region Based 51.659 92.747 63.580 98.032 8.411 4.515 0.073 59.856
SOM Geodesic
ACM 50.787 86.372 120.151 90.906 17.115 5.469 0.136 56.908
SOM Chan vese 71.009 95.031 42.561 98.271 5.765 4.166 0.050 61.599
SOM Region
Based 73.827 95.499 38.664 98.410 5.213 4.029 0.045 62.020
Proposed 75.098 95.826 35.747 98.631 4.823 4.004 0.042 62.388
Table 1. Average Performance evaluation for different segmentation algorithms for
25 MRI brain images
In Table I, performance of the developed algorithm is compared to the other
algorithms for 25 MR images of the patient. From the scores in Table I, it can be
seen that that our proposed algorithm performs better than other methods. It
clearly shows that on comparing with SOM method, the accuracy value and the
PSNR value of the proposed method increased by 4.763 and 3.691 respectively. Also
the MSE value of the proposed method decreased by 0.047. On comparing with
Geodesic ACM, the proposed method increased with the accuracy value of 10.842,
the PSNR value of 5.888 and also it is decreased with the MSE value of 0.108. Like
wise on comparing with Chan vese method, the accuracy value of the proposed
method increased by 0.853, the PSNR value increased by 0.874 and the MSE value
decreased by 0.008. Also on contrasing with region based ACM, the proposed
method got the increased accuracy value and PSNR value by 3.079, 2.532
International Journal of Pure and Applied Mathematics Special Issue
2490
correspondingly and it is reduced by 0.031. It obviously demonstrates that on
contrasting with SOM & geodesic ACM technique, the proposed method increased
with the accuracy value of 9.454, the PSNR value of 5.48 and also it is diminished
with the MSE value of 0.094. It absolutely shows that on comparing with SOM &
Chan vese method, the accuracy of the proposed method increased by 0.795, the
PSNR value increased by 0.789 and the MSE value reduced by 0.008. Also it
undoubtedly proves that the proposed method acquired the increased accuracy
value of 0.327, PSNR value of 0.368 and the reduced value of 0.003 on comparing
with SOM & region based method.
Sensitivity Performance
Sampl
e
Image
SOM Geo_
ACM
Chan
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image
1
37.24
8
51.99
0
74.20
2
80.49
1 52.039 75.135 79.951 83.047
Image
2
35.01
8
55.40
5
74.20
7
70.32
9 47.709 74.148 76.146 76.322
Image
3
45.50
9
49.07
8
76.97
8
71.86
2 50.089 77.811 79.120 80.845
Image
4
52.08
9
35.04
2
73.03
6
39.88
9 60.501 71.421 74.373 76.323
Image
5
51.16
8
36.37
4
71.02
3
39.87
8 55.895 69.577 73.582 75.362
Table 2. Sensitivity evaluation for segmentation algorithms
Accuracy Performance
Sampl
e
Image
SOM Geo_
ACM
Cha
n
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image
1
90.6
9
85.4
6
95.5
4 96.51 85.24 95.77 96.37 97.08
Image
2
90.7
5
88.6
2
95.4
7 94.97 86.76 95.51 95.77 95.92
Image
3
91.0
8
87.8
5
95.8
5 95.24 87.62 96.04 96.45 96.77
Image
4
92.4
1
86.5
1
95.4
3 91.90 89.31 95.34 95.84 96.22
Image
5
92.4
4
86.6
9
95.1
3 91.90 88.80 95.07 95.64 96.08
Table 3. Accuracy evaluation for segmentation algorithms
International Journal of Pure and Applied Mathematics Special Issue
2491
Border Error Performance
Sampl
e
Image
SOM Geo_
ACM
Chan
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image
1
74.93
9
117.10
1
35.92
1
28.10
8 118.870 34.054 29.189 23.538
Image
2
89.07
2
109.51
8
43.59
6
48.41
4 127.438 43.243 40.717 39.248
Image
3
86.97
2
118.38
2
40.45
2
46.40
1 120.643 38.549 34.622 31.469
Image
4
69.24
8
123.12
0
41.67
1
73.92
8 97.605 42.563 37.939 34.485
Image
5
68.91
0
121.24
6
44.38
3
73.80
4 102.058 44.883 39.711 35.762
Table 4. Border Error evaluation for segmentation algorithms
Specificity Performance
Sampl
e
Image
SOM Geo_
ACM
Chan
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image
1
98.27
2
90.20
1
98.56
4
98.78
0 89.944 98.697 98.704 99.066
Image
2
97.20
8
92.47
4
97.93
6
97.82
7 91.289 97.984 98.045 98.195
Image
3
96.28
7
92.28
7
98.00
7
97.91
2 91.913 98.130 98.429 98.592
Image
4
97.37
5
92.84
4
98.19
0
98.30
0 92.851 98.280 98.485 98.670
Image
5
97.52
5
92.89
7
98.10
1
98.31
4 92.856 98.218 98.361 98.629
Table 5. Specificity evaluation for segmentation algorithms
Hammoude Distance Performance
Sampl
e
Image
SOM Geo_
ACM
Cha
n
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image 10.81 18.41 5.16 4.036 18.743 4.893 4.194 3.370
International Journal of Pure and Applied Mathematics Special Issue
2492
1 5 2 9
Image
2
10.62
2
13.72
9
5.16
0 5.737 16.183 5.116 4.814 4.633
Image
3
10.32
7
14.66
6
4.71
9 5.418 15.007 4.491 4.022 3.649
Image
4 8.750
16.31
6
5.22
2 9.253 12.934 5.329 4.740 4.300
Image
5 8.710
16.08
9
5.57
7 9.254 13.548 5.633 4.977 4.470
Table 6. Hammoude Distance evaluation for segmentation algorithms
Hausdorff Distance Performance
Sample
Image SOM
Geo_
ACM
Cha
n
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image1 5.09
9
5.38
5
4.00
0 3.606 5.568 3.873 3.742 3.606
Image2 5.19
6
5.29
2
4.79
6 4.359 5.292 4.690 4.359 4.123
Image3 5.09
9
5.29
2
4.24
3 4.359 5.745 3.873 3.873 3.606
Image4 5.09
9
5.09
9
4.12
3 4.690 5.477 4.123 4.000 4.000
Image5 5.19
6
5.38
5
4.24
3 4.899 5.477 4.472 4.359 4.123
Table 7. Hausdorff Distance evaluation for segmentation algorithms
MSE Performance
Sample
Image SOM
Geo_
ACM
Cha
n
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image1 0.09
3
0.14
5
0.04
5 0.035 0.148 0.042 0.036 0.029
Image2 0.09
3
0.11
4
0.04
5 0.050 0.132 0.045 0.042 0.041
Image3 0.08
9
0.12
1
0.04
2 0.048 0.124 0.040 0.036 0.032
Image4 0.07
6
0.13
5
0.04
6 0.081 0.107 0.047 0.042 0.038
International Journal of Pure and Applied Mathematics Special Issue
2493
Image5 0.07
6
0.13
3
0.04
9 0.081 0.112 0.049 0.044 0.039
Table 2. MSE evaluation for segmentation algorithms
PSNR Performance
Sampl
e
Image
SOM Geo_
ACM
Chan
vese
Regio
n
SOM
_Geo_AC
M
SOM_Cha
n vese
SOM
_Regio
n
Propose
d
Image
1
58.44
2
56.50
4
61.63
6
62.70
1 56.439 61.868 62.537 63.472
Image
2
58.46
8
57.57
1
61.57
1
61.11
6 56.912 61.606 61.868 62.027
Image
3
58.62
6
57.28
7
61.95
0
61.35
4 57.204 62.159 62.626 63.040
Image
4
59.33
0
56.83
1
61.53
6
59.04
6 57.840 61.444 61.944 62.358
Image
5
59.34
4
56.89
0
61.25
5
59.04
6 57.639 61.206 61.738 62.193
Table 2. PSNR evaluation for segmentation algorithms
The table 2 shows the sensitivity evaluation of 5 MRI brain images. In table
2, for example we take image 3 which shows the sensitivity difference of 30.756 in
SOM & geodesic ACM, 3.034 in SOM & chan vese and 1.725 in SOM & region based
ACM on compared with the proposed method. The other images are also shows the
sensitivity difference similar to image3. The table 3 illustrates the accuracy
estimation of 5 MRI brain images. In table 3, for instance the image 3 which shows
the accuracy difference of 9.15 in SOM & geodesic ACM, 0.73 in SOM & chan vese
and 0.32 in SOM & region based ACM on compared with the proposed method. The
other images are also shows the accuracy difference similar to image3.
The border error rate evaluation and MSE evaluation is shown in table 4 and
table 8 respectively. For example an image 3 is chosen for evaluation of both
performance metrices. The border error rate value and MSE value decreased by
89.174, 5.836 respectively in SOM & geodesic ACM, 7.08, 0.881 respectively in
SOM & chain vase and 3.153, 0.414 in SOM & region based ACM on comparing
with the proposed method. The other image is also shown like this. The table 5
displays the specificity analysis. It obviously demonstrates that in image 3, the
proposed method offes the increased specificity vaue of 0.679 than SOM & geodesic
ACM technique, 0.462 than SOM & chan vese and 0.163 than SOM & region based
ACM. The other images are also displays the specificity difference on comparing
with the proposed method.
The table 6 and table 7 illustrates the evaluation of Hammoude distance and
Hausdorff distance. We choose image 3 for example. For both the Hammoude
distance and Hausdorff distance evaluation, the proposed method decreased by the
International Journal of Pure and Applied Mathematics Special Issue
2494
value of 11.358, 0.753 respectively than SOM & geodesic ACM, 0.842, 2.139 than
SOM & chan vese and 0.373, 0.267 than SOM & region based ACM.
The table 8 demonstrates the PSNR value analysis of 5 MRI brain images.
For instance an image 3 is chosen for analysis. It clearly shows that on comparing
with SOM & geodesic ACM, SOM & chan vese and SOM & region based ACM , the
PSNR value of the proposed method increased by 5.836, 0.881 and 0.414
respectively.
International Journal of Pure and Applied Mathematics Special Issue
2495
International Journal of Pure and Applied Mathematics Special Issue
2496
Figure 4: Graphs showing the performance analysis of segmentation with respect to
measurement parameters for all the 20 MRI brain images.
Conclusion:
Segmentation of MRI data is a challenging and time consumption task.
Segmentation techniques must afford efficient and accurate results for the purpose
of analysis and diagnoising the abnormalities present in the medical image.
Generally, boundaries of different tissues in MRI brain images are not clear because
white and gray matter intensities were very close which makes the boundary
identification to be difficult. We have proposed a method that can overcome the
difficulty of accuracy and diagnosis time and results with increased yield. By
comparing the results, our proposed method has higher performance and it offer
better accuracy (atleast 10% higher) and least Mean Square Error in comparison to
other models. Hence, the proposed method yields the better segmentation results
than the existing methods.
References:
International Journal of Pure and Applied Mathematics Special Issue
2497
1. Iffat Zahra Rizvi, Syed Zuhair Haider, Skin Texture based Smart Detection of
Skin infections
by Image Segmentation Using Hybrid DWT and Watershed Transform,
International Journal
of Research and Development in Applied Science and Engineering (IJRDASE)
ISSN: 2454-
6844.
2. Ayşe Demirhan, Mustafa Törü, İnan Güler, Segmentation Of Tumor And Edema
Along With
Healthy Tissues Of Brain Using Wavelets And Neural Networks, IEEE Journal
of Biomedical
and Health Informatics, DO10.1109/JBHI.2014.2360515.
3. P.Vidhya, Dr.B.Rajeshkumar, B.Tamilselvi, A Novel Approach For Segment
Tumor And
Edema Along With Healthy Tissues Of Brain, International Journal of Advanced
Research in
Biology Engineering Science and Technology (IJARBEST), Vol. 2, Special Issue
10, March
2016.
4. E. Synthiya Judith Gnanaselvi, M. Fathima Zahira, Dr. M. Mohamed Sathik,
Comparative
Study On Brain Tumor Segmentation Techniques, International Journal of
Advanced Research
in Science and Engineering, Vol. No. 5, Issue No. 09, September 2016.
5. Mohammed M. Abdelsameaa, Giorgio Gneccoa,n, Mohamed Medhat Gaberb, An
efficient
Self-Organizing Active Contour model for image segmentation, ScienceDirect,
journal
homepage: www.elsevier.com/locate/neucom
6. Phyo Thant Thant Aung, Aung Soe Khaing, Hla Myo Tun, MR Brain Image
Segmentation
Using Region Based Active Contour Model, International Journal Of Scientific &
Technology
Research Volume 5, Issue 06, June 2016.
7. A. Shenbagarajan, V. Ramalingam, C. Balasubramanian and S. Palanivel, Tumor
Diagnosis in
MRI Brain Image using ACM Segmentation and ANN-LM Classification
Techniques
Indian Journal of Science and Technology, Vol 9(1),
DOI:10.17485/ijst/2016/v9i1/78766,
January 2016.
8. Xiuming Li, Dongsheng Jiang, Yonghong Shi,and Wensheng Li, Segmentation of
MR image
International Journal of Pure and Applied Mathematics Special Issue
2498
using local and global region based geodesic model, Biomed Eng Online. 2015;
14: 8,
Published online 2015 Feb 19. doi: 10.1186/1475-925X-14-8.
9. Robert Crandall, Image Segmentation Using the Chan-Vese Algorithm,
Geometric Models
and Level Set Methods, ECE 532 Project Fall, 2009.
10. T. Logeswari and M. Karnan, An improved implementation of brain tumor
detection using
segmentation based on soft computing, Journal of Cancer Research and
Experimental
Oncology Vol. 2(1) pp. 006-014, March, 2010.
11. Vicent Caselles, Ron Kimmel, Guillermo Sapiro, Geodesic Active Contours,
International
Journal of Computer Vision 22(1), 61–79 (1997), ©1997 Kluwer Academic
Publishers.
Manufactured in The Netherlands.
12. Albert Huang, Rafeef Abugharbieh, Roger Tam, Anthony Traboulsee, Brain
Extraction
Using Geodesic Active Contours, Medical Imaging 2006: Image Processing, doi:
10.1117/12.654160, Proc. of SPIE Vol. 6144 61444J-1.
13. Boykov, Kolmogorov, Computing geodesics and minimal surfaces via graph
cuts, Ninth
IEEE International Conference on Computer Vision, 2003.
14. A. Shenbagarajan, V. Ramalingam, C. Balasubramanian and S. Palanivel,
Tumor Diagnosis
in MRI Brain Image using ACM Segmentation and ANN-LM Classification
Techniques,
Indian Journal of Science and Technology, Vol 9(1),
DOI:10.17485/ijst/2016/v9i1/78766,
January 2016.
15. Payal Gupta, Er. Satinderjeet Singh, Review Paper On Brain Image
Segmentation Using
Chan-Vese Algorithm And Active Contours, International Journal of Advanced
Research in
Computer Engineering & Technology (IJARCET) Volume 3 Issue 11, November
2014.
16. Mohammed M. Abdelsamea, Giorgio Gnecco and Mohamed Medhat Gaber, A
SOM-based
Chan–Vese model for unsupervised image segmentation, Methodologies And
Application,
Soft Comput (2017) 21:2047–2067 DOI 10.1007/s00500-015-1906-z.
International Journal of Pure and Applied Mathematics Special Issue
2499
2500