International Journal of Engineering Trends and Technology (IJETT) – Volume 10 Number 7 - Apr 2014
A Comparison between ROI Extract and FCM Algorithms in Segmentation of
Tumor Region from MR Brain Images
A.Almusthaliba#1,G.Vishnuvarthanan*2,S.Meenakshi#3
#PG student, Department of ICE,
Kalasalingam University, Tamil Nadu, India
*Assistant Professor – I, Department of ICE
Kalasalingam University, Tamil Nadu, India
#
Assistant Professor, Department of EIE
Sethu Institute of Technology, Tamil Nadu, India
Abstract— Segmentation of brain images based on
Magnetic Resonance (MR) is used to segment and analyse
the slices of brain to indicate the presence of tumour.
Region of interest (ROI) extract is used to select the exact
spot in an image where the tumour is present. Primary
area of interest (tumour region) is chosen by the
algorithm. The chosen region is completely extracted out
from the input image.Fuzzy c-means algorithm is a soft
clustering method in which the data elements (pixel values)
are divided and clustered into two or more clusters. FCM
is carried out by clustering the similar pixel values into one
cluster and dissimilar pixel values into another cluster. A
comparison is carried out between these two algorithms
with the help of parameters such as Mean Square Error
(MSE), Peak Signal to Noise Ratio (PSNR), segmentation
accuracy, Jaccard index, computational time and memory
requirement for processing the algorithm. The efficiency
of either algorithm is proved using the comparison
parameters
Keywords— Magnetic resonance (MR) brain images,
Region-of-interest (ROI)and Fuzzy C-means (FCM)
I. INTRODUCTION
Magnetic Resonance Imaging (MRI) provides clear
images of living tissues and is used for both brain tissues and
human body studies. Data obtained from MR images is used
for detecting tissue deformities like cancer and injuries; MR is
also mainly used in the studies of brain pathology, where
Regions of Interest (ROI) are often examined in detail, for
example in multiple sclerosis (MS) studies. In order to obtain
good quantitative studies, ROI’s within the brain must be well
described. In conventional methods, a skilled operator
manually outlines the ROI’s with the use a mouse or cursor.
Recently, computer-assisted methods have been used for
specific tasks such as extraction of MS lesions from MRI
brain scans, or extraction of the cerebral ventricles in
schizophrenia studies. Many of these computer-assisted tasks
require segmentation of the whole brain from the head, either
because the whole brain is the ROI such as in Alzheimer’s
studies or because automatic ROI extraction using statistical
methods is made easier if the skull and scalp have been
removed[1].
ISSN: 2231-5381
II.MAGNETIC RESONANCE IMAGING (MRI)
Imaging plays a central role in the diagnosis of brain
tumors. Early imaging methods were invasive and sometimes
dangerous-such as pneumo-encephalography and cerebral
angiography have been abandoned in recent times because of
their non-invasiveness. High-resolution techniques especially
Magnetic Resonance Imaging (MRI) and Computed
Tomography (CT) scans are preferred these days. There are
various sequences in MR images like T1, T2, FLAIR and
MRS. Among these, T1, T2 and FLAIR are widely used.
III.REGION OF INTEREST
The idea of an ROI is commonly used in many
application areas. For example, in medical imaging, the
borders of a tumor may be defined on an image or in a
volume, to measure its size. The endocardial border may be
defined on an image, maybe during different phases of the
cardiac cycle, say end-systole and end-diastole, for the need of
assessing cardiac function. In geographical information
systems (GIS), an ROI can be taken actually as a polygonal
selection from a 2D map. In the view of computer imaging
system, the ROI defines the borders of an object under
consideration. In many applications, symbolic (textual) names
are added to a ROI, to describe its content in a compact
manner.
An ROI can contain many numbers of individual points called
Point of Interests (PoI)
An ROI is a form of elucidation, often associated with
categorical or quantitative information (e.g., measurements
like volume or mean intensity), displayed as text or in
structured form.
There are three fundamentally different means of encoding an
ROI:
 As a vital part of the sample data set, with a unique or
masking value that may or may not be outside the
normal range of normally occurring values and which
tags individual data cells
 As separate, completely graphic information, such as
with vector or bitmap (rasterized) drawing elements,
perhaps
with
some
accompanying
plain
(unstructured) text in the format of the data itself
http://www.ijettjournal.org
Page 366
International Journal of Engineering Trends and Technology (IJETT) – Volume 10 Number 7 - Apr 2014

As a separate structured semantic information (such
as coded value types) with a set of spatial and/or
temporal coordinates
‖
‖ = Measuring the similarity or mean distance of the
neighbouring voxels present in the data point
to the centre
vector (centroid value) of cluster j.
Degree of membership is given as,
A. Block diagram to identify tumor region
T1,T2 and FLAIR
weighted brain
input image
Image resized to
256 × 256 pixel
size
∑
Image
preprocessing
(RGB to gray
scale conversion)
(
‖
‖
‖
‖
)
m = Fuzziness co – efficient obtained from the overlapping of
clusters and it is defined as 1.5 ≤ m ≤ 2.5 for optimum
segmentation results [17].
Centre vector (Centroid voxel) for the cluster is defined as,
∑
∑
Segmented
image with
tumor region
identified
ROI Extraction
for segmentation
Image read by
ROI algorithm
FIG.1 BLOCK DIAGRAM TO IDENTIFY TUMOR REGION
Fuzzy C- means (FCM) algorithm is a clustering
technique developed by Dunn, improved by Bezdek and
further improved by Matteo Matteucci; in this technique the
voxels (data) of the Magnetic Resonance (MR) brain images
are classified and grouped under ‘n’ number of clusters
), the updated position
specified by PSO algorithm ⃑⃑⃑ (
vector of the particle (Voxel or pixel) derived with the help of
PSO algorithm acts as the Centroid value and based upon this
value, the number of clusters for FCM algorithm is also
assigned using PSO. Using the position vector, centre voxel
of each cluster is identified and assigned with a high
membership grade value through FCM. The neighboring
voxels of least mean distance from the centroid voxel are
assigned with low membership grade value and are grown
around the centroid value, hierarchically. The membership
grade and the cluster centers are iteratively updated to reduce
the objective function of grouping the voxels.
∑
‖
‖
Output obtained from the lesion segmentation based
on the ROI method and FCM methods are compared with the
gold standard. The similarity criteria (SI),overlap fraction
(OF) and extra fraction (EF), are calculated for the selected
slices. The SI is a criterion for the properly classified lesion
area relative to the total areas of lesion, in both the gold
standard and in the segmented image. The OF and EF
specifies, the areas which have been correctly and falsely
classified as lesion area relative to the lesion area in the gold
standard. The similarity criteria are defined by (Fig. 2):
(
)
[6]
[2]
N = Number of data points given as an input to the algorithm
(Total number of voxels or pixels)
K = Number of iterations to be performed.
) – Number of clusters into which the voxels are
C = ⃑⃑⃑ (
to be grouped. ‘C’ is calculated using the position vector
obtained from PSO algorithm.
Centre vector for the clusters (Centroid value of the
voxels).
= Data points (voxels or pixels).
= Degree of membership for the ith data point of
in
cluster j.
ISSN: 2231-5381
.
(Usually
.
V. EVALUATION CRITERIA
IV. FUZZY C-MEANS (FCM) ALGORITHM
∑
During the initial processing of the algorithm,
= A random value initialized as,
expressed as 0, 0.1, 0.2, 0.4….1), such that ∑
In these equations, TP stands for true positive, FP for false
positive, and FN for false negative. SI and OF for a good
segmentation should be closer to 1 and EF be close to 0. In
real time, a value for SI more than 0.7 represents a very good
segmentation. The mean values of similarity criteria are
categorized to the three patient categories and volumetric
assessment of lesions volume between the fully automated
segmentation and the gold standard are performed using
correlation co-efficient (CC).
http://www.ijettjournal.org
Page 367
International Journal of Engineering Trends and Technology (IJETT) – Volume 10 Number 7 - Apr 2014
The normal value or the desired value for a quality image
should always range from 40 to 60 dB.
C. Jaccard Index:
The Jaccard index, also known as the Jaccard similarity
coefficient (developed by Paul Jaccard), is a statistic used for
comparing the similarity and dissimilarity of sample sets. The
Jaccard coefficient measures similarity between sample sets,
and is defined as the ratio of size of intersection and the size
of the union of the sample sets.
Fig. 2. Comparison of a binary segmentation with the reference image (Ref).
TP, TN, FP, and FN represent the true positive, true negatives, false positives,
and false negatives voxels, respectively [1].
VI.
Mean Square Error MSE (Mean Square Error) refers
to the process of squaring the differentiated quantities. It is
defined as the average of the squares of the errors obtained by
subtracting the input and the output values. MSE (Mean
Square Error) is the cumulative squared error value between
the input image ( ) and the segmented image K( ).
)
∑∑ ( )
( )
Welstead and Stephen T Fractal (ISBN 978-0-8194-3503-3)
m - Number of rows present in an input image represented as a
matrix.
n - Number of columns present in an input image represented
as a matrix.
i = m, to perform incremental operation for ‘rows’ in a loop
statement.
j = n, to perform incremental operation for ‘columns’ in a loop
statement.
MSE value of the resulting image should be
minimized. Reduced MSE values offer minimized error
occurrence in the segmented MR brain images
.
B. Peak signal-to-noise ratio (PSNR)
Peak signal-to-noise ratio, is an engineering term which
ratios between the maximum possible power of a signal and
the power of corrupting noise that affects the reliability of its
representation
PSNR = 20.log(MAXi) – 10.log(MSE)
Where,
MAXi - the maximum pixel value of the image
MSE - Mean Squared Error
ISSN: 2231-5381
[4]
The presence and absence of data is given by,
(
COMPARISON PARAMETERS
A. MSE (Mean Square Error)
(
(
)
D. Dice Overlap Index:
)
Here, A and B indicates the number of species in samples A
and B, respectively, and C represents the samples shared by
both A and B; QS usually in the range of 0 to 1, is the quotient
of similarity. This expression is easily extended to
profusion instead of presence/absence of species. The
Sorensen index resembles to Dice's coefficient which also has
the same range.
D. Memory Storage Required:
Amount of space or memory required for the storage of
output data is called as Memory storage required. If the
number of images is increased in a process obviously there is
a need to increase the memory required. So, if the memory
required for storage is lower, then there can be more space to
work on with the images.
E. Segmentation Accuracy:
Image segmentation is the course of grouping a digital
image into multiple segments which are also known as super
pixels. The need of segmentation is to simplify and/or change
the representation of an image into pixels which can give
higher results. Image segmentation is usually used to locate
objects and boundaries (lines, curves, etc.) in images. As a
whole image segmentation is labeling every similar group
which can share some characteristics
Previously mentioned values like the TP, FP, TN and FN
are the four vital elements that are used to calculate the
segmentation accuracy.
VII.RESULTS AND DISCUSSION
Input image is called upon using the algorithm with
the help of MATLAB program where the region of interest
(tumor region) is to be selected is shown in fig 3. Unlike the
crop function which is present in the MATLAB tool box to
crop out the specified region, here free hand sketch is applied
to concentrate on the required ROI. In the earlier, either a
square or rectangular part can alone be cropped whereas in
this ROI method, it can of free form which is indicated in
http://www.ijettjournal.org
Page 368
International Journal of Engineering Trends and Technology (IJETT) – Volume 10 Number 7 - Apr 2014
fig 4. The ROI region is completely extracted from the
original image which is illustrated in fig 5.
TABLE 1
COMPARISON PARAMETERS
Choose ROI from the figure
http://www.ijettjournal.org
FCM
0.5910
0.5910
0.5910
0.5910
0.5910
0.5910
0.5910
0.4692
0.4702
0.4692
ROI
0.8197
0.8213
0.8245
0.7461
0.7585
0.7330
0.6743
0.6860
The tumor region which is obtained by using FCM algorithm
shows more accurate segmentation of the image because of
the increased values in the number of TP values obtained. The
images thus obtained are listed and the values are compared
with the results of the ROI algorithm and are listed below.
ISSN: 2231-5381
0.6718
FCM
0.4194
0.4194
0.4194
0.4194
0.4194
0.4194
0.4194
0.3065
ROI
0.6944
0.6967
0.7014
0.5950
0.6110
0.5662
0.5086
0.5221
0.8164
3.7864
1.5382
0.3065
5.2893
4.1661
0.3074
54.2736
62.3181
DOI
0.5059
43.3142
45.4752
TC
0.6898
FCM
2.8035
1.9221
3.2367
3.4144
1.1993
1.2100
1.3583
0.6477
0.1029
3.1626
ROI
3.5522
4.4290
4.0411
5.3363
5.7256
5.3605
5.6804
70.2300
3.0315
1.8678
60.5427
T2
axial
5.9172
FCM
59.2936
60.9815
62.2533
ROI
46.2343
47.1381
47.1664
49.276
2
63.519
50.747
5
73.474
8
41.914
3
61.313
6
42.4074
ENTROPY
T2
coronal
44.0061
FCM
0.8451
0.5730
0.4275
0.3194
ROI
1.5458
1.2568
1.2486
0.7682
0.5474
0.0323
0.5308
0.0681
Fig.9 Cluster formation 3
PSNR
0.1529
Fig.8 cluster formation 2
4.1845
T1
coronal
T2
coronal
Fig .7 Cluster formation 1
T1
sagittal
Fig.6 Input Image
T1
coronal
T1
Axial
T1
sagittal
Fig.5 ROI selected is completely extracted from the image
3.7357
T1
Coronal
Fig.4 Selected ROI showing
the masked part
T1
axial
Fig.3 Input image
MSE
2.5851
Name
of
the
Image
Page 369
International Journal of Engineering Trends and Technology (IJETT) – Volume 10 Number 7 - Apr 2014
the objective function which looks for the best possible tumor
segmentation from the given input image. Also, the memory
required for the storage of the image is also minimized.
Hence, on the lookout for the best possible algorithm for
segmentation of tumor part from MR brain images, Fuzzy cmeans (FCM) algorithm outsmarts the Region of Interest
(ROI) algorithm with the minimized values of various
comparison parameters.
REFERENCES
[1]
Fig.10 Output Image showing segmented tumor region using FCM algorithm
TABLE 2- SIMILARITY CRITERION
[2]
Similarity
Criterion
No. of
Images
10
Over Fraction
Extra Fraction
ROI
FCM
ROI
FCM
ROI
FCM
0.8824
0.3636
0.8824
0.3333
0.1176
0.5000
[3]
[4]
[5]
VIII.
CONCLUSION
The comparison parameter values which are
tabulated for both the Region of Interest (ROI) and Fuzzy C
Means (FCM ) algorithms, clearly indicates that the later
stands out best in the values like Mean Square Error (MSE),
the Peak to Signal Noise Ratio and entropy values. The
images obtained are also clearer and time consumption is
comparatively low in that of the FCM algorithm. The number
of true positive (TP) values is also high in number because of
ISSN: 2231-5381
[6]
Juin-Der Lee, Hong-Ren Su, PhilipE. Cheng, Michelle Liou, John
A. D. Aston, Arthur C. Tsai, and Cheng-Yu Chen, “MR Image
Segmentation Using A Power Transformation Approach”, IEEE
Transactions on Medical Imaging, vol. 28, no. 6, June 2009.
Hassan Khotanlou, MahlaghaAfrasiabi, “Segmentation Of
Multiple Sclerosis Lesions In Brain Mr Images Using Spatially
Constrained Possibilistic Fuzzy C-MeansClassification”, Journal
of Medical Signals & Sensors, Vol 1, Issue3, Sep-Dec 2011.
S.Satheesh, Dr.K.V.S.V.R Prasad, Dr.K.Jitender Reddy, “Tumor
Extraction And Volume Estimation For T1-Weighted Magnetic
Resonance Brain Images”, Global Journal of Computer Science
and Technology Neural & Artificial Intelligence, Volume 12 Issue
12 Version 1.0 Year 2012.
AndacHamamci, Nadir Kucuk, KutlayKaraman, KayihanEngin,
and GozdeUnal, “Tumor-Cut: Segmentation Of Brain Tumors On
Contrast Enhanced Mr Images For Radiosurgery Applications”,
IEEE Transactions on Medical imaging, vol. 31, no. 3, March
2012.
Ortiza, J.M. Górrizb, , J. Ramírezb, D. Salas-Gonzálezb, J.M.,
“Two Fully-Unsupervised Methods For Mr Brain Image
Segmentation Using Som-Based Strategies”, Journal of Applied
Soft Computing 13 (2013) 2668–2682.
RasoulKhayatia, MansurVafadusta, FarzadTowhidkhaha, S.
MassoodNabavib, “Fully Automatic Segmentation Of Multiple
Sclerosis Lesionsin Brain Mr Flair Images Using Adaptive
Mixtures Method And Markov Random Field Model”, Computers
in Biology and Medicine 38 (2008) ,pp.379–390.
http://www.ijettjournal.org
Page 370