Brain Tumor Detection using Clustering Algorithms in MRI Images

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 06 | June -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 1587
Brain Tumor Detection Using Clustering Algorithms in MRI Images
Nikhita Biradar1, Prakash H. Unki2
1M.Tech Student, Dept. of Computer Science and Engineering,
BLDEA’s V.P. Dr. P.G. Halakatti College of Engineering and Technology, Vijayapur, Karnataka, India
2Associate Professor, Dept. of Computer Science and Engineering,
BLDEA’s V.P. Dr. P.G. Halakatti College of Engineering and Technology, Vijayapur, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Brain tumor is an accumulation of abnormal
cells in the brain. Detection of tumorfrommagneticresonance
imaging (MRI) brain scan is one of the most promising
research topics in medical image processing. This paper
presents a novel tumor detection system in MRI images using
k-means technique integrated with Fuzzy c-means (FCM)
clustering algorithm and artificial neural network (ANN).
ANN is used to classify the MRI images into two categories;
normal and tumor image. The proposed system takes benefit
of both integrated algorithms in the aspect of minimal
computation time and accuracy. It accurately extracts the
tumor region and calculates the tumor area. The accuracy is
calculated by comparing results with the ground truth (GT)of
processed image.
Key Words: Magnetic resonance imaging (MRI),k-means
clustering, fuzzy c-means (FCM) clustering, artificial
neural network (ANN), ground truth (GT).
1. INTRODUCTION
Brain tumors are formed by collection of abnormal cellsthat
grows uncontrollable. Diagnosis of brain tumors is done by
detection of the abnormal brain structure. The internal
structure of brain can be viewed by magnetic resonance
imaging (MRI) and computed tomography (CT) scans.
Compared to CT scan, MRI scan is more efficient and it
doesn’t affect the patient body asnoradiationsareused. MRI
scanning is done by using radiofrequency andmagneticfield
[1]. MRI images are analyzed by the radiologists to diagnose
the tumor. Segmentation of images is important as large
numbers of images are generated during the scan and it is
unlikely for clinical experts to manually divide these images
in a reasonable time.
Image segmentation refers to segregation of given image
into multiple non-overlapping regions. Segmentation
represents the image into sets of pixels that are more
significant and easier for analysis. It is applied to
approximately locate the boundaries or objects in an image
and the resulting segments collectively cover the complete
image [2]. The segmentation algorithms works on one of the
two basic characteristics of image intensity; similarity and
discontinuity [3]. In the former, segmentation technique is
based on dividing an image into set of pixels that are similar
to the somepredefined criteria. Thelatterpartitioningworks
on the changes in intensity of an image, such as corners and
edges. Segmentation has a significant part in clinical
diagnosis and can be useful in pre-surgical planning and
computer assisted surgery. Therefore, numerous
segmentation techniques are available which can be used
widely, such as threshold based segmentation, histogram
based methods, region-based (region growing, splitting and
merging methods), edge-based and clustering methods
(expectation maximization, k-means, FCM and mean shift)
[4]-[6]. Clustering methods aremostpromisingtechniquefor
processing the medicalimages.Clusteranalysiscanbesetout
as a pre-processing stage for other methods, namely
classifiers that would then run on selected clusters [7].
Therefore in our system, we have used clustering
segmentation techniques for diagnosis of tumor and
calculating tumor area in MRI images.
This paper presents an effective tumor detection system
for MRI images by integrating k-means with FCM clustering
techniques. This system gets benefit of the k-means in the
aspect of minimal computation timeandfuzzyc-meansinthe
aspect of accuracy.k-meansalgorithmistoperformtheinitial
segmentation. Then, on the criteria of updated membership
set and exact cluster selection, an approximate segmented
tumor is located from FCM technique. Even the minute
changes in intensities of normal and tumor tissue is
recognized by this method. ANN classifies MRI into two
categories, normal image andimagewithtumorbypreparing
pertinent training, target data. Finally, the reliability of the
system is calculated by comparing the result with GT of the
processed image. Essential utilization of this technique is to
get measure and location of tumor, which will help in
organizing of treatment and surgery.
2. RELATED WORK
Many researchers havesuggestedseveralmethodologiesand
procedures for medical image segmentation and techniques
for tumor detection. These include region growing,
thresholding, mean shift methods, clustering and statistical
model.
H. Suzuki and J. Torwaki [8], developed an algorithm for
automatic segmentation of head MRI images using
thresholding techniques. The algorithm consists of three
incremental steps; histogramanalysis to locatethe brainand
next step is to create a mask using nonlinear anisotropic

diffusion and thresholding to segregate brain from detected
head region.Finallyactive contour model is applied todetect
intracranial boundary.
Bandhyopadhyay sk, paul [9], proposed a segmentation
technique based on k-mean clustering with dual localization
that effectively segment tumor from brain MR images. After
optimal segmentation, histogram and line scan method are
applied to estimate breadth, length of tumor along xy-axis.
Shen, William sansharm [10], designed a fuzzy brain
tissue segmentation method by applying new extension to
FCM. In segmentation,theyconsideredtwoinfluentialfactors
to address the issues in neighborhood attraction. Finally,
proposed techniques were tested for simulated, square and
hospital collected MR images, at different noise levels.
Lemieux, G. Hahemann et al. [11], presentedanautomatic
segmentationtechniqueforbraininT1-weightedMRimaging
data by applying thresholdingandmorphologicaloperations.
Their segmentationtechniqueisindependentofimagingscan
orientation. They evaluated the performance by comparing
results with semi-automated measurements.
Noreen, Hayatand S.Madami[12],introducedatechnique
to segment MRI images using discrete wavelet transform
(DWT) andFCM. They extracted high pass image byapplying
DWT, which is robust to noise and FCM further enhanceedge
details in an image. This combination carries benefit of both
the algorithms and gives better segmented result.
3. CLUSTERING TECHNIQUES FOR IMAGE
SEGMENTATION & CLASSIFIER
The clustering segmentation techniques used here are k-
means and fuzzy c-means (FCM). In k-means a pixel point
belongs to only one cluster, where asinc-meanswithcertain
probability it may belong to more than cluster.Classification
is an important technique used widely to differentiate
normal and tumor brain images.
3.1 k-means clustering
k-means clustering aims to divide the set of pixels X= {xi
|i=1, 2 . . ., N} into k clusters [13]. It supports
multidimensional vectors and has highest computational
efficiency. Initially the no. of cluster values are defined as k.
Then k cluster centers C= {cj | j=1, 2 . . ., k } are chosen
randomly. Euclidiandistanceiscalculatedbetweeneachpixel
point to each cluster centers using Eq.1.
(1)
Next, the pixel point is assigned to the cluster which has
minimal distance among all. Then the newclustercentersare
re-calculated. Again pixel distances are compared and re-
assigned to new clusters. The process continues until no
points were reassigned. The output of this technique is
cluster image where each pixel belongs to one of the closest
cluster. The main merits of this technique are its simplicity
and minimal computational time.
3.2 Fuzzy c-means (FCM) clustering
FCM is introduced to divide the set of pixels X= {x1, x2, . . .,
xN} into C fuzzy clusters where each point has a degree of
belonging to clusters. It allows a point to belong to more
than one cluster as per its membership value. It is an
iterative process for minimizing objective function, related
to fuzzy membership set U of cluster centers C:
(2)
Where, uij is the membership table, m is a cluster fuzziness
factor and (xi – cj) is euclidean distance.
The data points nearer to center of cluster have highest
degree of membership than the points on edge [14]. FCM
initially guess the cluster centers and assigns every point a
membership grade for each clusters. Then, it moves the
cluster center to right location by iteratively updating the
centers within a data set. The membership defines the
fuzziness of an image and alsodefinesinformationcontained
in an image.
3.3 Artificial Neural Network (ANN) classifier
ANN is a computational model inspired by functional
and/or structural aspects of the biological neural networks.
The artificial neurons are organized into 3-layers; input,
hidden and output layer. The algorithm uses multi-layer
interconnected units of artificial neurons (processing
elements), to solve the specific problem. Classification
involves feature extraction which gives important
characteristics of an image. The classification process is split
into training/learning phase and the testing stage. In the
training, features are extracted from the diagnosed(known)
images and are stored in knowledgebase.Thenintesting,the
features of unknown image are compared with the stored
data for classification.
4. PROPOSED SYSTEM
The proposed tumor detection system gets benefit from last
two algorithms. The main motivation for combining these
algorithms is to reducethetotaliterationsdonebyinitializing
exact cluster to the FCM clustering. Fig-1describestheentire
flow for detecting tumor in brain MRI images.

Fig -1: Block diagram for proposed tumor detection
system
In this block diagram, initially we perform
acquisition of brainMRIimages.Imagesarepreprocessedfor
removing noise and skull part using erosion and dilation
morphological techniques. The output of this step is noise-
free MRI containing onlythe brain part.Theprocessedimage
is given as input for segmentation step where, the
combination of k-means and fuzzy c-means clustering
algorithms are used. In feature extraction stage, we have
extracted different features from sharpened image like
entropy, energy, contrast, intensity, homogeneity, etc.These
features are used as an input to classifier for tumor
detection. The neural network classifier is appliedtoclassify
the brain MRI images. If the brain image has the tumor
region, then the image is further processed to detect the
tumor accurately. Next, the tumor region is marked using
morphological operations and area of the tumor region is
calculated. Finally in validation stage, segmented image by
integrated k-means and FCM are compared to the ground
truth (GT) of the processed image. The results are assessed
by performancematrixthatcontainstheaccuracy,sensitivity
and specificity.
5. RESULTS AND DISCUSSION
In the proposed system, detection of tumor in MRI images is
done using combined k-means and FCM clustering
techniques. When the inputimageisloaded,preprocessingis
done for the removal of noise and skull part using the
morphological operations. The resulted image of the
preprocessing step is as shown in Fig -2.
(a)
(b)
Fig -2: (a) Original input images (b) preprocessed images
After preprocessing step,theimageisgivenasinputtothe
image segmentation process. The integrated k-means and
FCM clustering technique is applied to get the segmented
image.k-meanclusteringdividesthepreprocessedimageinto
the clusters having same or nearby intensity values.
Therefore it reduces the total iterationsbyinitializingproper
cluster to the FCM clustering and the region of interest is
detected and segmentedtoseparateimageasshowninFig-3.
(a)
Acquisition
(Brain MRI Image)
Image Preprocessing
Initial Segmentation using
k-means clustering
Tumor area calculation
Marking tumor position
Validation using
Ground truth
Segmentation using FCM
clustering
Feature Extraction
ANN classifier
Image with Tumor Normal Image

(b)
Fig -3: (a) Output of combined k-mean and FCM
technique (b) Final segmented result
Image classification has been done by extractingdifferent
features like energy, entropy, homogeneity etc. The MRI
image is classified into two categories, normal and abnormal
brain image. If the MRI image has the tumor then the tumor
region is marked using morphological operationandalsothe
area of tumor is calculated. The below table -1 depicts the
extracted feature and calculated tumor area for sample
images.
Table -1: Feature Extraction and calculated tumor area
Features
Sample
Image 1
Sample
Image 2
Energy 3.43e-03 1.02e-03
Contrast 1.92e+03 2.53e+02
Entropy 5.68e+00 6.89e+00
Homogeneity 2.30e-02 1.32e-01
Dissimilarity 4.33e+01 1.33e+01
Calculated Tumor
Area (pixel)
296 992
The tumor region is marked using morphological
operations. The output image of the detection process is as
shown in Fig -4.
(a)
(b)
Fig -4: (a) Input image with tumor (b) Detected brain
tumor images
In validation stage, experimental resultsofthesystemare
compared with GT of the processed image. This comparison
is assessed by performance matrix that contains accuracy,
sensitivity and specificity. The performance matrix for two
sample input images is listed in table -2.
The accuracy, sensitivity and specificity are calculated using
following equations:
(3)
(4)
(5)
where, True Positive (TP)= intersection between resulting
image and GT, False Positive (FP)= output segmented image
not overlapping the GT, False Negative (FN)= missed part of
GT and True Negative (TN)= part of image beyond the union
output image + GT.
Table -2: Performance matrix (Validation of results by
comparing it with GT of the processed image)
Sample
Input
Image
Output
Image
Accuracy 0.99866 0.98474
Specificity 0.97872 0.92209
Sensitivity 0.99895 0.98821
6. CONCLUSION
In this work, we have presented a new methodology for
detection of tumor in MRI imagesbycombiningk-meansand
FCM techniques. It is applied to remove the constraints of
the k-mean and FCM clusteringalgorithms.Theclassification
of tumor in MRI image is done usingartificial neural network
(ANN) based on similarity between the featurevectors.ANN
classifies the brain MRI images into two categories, normal
and tumor images. In this approach, k-mean algorithm is
used to perform the initial segmentation of MRI images. On
the criteria of updated membership set and appropriate
cluster selection, a final segmented tumor image is obtained
using FCM. The segmented tumor region is marked and the

tumor area is calculated. The effectiveness of the current
method is determined by comparing results with GT of the
processed image. The experimental results clarify that the
suggested approach gives better segmentation result and
accuracy in minimal execution time.
REFERENCES
[1] Patel J, Doshi K, “A study of segmentation methods for
detection of tumor in brain MRI”, Advance in Electronic
and Electric Engineering, Vol. 4(3), 2014, pp. 279-284.
[2] Janani V, Meena P, “Image segmentation for tumor
detection using fuzzy inference system”, International
Journal of Computer Science andMobileComputing,Vol.
2(5), 2013, pp. 244-248.
[3] Acharya J, Gadhiya S and Raviya, “Segmentation
techniques for image analysis: a review”, International
Journal of ComputerScience andManagementResearch,
Vol. 2(4), 2013, pp. 1218-1221.
[4] Naik D, Shah P, “A review on image segmentation
clustering algorithms”, International Journal of
Computer science and Information Technologies, Vol.
5(3), 2014, pp. 3289-3293.
[5] Christ SA, Malathy K, Kandaswamy A, “Improved hybrid
segmentation of brain MRI tissue and tumor using
statistical feature”, ICTACT Journal on Image and video
processing, Vol. 1(1), 2010, pp. 34-49.
[6] Seerha GK, Kaur R, “Review on recent image
segmentation techniques”, International Journal of
Computer Science Engineering, Vol. 5(2), 2013,pp.109-
112.
[7] Panda M, Patra MR, “Some clustering algorithms to
enhance the performance of the network intrusion
detection system”, Journal of Theoretical and Applied
Information Technology, Vol. 4(8), 2008, pp.795-801.
[8] H. Suzuki, J. Toriwaki, “Automated segmentationofhead
MRI images by knowledge guided thresholding”,
Computed Medical Image Graphics, Vol. 15(4),1991,pp.
233-240.
[9] Bandhyapadhaya SK, Paul TU, “Automatic segmentation
of brain tumor from multiple images of brain MRI”,
International Journal of Application or Innovation in
Engineering and Management, Vol. 2(1),2013,240-248.
[10] S. Shen, W. Sandham, et al., “MRI fuzzy segmentation of
brain tissue using neighborhood attraction with neural
network optimization”, IEEE Transactions and
Information Technology in Biomedicine,2013,Vol.9(3),
pp. 459-467.
[11] L. Lemieux, G. Hagemann, et al., “Fast accurate, and
reproducible automatic segmentationofthe braininT1-
weighted volume MRI data”, Magnetic Resonance
Medical, 1991, Vol.42, pp. 127-135.
[12] N. Noreen, K. Hayat et al., “MRI Segmentation through
wavelets and Fuzzy C-Means”, World Applied Science
Journal 13(Special Issue of Applied Math), 2011, pp. 34-
39.
[13] W. Narbuakaew, H. Nagahashi, et al., “Integration of
Modified K-Means Clustering and Morphological
Operations for Multi-Organ Segmentation in CT Liver-
Images”, Recent Advances in Biomedical AND Chemical
Engineering and Materials Science, 2014, pp. 34-39.
[14] Lei Jiang, Wenhui Yang, “Fuzzy C Means algorithm for
segmentationofmagneticresonanceimages”,Proc.VIIth
Digital Image Computing: Techniques and Applications,
(Eds.), 2013, pp.10-12.

Brain Tumor Detection using Clustering Algorithms in MRI Images

More Related Content

What's hot (20)

Similar to Brain Tumor Detection using Clustering Algorithms in MRI Images (20)

More from IRJET Journal (20)

Recently uploaded (20)

Brain Tumor Detection using Clustering Algorithms in MRI Images