BRAIN TUMOR ANALYSIS USING K-MEANS CLUSTERING

AND EDGE DETECTION:

T. Usha Sree, J. Govinda Lahari, Christopher Clement. J

ABSTRACT:
Precise and detailed anatomical information can be taken from the segmentation of magnetic
resonance response (MR) images, which helps for accurate diagnosis of the brain. For this,
medical image segmentation plays a crucial role. Because of the complex structure of the brain
and the equipment noise, the uncontrolled production of cells causes abnormal growth resulting
in brain tumors. The most common device for diagnosis is a magnetic resonance image (MRI).
When it comes to the field of a medical imaging system, this MRI stood as an emergent research
area. In the diagnosis of a tumor, the accurate detection and depiction of the tumour's size and
brain's location play a vital role. For the extraction of suspicious regions from the medical
images, this image segmentation plays a significant role in image processing. This work
documents the research on an efficient algorithm for tumour detection based on brain magnetic
resonance images (MRI) through K-means clustering.

------------------------------------------------------------------------------------------------
1. INTRODUCTION:
Hence image segmentation is that the Anatomy of the brain is often viewed by the
fundamental problem utilized in tumor MRI scan or CT scan [2] for diagnosis.
detection. Image segmentation is often Differing types of algorithms were
defined because the partition or developed for brain tumor segmentation.
segmentation of a digital image into similar But they'll have some drawbacks in
regions with the main aim to simplify the detection and extraction and result in people
image into account into something more that have brain tumours died. K-mean
meaningful and easier to research visually. clustering algorithm is employed for the
Image segmentation methods [1] are often accurate detection of the brain tumour. The
classified as thresholding, region-based, tumour could also be primary or secondary.
supervised, and unsupervised techniques. A The brain tumour is [3] primary when it's an
tumour is often defined as a mass that grows origin and therefore the brain tumour is
with none control of normal forces. Real- secondary when the part of the tumour is
time diagnosis of tumours by using more spread to a different place and grown.
reliable algorithms has been lively of the
Examples: contrast enhancement, edge
newest developments in medical imaging
detection, sharpening, zooming, etc. Edge
and detection of a brain tumour in MR and
detection may be a problem of fundamental
CT scan images.
importance in image analysis. In typical
For the detection of the range and shape of images, edges characterize object
the tumour in the brain MRI image, this boundaries and are therefore useful for
segmentation deals with the implementation segmentation [4] registration and
of a straightforward algorithm. Normally the identification of objects during a scene.
Localization, detection, and extraction of
objects or details of a given type on the 2. METHODOLOGY:
complex image are long recognized as a
number of the foremost challenging Methods that we use to detect brain
problems within the field of image tumors from MRI images are watershed
processing. segmentation and contour of the image.
Before we start the segmentation, we've to
Brain tumors affect humans badly, due to filter the MRI image noise. We use during
the abnormal growth of cells within the this step the digital filter from Matlab
brain. It can disrupt proper brain function ("Sobel edge masks") which can show the
and be life-threatening. Two sorts of brain gradient is high at the borders of the MRI
tumors are identified as benign tumors and objects and low (mostly) inside the image.
malignant tumors[5]. A brain tumor may be
a fatal disease that can't be confidently 2.1. SOME IDEAS FOR MRI IMAGE
detected without MRI. Within the project, PROCESSING: -
it's tried to detect whether a patient's brain
has a tumor or not from an MRI image using Detecting brain tumor and automatic brain
tissue classification from resonance
MATLAB simulation. To pave the way for
images (MRI) is extremely important for
morphological operation on MRI image[6],
research and clinical studies of the
the image was first filtered using an traditional and diseased human brain. The
Anisotropic Diffusion Filter to scale back foremost important method to process an
contrast between consecutive pixels. Then MRI image is a segmentation of the image
the image was resized and utilizing a which works by dividing the objects
threshold value image was converted to a within the image and processing them
black and white image manually. This separately. There are three important
primary [7] filter the plausible location for segmentation methods used with MRI
tumor presence. The larger goal of the imaging and they are:
project is to create a knowledge base of 2D 1. Classification based
image data of tumors from the MRI images 2. Region-based
taken from a different angle of a specific 3. Contour based
human and by analysing them to means the
2.2. K-MEANS CLUSTERING
precise 3D location of the tumor. to satisfy
this, 2Dtumor detection and segmentation Clustering is one of the main data analysis
are developed to raised accuracy so that 3D methods widely utilized in many practical
detection is often more reliable. this is often applications of emerging areas. Clustering
the first target of the project. Biomedical is that the process of finding groups of
images[8] are different because of the areas objects such the objects during a group is
of the physical body. for instance, to review going to be almost like each other and
soft tissue within the physical body, we've to different from the objects in other groups.
use the MRI scan for soft tissue images like Top quality clusters with high intra-
Brain liver, and other soft tissue within the cluster similarity and low inter-cluster
physical body. However, those[9] curious similarity this is often an honest clustering
about studying hard tissuelike bone or method. Within the k-means algorithm
initially, we've to define the amount of
cartilage should use an X-ray for a tough
clusters k. Then k cluster centre is chosen
tissue image instead of the MRI. The randomly. The space between each pixel
difference within the biomedical image isn't to every cluster's centres are calculated.
just within the area but also different within Space could also be of straightforward
the manner of processing. Euclidean function. A single pixel is
compared to all or any cluster centres corresponding to edge chains. The picture
using the space formula. obtained utilizes one of the basic image
attributes: coherency of the contour
Algorithm: - elements.
1. Give the no of cluster value asks. FLOW CHART:-
2. Randomly choose the k cluster centres
3. Calculate mean or centre of the cluster
4. Calculate the distance b/w each pixel to MRI Image Reading
each cluster centre
5. If the distance is near to the centre then
move to that cluster.
6. Otherwise move to the next cluster.
7. Re-estimate the centre. Image resizing
8. Repeat the process until the centre
doesn't move

2.3. EDGE DETECTION:-

Noise removal
Edge detection may be a problem of
fundamental importance in image
analysis. These typical images can be used
for segmentation, registration, and Converting into matrix
identification of objects during the use
cause of its edges characterizes object
boundaries. Localization, detection, and
extraction of objects or details of a given
type on the complex image are long RGB to GREY Conversion
recognized as a number of the foremost
challenging problems within the field of
image processing. Edge-based
segmentation is one of the earliest
segmentation approaches and remains Adjusting the image and
vital. It depends on the edges found in an converting into kirsch edges
image and these edges mark's the
locations of discontinuities in grey-level,
color, texture. The target is to group local
edges into a contour picture. Existing Converting into Sobel edges
objects or image parts are present

Converting into k-cluster for

better analysing.

This flow chart explains the procedure of

analysing the brain tumor using both k-
means clustering and edge detection
methods.
3. RESULTS:

We have taken few MRI images of brain tumor and analysed using MATLAB.

Fig.1 Original Image

Fig.2 RGB color of brain tumor

Fig3. K-means clustering output

Analysing the output of Matlab we found that k –means clustering converts

the original image into [fig.2] RGB color format and then it converts the RGB
output to black and white [fig.3] which results in greater analysis of tumor.
 Fig4.Edge detection output

From edge detection, we can say that [fig.4] Original image is then converted into Sobel
Horizontal where it detects all the horizontal Edges and then it detects all Sobel vertical by
calculating the gradient intensity at each and every pixel .Then canny edge process finds the
maxima and detects it as edge.

Fig5. K-means and edge detection output

Finally with the help of Matlab we analyzed that by combining both the methods the analyzing
performance is better .so we found that tumor is clear and can be analyzed easily using both k-
Means and edge detection and we clustered it up to 3 iterations on further increment of clusters
can increase greater vision of tumor
4. CONCLUSION: - [4] A. Bal and R. Saha, “An improved
method for
Abnormal growth of tissues within the brain handwritten documentanalysis using
which affect proper brain functions is taken segmentation, baseline recognition and
into account as a brain tumor. the most goal writing pressuredetection,” Procedia
of medical image processing is to spot Computer Science, vol. 93, pp. 403–415,
accurate and meaningful information using 2016.
images with minimum error possible. Brain [5] D. Kwon, R. T. Shinohara, H. Akbari,
tumor identifications through MRI images is and C. Davatzikos,
difficult task due to the complexity of the “Combininggenerative models for
brain. These tumors are often segmented multifocal glioma segmentation
using various image segmentation and registration,” in International
techniques.There are differing types of Conference on Medical Image
tumors are available. they'll be as mass in Computing andComputer-Assisted
brain or malignant over the brain. Suppose Intervention. Springer, 2014, pp. 763–770.
if it's a mass then K- means algorithm is [6] A. Bal and R. Saha, “An improved
enough to extract it from the brain cells. If method for text
there's any noise are present within the MR segmentation andskew normalization of
image it's removed before the K-means handwriting image,” pp. 181–196, 2018.
process. The noise free image is given as a [7] M. Havaei, A. Davy, D. Warde-Farley,
input to the k-means and tumor is extracted A. Biard, A. Courville, Y. Bengio,C. Pal, P.-
from the MRI image. For accurate M. Jodoin, and H. Larochelle, “Brain tumor
segmentation of brain tumor K-means segmentation with deep neural networks,”
clustering algorithm is more efficient. Medical image analysis, vol. 35, pp. 18–31,
2017.
[8] A. Bal and R. Saha, “An efficient method
5. REFERENCES: for skew
[1] A. Hamachi, N. Kucuk, K. Karaman, K. normalization ofhandwriting image,” in
Engin, and G. Unal, “Tumor-cut: Progress in 6th International Conference
segmentation of brain tumors on contrast onCommunication Systems and Network
enhanced mr imagesfor radiosurgery Technologies (CSNT-2016), 2016.
applications,” IEEE transactions on [9] S. Taheri, S. H. Ong, and V. Chong,
medical imaging,vol. 31, no. 3, pp. 790– “Level-set segmentation
804, 2012 of braintumors using a threshold-based
[2] M. Havaei, H. Larochelle, P. Poulin, speed function,” Image
and P.-M. Jodoin, “Within- and VisionComputing, vol. 28, no. 1, pp.
brain classification for brain tumor 26–37, 2010.
segmentation,” International [10] K. Shanthi and M. S. Kumar, “Skull
journal ofcomputer assisted radiology and stripping and automatic seg-mentation of
surgery, vol. 11, no. 5, pp. 777–788,2016. brain mri using seed growth and
[3] N. J. Tustison, K. Shrinidhi, threshold techniques,”in Intelligent and
M. Wintermark, C. R. Durst, B. Advanced Systems, 2007. ICIAS
M. Kandel,J. C. Gee, M. C. Grossman, and 2007. InternationalConference on. IEEE,
B. B. Avants, 2007, pp. 422–426.
“Optimal symmetricmultimodal templates [11] L. M. Fletcher-Heath, L. O. Hall, D.
and concatenated random forests B. Goldgof, and F.
for supervisedbrain tumor segmentation R. Murtagh,“Automatic segmentation of
(simplified) non-enhancing brain tumors
with antsr,” Neuro informatics,vol. 13, no. in magneticresonance images,” Artificial
2, pp. 209–225, 2015.
intelligence in medicine, vol. 21, wavelets as input to support vector ma-chine
no. 1,pp. 43–63, 2001. and neural network,” Biomedical Signal
[12] S. Chaplot, L. Patnaik, and N. Processing and Control,vol. 1, no. 1, pp. 86–
Jagannathan,2“Classification 92, 2006.
of magneticresonance brain images using