You are on page 1of 4

A Survey on Brain Tumor Detection Using Image

Processing Techniques

Luxit Kapoor Sanjeev Thakur, Professor


Amity School of Engineering and Technology Amity School of Engineering and Technology
Amity University, Noida Amity University, Noida
luxit.kapoor@student.amity.edu sthakur3@amity.edu

Abstract: Biomedical Image Processing is a growing and However, as MRI uses magnetic waves, so it is unsuitable
demanding field. It comprises of many different types of for patients with pacemakers and metal implants. Now once
imaging methods likes CT scans, X-Ray and MRI. These we have the scanned image of the brain, it is important to
techniques allow us to identify even the smallest accurately detect the tumor, its size, and its location. All this
abnormalities in the human body. The primary goal of information is necessary for the Neurosurgeon to complete
medical imaging is to extract meaningful and accurate his diagnosis. This is where Computerized Image Processing
information from these images with the least error comes to help. With the use of different segmentation
possible. Out of the various types of medical imaging techniques and feature extraction method, we can accurately
processes available to us, MRI is the most reliable and detect the tumor.
safe. It does not involve exposing the body to any sorts of
harmful radiation. This MRI can then be processed, and II. LITERATURE SURVEY
the tumor can be segmented. Tumor Segmentation Processing an image is a complicated task. Before any
includes the use of several different techniques. The image can be processed, it is important to remove any
whole process of detecting brain tumor from an MRI can unwanted artifacts it may hold. Only then can the image be
be classified into four different categories: Pre- processed successfully. Processing a medical image involves
Processing, Segmentation, Optimization and Feature two main steps. The first is the pre-processing of the image.
Extraction. This survey involves reviewing the research This involves performing operations like noise reduction and
by other professionals and compiling it into one paper. filtering so that the image is suitable for the next step. The
second step is to perform segmentation and morphological
operations. These determine the size and the location of the
Keyword: Image Processing, Segmentation, Filtering tumor
Techniques, Tumor Detection A.Image Pre-Processing: Before we begin processing our
image it is important that the image doesn't consist of any
I. INTRODUCTION unwanted data and is in the right format for processing so
Intracranial Neoplasm or Brain tumor is abnormal growth of that the results are accurate. This preparation is classified as
cells in the brain. Brain is the most complicated part of our pre-processing. Pre-processing involves processes like
body. The symptoms of a tumor may be frequent headaches conversion to greyscale, noise reduction and noise removal,
and migraines. Over the years it may even lead to vision image reconstruction, image enhancement and regarding
loss. At this moment science is scarce about the origins and medical images it may involve steps like skull removal from
factors leading to this abnormal growth. Tumors are an MRI.
classified on two bases: whether they are cancerous or not One of the most common pre-processing practice is the
and their place of origin. The noncancerous form of the conversion to a greyscale image. It is often considered that a
tumor is referred to as Benign. These are easily grey scale image is just a black and white image however
distinguishable and have a slow growth rate. Cancerous this is not true. A black and white image has only 2 shades
tumors are called Malignant. These are very aggressive and i.e black and white so at a point, the intensity can either be 1
can be life threatening as these are hard to detect. or 0. However, a greyscale image consists of shades of grey
When it comes to detecting a tumor, doctors can opt for with no apparent color. This means that every pixel
either an X-ray or an MRI. MRI’s are appropriate when all represents the intensity value at that pixel without showing
other test fail to provide sufficient information. An MRI scan any color. Also unlike a black and white image, a greyscale
uses the properties of magnetism and radio waves to produce image has many different shades with white being the
accurate images. Neurosurgeons most commonly prescribe lightest shade and black being the darkest. Also in a
MRI’s as it provides them with sufficient information to greyscale image the intensity values at a pixel are not
detect even the smallest abnormalities. absolute and can be in fractions. Grey scaling is important as
it provides a more accurate color information which aids

978-1-5090-3519-9/17/$31.00 2017
c IEEE 582
during segmentation. This is the primary step taken by all of The paper by Sneha Khare[9] uses Segmentation based on
the researchers. Genetic Algorithm. These are algorithms that are based on
Once the image is converted to a greyscale image, it is then natural selection and evolution. These belong to a subclass
filtered to remove excess noise. Filters are of two types, one of an evolutionary algorithm, as it uses evolutionary
that allows the low-end frequencies to pass or filters that techniques and natural selection to solve optimization
allow the high-end frequencies to pass. A filter can either problems. It works on a heuristic and an iterative model.[9]
flatten the image or sharpen the image. When a filter is used Thresholding is also a very popular and straightforward
to flatten the image, the noise is blurred leaving behind a segmentation technique in use. It creates a binary segmented
smooth image however the finer details of the image are image from a greyscale image. All it does is replace the
lost. If the image is to be sharpened than the filter enhances pixel with a black pixel at a certain point if the intensity at
the finer details, however this leads to an increased amount that point is less than a certain intensity or replace it a with a
of noise in the image. This noise should be clipped before white pixel if the intensity is more than that.
further processing as it can interfere with the accuracy of the An extension of this is Otsu Thresholding. In this, we
detection program. classify the image as having 2 classes, the background and
A. Filtering: Median Filter is one of the most common noise in the foreground. We try to determine class in which the
removal techniques in use today. The reason behind its pixel is present. To do this we iterate through the various
widespread usage is that it preserves the edges of the image. values on either side until the inter-class variance is
As the name suggests, in this each entry is replaced with the minimum. This technique was used by Rajesh C. Patil[3]
median of its adjacent entries. This filter is very effective for and Deepthi Murthy[6].
removing salt and pepper noise and poisson’s noise. This Various Clustering Techniques are often used to segment
filter works by sweeping the entire signal in a pattern. The images. A cluster is a set of pixels that share some similar
intensity of the median of the pixels in the pattern becomes characteristics. It involves classifying objects on the basis of
the output intensity. Dr. M. Karnan[1], A.Lakshmi[2] and their similarity to each other. Clustering techniques can be
Dr. A. S. Bhalchandra[3] all opted for using a median filter further divided into 2 types. Hard type clustering techniques
to remove noise in their experiment. state that an object can only belong to one cluster only. This
Ming-Ni Wu[4] proposed a Pseudo Color Translation results in very definitive segmentation. However, if the
method to be applied to the greyscale image. A pseudo color image has low resolution and contrast then this technique
image or a false color image differs from a regular image in becomes very hard. An example of Hard clustering
the fact that here the detection of colored objects is different algorithm is the K-Means Clustering Algorithm. The aim of
than from the actual image. A greyscale image is converted the K-Means algorithm is to segment the image into n
to a pseudo color image by assigning a color value to each partitions with each object belonging to the cluster with the
intensity value based on some function. Examples of pseudo nearest mean. Here each point can belong to only one
color images are thermal imaging. It should be used when cluster. This technique was used by A.Lakshmi[2] and in the
only a single channel of data is available. research of Ming-Ni Wu[4]. Another type of clustering
In their research, R. B. Dubey[5] used a Gaussian Filter to technique is the soft clustering technique. An example of
remove noise from the input MRI image. Applying a this is the Fuzzy C-Means Algorithm. Here, unlike K means
Gaussian filter is similar to Weierstrass Transform which algorithm, we assume that each object has the probability of
involves convolving using a Gaussian Function. The belonging to each cluster rather then just a single cluster. It
application of a gaussian filter results in a smooth image. further assumes that an object may belong partially to more
This is similar to viewing the image through a translucent then one cluster. This method is very commonly used in the
screen. As noise is usually present in the high -frequency field of image segmentation. Dr. M. Karnan[1],
regions of an image, so a gaussian filter is a low pass filter. A.Lakshmi[2] both use this in their research.
Research by Deepthi Murthy[6] involved the use of Sobel R. B. Dubey[5] proposed the use of a Level Set
Filter. ”The Sobel–Feldman operator is based on convolving Segmentation method. It works on the principle of partial
the image with a small, separable, and integer-valued filter differential equations. It works by continuously calculating
in the horizontal and vertical directions and is therefore the differences between pixels. Now the image can be
relatively inexpensive in terms of computations”[7]. It is a segmented easily by using well defined mathematical
type of derivative mask which works by calculating the techniques and methods.
difference in pixel intensitites. The Sobel filter is used along Another common type of segmentation technique is the
with edge detection algorithms where the output image Watershed Segmentation technique. It is a region based
consists of well-defined edges. segmentation technique meaning that it searches of pixel and
B. Segmentation: The process of splitting an image into region similarities. So for every pixel, our goal is to
multiple parts is known as segmentation. It creates various determine what region the pixel belongs to. We start by
sets of pixels within the same image. Segmenting an image defining the watershed transform function using elementary
makes it easier for us to further analyze and extract morphological operations. Next, we create an intensity
meaningful information from it. It is also described as “The gradient of the input image. Then we perform the final
process of labeling each pixel in an image such that they contour search on the gradient image with the help of the
share the same characteristics”.[8] The process results in watershed transform function. Sometimes this results in an
pixels sharing a common property . over-segmented image. To solve this problem we use
Marker-Controlled Watershed Segmentation. In this we

2017 7th International Conference on Cloud Computing, Data Science & Engineering – Confluence 583
predetermine certain elements of the image. Dr. A. S. noise. It works on the principle of convolution. Median
Bhalchandra[3], Ehab F. Badran[10] and Shantaram filter, unlike a Gaussian filter is a nonlinear filter. The result
Vasikarla[5] all use this technique. of this is that it is an edge preserving filter. On the other
hand in a Gaussian filter, as it is a low pass filter so the edge
C. Post-Processing: Successful Segmentation of the image information is lost and edges appear blurry and displaced.
is followed by the post-processing of the image. Post However, Gaussian filter is cheap to implement and less
Processing of the image involves steps to judge the size of complex than the median filter. Gaussian filter is very
the tumor and its type. Post processing may also involve effective in smoothening Gaussian noise. If edge
various optimization techniques to further improve the preservation is the main goal of pre-processing, then Sobel
result. Filter is a better option then both Median and Gaussian filter.
Dr. M. Karnan[1] proposed the use of Particle Swarm When it comes to segmentation, Thresholding is the easiest
Optimization. It is a meta-heuristic technique as it doesn't to implement and is widely used. This technique works well
require any initial assumptions. It works by creating a when the contrast between the background object and the
sample population and iteratively searches for the optimum foreground objects is relatively high, so that this difference
solution by upgrading generations. It is very similar to the in dynamic range can be thresholded. As segmentation is the
genetic algorithm, however, unlike the genetic algorithm it most important process while detecting tumor’s from an
does not require an evolutionary operator. MRI Image, the following table describes the typical usage
Another common post processing technique is Canny Edge and susceptibility factors for the various segmentation
Detection. It uses the principle that in an image there is a methods[11]. However, as thresholding is solely based on
sharp change in the contrast of the image at the boundaries the contrast of the image, this is not very useful as it does
of the objects present in the image. These points are known not extract much information from the input MRI. Due to its
as edges. It is at these points that there is a sharp change in simple nature, it is ideal for use in the initial stages of
the brightness. So to detect these edges we can use a processing.
gaussian filter to remove the noise and then use hysteresis to TABLE 1
detect the edges. This technique was used by A.Lakshmi[2] ADVANTAGES AND WEAKNESSES OF VARIOUS SEGMENTATION
TECHNIQUES
in their research.
In their paper, Sneha Khare[9] proposed using a Support
Vector Machine along with Curve Fitting for further
classification. Support Vector Machines are machine S Segmentatio Usage Susceptible To
learning models with learning algorithms which are used for No n Technique
regression analysis and classification. It learns at the initial
stages where the data should be predefined and label. It has
proven to be more accurate than other classification 1 Threshold • Should be used when the • Not able to extract
methods. Based contrast intensity is high much i n f o r ma t i o n
Segmentation • Less susceptible to noise from the input MRI
Morphological operations are also commonly used during than other images so can image and hence the
image post processing. These involve the use of a be used in conditions analysis may or not
mathematical structuring element to define the size of the where no pre-filtering is prove helpful
tumor and extracting meaningful information from the done • Cannot be used for
• When quick analysis of images with poor
image. Ehab F. Badran[10] proposed using a Harris-Laplace the image is to be done contrast or images
Or LOG-Lindeberg algorithm to further optimize the rather then exhaustive with a lot of
segmentation results whereas T.S. .Sadashivappa[6] used analysis background and
morphological operations like Binary Dilation and Binary foreground artifacts
Erosion for further optimization.
Chia-Chen Lin[4] and Deepthi Murthy[6] proposed using
Histogram Equalization. This is a contrast adjustment 2 Region Based • Watershed segmentation • Noise may lead to
technique and is very useful in images where the and region seed growing undesired artifacts in
background and foreground are both dark. It has widespread with well selected seed final result
use in X-Ray images as it enhances the visible bone regions can be used to • Susceptible to human
accurately extract object error as seed region is
structure in the image. of interests like tumors manually selected
The research by Ehab F. Badran[10] also proposed the use form input MRI’s
of a Convolutional Neural Network. A neural network is
inspired by the biological neural network of humans. It is of 3 Fuzzy C • Unsupervised techniques • Nois e has s e v e r e
great interest as it is a machine learning model that can learn Means, K and require least human affects on the final
and make interpretations from the input data thus further Means and interaction product
Level Set • Useful for large images • Sample selection and
optimizing the results and increasing the accuracy. Techniques with poor contrast establishing fuzzy sets
• Level set techniques can may be tedious
III. COMPARISON be used for computer-
Out of the many different filtering techniques available, the aided vision
median filter is the most commonly used. This is due to its
simplicity and its efficiency in removing salt-and-pepper

584 2017 7th International Conference on Cloud Computing, Data Science & Engineering – Confluence
Another common segmentation technique is watershed 7. Wikipedia contributors,"Sobel operator,” Wikipedia, The Free
segmentation. Unlike thresholding, this is a slow and Encyclopedia, https://en.wikipedia.org/w/index.php?
title=Sobel_operator&oldid=750123462 [Accessed: 21-May-2016].
demanding calculation. The success of the segmentation is 8. Wikipedia contributors, "Image segmentation," Wikipedia, The Free
dependent on the selection of the seed region. Also unlike E n c y c l o p e d i a , https://en.wikipedia.org/w/index.php?
thresholding, which is hardly effected by the presence of title=Image_segmentation&oldid=753820579.[Accessed 23-May-201]
noise, water shed segmentation isn’t. However, noise may 9. S. Khare, N. Gupta, and V. Srivastava, “Optimization technique, curve
fitting and machine learning used to detect Brain Tumor in MRI,”
lead to holes in the segmented image. Another common Proceedings of IEEE International Conference on Computer
technique is Fuzzy C Means Clustering and Fuzzy K means. Communication and Systems ICCCS14, 2014.
Both of these make use of fuzzy logic and are unsupervised 10. E. F. Badran, E. G. Mahmoud, and N. Hamdy, “An algorithm for
techniques and involves the generation of clusters. However detecting brain tumors in MRI images,” The 2010 International
this determination of fuzzy membership is a difficult task Conference on Computer Engineering & Systems, 2010.
11. Merlyn Mary Michael, “Survey on brain segmentation techniques,”
and hence makes it more CPU and memory intensive then International Journal of Modern Trends in Engineering and Research,
Watershed segmentation and Thresholding. Also, these vol. 1, no. 6, pp, 187-192, December 2014.
pixel-based segmentation techniques are highly susceptible 12. Daizy Deb, Bahnishikha Dutta and Sudipta Roy, “A noble approach for
to noise and hence proper preprocessing must be done[11]. noise removal from brain image using Region Filling,” 2014 IEEE
International Conference on Advanced Communications Control and
Comparison of segmentation techniques can also be made Computing Technologies, 2014.
on the basis of human interaction involved. In this case, 13. Al-Mohair, Hani K., Junita Mohamad Saleh and Shahrel Azmin Suandi.
segmentation can be classified as Manual Segmentation, "Hybrid Human Skin Detection Using Neural Network and KMeans
Semi Auto and Fully Automated[11]. When dealing with Clustering Technique, ” Applied Soft Computing, vol. 33, Issue C, pp.
digital images we hardly use Manual or the Semi Auto 337-347, August 2015.
14. Ibtihal D. Mustafa and Mawia A. Hassan “A Comparison between
techniques as the results will be poor and redundant to Different Segmentation Techniques used in Medical Imaging,”
errors. Also using manual segmentation techniques for American Journal of Biomedical Engineering, vol. 6, no. 2, pp. 59-69
computerized image processing renders its whole purpose 2016.
moot. 15. Kelvin K.L. Wong, Jiyuan Tu, et al “Cardiac flow component analysis,”
CONCLUSION Medical Engineering & Physics, 2010.
This paper surveys the various techniques that are part of
Medical Image Processing and are prominently used in
discovering brain tumors from MRI Images. At first the
various methods, that are being currently used in medical
image processing were extensively studied. This involved
studying the available research. Based on that research this
papers was written listing the various techniques in use. A
brief description of each technique is also provided. Also of
all the various steps involved in the process of detecting
tumors, Segmentation is the most significant and propitious.

REFERENCES
1. N. N. Gopal and M. Karnan, “Diagnose brain tumor through MRI using
image processing clustering algorithms such as Fuzzy C Means along
with intelligent optimization techniques,” 2010 IEEE International
Conference on Computational Intelligence and Computing Research,
2010.
2. J.selvakumar, A.Lakshmi and T.Arivoli, “Brain Tumor Segmentation
and Its AreaCalculation in Brain MR Images using K-Mean Clustering
and Fuzzy C-Mean Algorithm” 2012 IEEE-International Conference
On Advances In Engineering, Science And Management (ICAESM
-2012) March 30, 31, 2012.
3. Rajesh C. Patil, Dr. A. S. Bhalchandra, “Brain Tumour Extraction from
MRI Images Using MATLAB” International Journal of Electronics,
Communication & Soft Computing Science and Engineering ISSN:
2277-9477, vol. 2, no. 1, April 2012.
4. M.-N. Wu, C.-C. Lin, and C.-C. Chang, “Brain Tumor Detection Using
Color-Based K-Means Clustering Segmentation,” Third International
Conference on Intelligent Information Hiding and Multimedia Signal
Processing (IIH-MSP 2007), 2007.
5. R. Dubey, M. Hanmandlu, and S. Vasikarla, “Evaluation of Three
Methods for MRI Brain Tumor Segmentation,” 2011 Eighth
International Conference on Information Technology: New
Generations, 2011.
6. T. S. D. Murthy and G. Sadashivappa, “Brain tumor segmentation using
thresholding, morphological operations and extraction of features of
tumor,” 2014 International Conference on Advances in Electronics
Computers and Communications, 2014.

2017 7th International Conference on Cloud Computing, Data Science & Engineering – Confluence 585

You might also like