Performance Analysis of SVM Classifier for Classification of MRI Image

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 828
Performance Analysis of SVM Classifier for Classification of MRI Image
Aditi1, Sheetal Chhabra2
1M-Tech Student, Dept. of CSE, Maharishi Ved Vyas Engineering College, Kurukshetra University, Haryana, India
2Associate Professor, Dept. of CSE, Maharishi Ved Vyas Engineering College, Kurukshetra University, Haryana,India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In this research work classification of Magnetic
Resonance images is performed using support vectormachine
technique. For filtering the biomedical image median and
Gaussian filters are used and to enhancethe imagequalityand
finding region of interest k-mean clustering technique is
proposed. In this work two type of MRI tumors are taken into
consideration one is benign image and other is malignant
image. SVM classifier is used to classify within normal image,
benign image and malignant image using GLCM (graylevel co
matrix) feature extraction technique.
Key Words: Biomedical Imaging, MRI Image, Machine
Learning, SVM, GLCM
1. INTRODUCTION
Images play a significant role in today’s age of compact
information. The world of image method has exhibited
monumental progress over past few decades. Generally, the
photographs dealt in virtual environments or diversion
applications possess replication resulting in large storage
desires. Footage may bear distortions throughout
preliminary acquisitiontechnique,compression,restoration,
communication or final show. So image quality measuring
plays a serious role inseveral image-processingapplications.
Image quality, for scientific and medical functions, is made
public in terms of well desired information that may be
extracted from the image. An image is presupposed to own
acceptable quality if it shows satisfactory utility that
suggests discriminability of image content and satisfactory
naturalness that suggests identifiability of image content.
Digital storage of images has created a significant place in
imaging. Image quality metrics sq. necessary performance
variables for digital imaging systems and live unit of
measurement accustomed measure the visual quality of
compressed footage.
Brain tumour primarily based diseases are the highest
causes of death for person throughout the planet. The
common human has one chance in twelve of generating
malignant neoplastic disease throughout his life.
Abnormalities in brain decrease the thinking power thanks
to malignant neoplastic disease is being recognized by the
primary detection of malignant neoplastic disease pattern
the screening process [1] [2]. But it's presently a tricky call
to acknowledge the unsure irregularities in brain MRI
pictures rather than the increase in technology. It’s
ascertained that the primary process informs relatively low
distinction footage, notably inside the case of significant
brain condition and second; signs of irregular tissue may
keep dead refined. For example, speculated masses which is
able to indicate a malignant tissue at intervals the brain unit
of measurement usually robust to watch, notably at the
primary stage of development [3]. The recent use of the
second order statistics and machinelearning(ML)classifiers
has established a different analysis direction to watch
masses in digital X ray. The second order statistics of brain
tissue on brain MRI pictures remains a problem of major
importance in mass characterization. Compared to different
mammographic diagnostic approaches, the mammographic
has not been studied full thanks to the inherent drawback
and opaqueness [4].
2. RELATED WORK
MRI images can reflect details of different features to
provide an important basis for the diagnosis and treatment
of illness for patients. However, there are still some
restrictions in computer-based analysis of MRI medical
images, such as: the differences of imaging equipment,
imaging environment and imaging parameters among
patients; the redundancy, noise and other interference
factors inevitably from the formationoftheimages;thelarge
amount of image data from multiple sequences to be dealt
with; the uniform patients’ conditions among large
individuals; the lack of available priori knowledge; the
complex structure of MRI images, including differenttissues
in the internal and external region of the tumors; and the
lack of clarity aliasing of the tumor borders due to the
invasive characteristics of gliomas.
The traditional analysis on MRI images can ignore the
medical information implied in the images, and deal with
them directly as a general image processing problem to
operate. Corresponding to each two-dimensional MRI slice,
using some image processing methods to grasp regionswith
the same or similar characteristics toachievetheseparation.
Regions with coherent characteristics are not necessarily
consecutive, and there may be the case where the region is
composed of blocks. And other regions of different
characteristics are distinguished from other parts to
complete the preliminary analysis of the two-dimensional
slice. It can be observed that, the traditional tumordetection
in MRI images is a typical image analysis problem.

A variety of traditional image processing algorithms can be
applied to try to resolve this problem. The most common
algorithm is the threshold-based method [15], which is
always integrated with histogram analysis [16] to first
obtain the overall histogram distribution of the image, then
to select the appropriate threshold basedonthedistribution
of the binary image, and finally to finish segmenting the
image area supplemented post-processing processing of
morphological operations [16],suchasholefilling,boundary
improvement, and so on. However,thismethodistoosimple.
If the histogram of the image does not containobviouspeaks
or valleys, it will be difficult to selecttheoptimal threshold.A
commonly automatic threshold selection method is called
Otsu Threshold [17], but its accuracy cannot still reach a
higher level, that is because the algorithm itself is too simple
and cannot be adaptive to the complex situations of images.
Considering the characteristics of the whole tumor area,
such as the similarity, uniformity or gradual change among
the gray levels, region segmentation methods can be used to
achieve the detection of tumors. Commonly used algorithms
are region growing algorithm and regional separation and
merger algorithm [18]. The normal human brain tissue
indicates a symmetricstructure,thereforethealgorithm pre-
create multiple templates to construct a template library.
The image needed segmenting will be registered to the
templates one by one through different linear, non-linear or
combination maps to establish the corresponding
relationship between the segmented image and the
templates, so as to achieve the purpose of segmentation and
classification [19-22].
The basic elements of MRI images are the pixels, and each
pixel contains a variety of features about image properties,
including the basic gray-scale [23], a variety of features
derived from the characteristicsofthebasic features,suchas
texture features (including mean value, standard deviation,
and the commonly used Gray Level Concurrence Matrix
(GLCM) to reflect statistical characteristics [24-25], and so
on), mathematical transformation features (such as wavelet
transform features [26-28]) and so on. Various features are
related to the corresponding meaningful information in the
relevant field, which can indicate a specific physical meaning
through certain numerical values and its scope of change.
Pattern recognition algorithms research on the pixels and
include features as the study objectives. By analyzing and
comparingdifferentcharacteristics,thecorrespondingpixels
are classified into different categories. Commonly used
pattern recognition algorithms are clustering, Bayesian
probability model, linear and nonlinear discriminant
classification methods [29]. Clustering mainly uses the gray
values of the image, and merges data points of different
types according to the clustering criteria basedonsimilarity.
A typical unsupervised clustering algorithm is the nearest
neighbor method, in which the data is clustered into the
same category as that of the nearest point. The improved
algorithms are K-nearest neighbor (KNN) method (the data
is clustered into the same category as that of the K nearest
points), edited nearest neighbor (nearest neighbor method
becomes to a two-step process: the first step is the pre-
classification of data to remove the misclassification data,
and the next step is to classify using the KNN criterion to
improve accuracy), and so on. Nearest neighbor method is
relatively simple: one point is just compared with another
data point in classification.
2. METHODOLOGY
The approach consists in the main of 2 half: the image
process half and also theclassificationandtestingpart.Many
approaches that may be helpful for the image process half
exist. The main target here is on segmentation that has been
shown to be terribly helpful in issues like the one at hand.
Within the segmentation stageK-meanbunchisemployedto
extract the hemorrhage region from the image. Within the
following step, discriminative options of the region of
interest (ROI) are extracted. Finally, within theclassification
stage the image is assessed supported the computedoptions
of the ROI. MATLAB code is written to hold out the image
process and segmentation moreover because the options
extraction elements, whereas the wood hen tool was used
for the classification and testing elements.
Images of the patient are non-inheritable and keep at a
standard place. Some pictures are of traditional patientsthat
aren't affected by any wellness and rest is abnormal brain
pictures. The pictures that are already categorized as
cancerous and non-cancerous are used forreferenceandare
used for checking the potency of planned system. Pictures
are scanned and checked by radiologists whereas
rummaging magnetic resonance imaging. The initial step
when assortment of mister image is to convert them in gray
level and playing preprocessing.
At first image pre-processing is finished through bar chart
deed. Tumor features a dark look within the imagetherefore
bar chart deed is employed to create the perimeters visible.
Moot information is gift within the medical pictures.
Noise and inconsistencies may be removed and quality of
image and improvement of options also are consummated
through image pre-processing techniques.
A method within which bar chart of a picture isobtainedand
its distinction is adjusted is termed as bar chart deed. Some
pictures are having backgrounds and foregrounds with
completely different intensities; in these cases this deed
technique may be applied. It’s additionally referred to as
spatial domain improvement technique. [2].
It is the foremost basic technique describing image
segmentation. Here every component is related to some
worth. A binary image is formed by scrutiny the worth of
pixels with some threshold worth and pixels are termed as
object pixels if their value is larger than a threshold.
Associate in Nursing object component is assigned a price
one and alternative a price zero representing background.

Black and white colors are assigned counting on component
worth and a binary image is obtained. The process can be
described by following equation:
(1)
Each pixel’s intensity value is compared with the threshold
value and all the pixels having lesser value than the
threshold are kept only and other pixels are removed. Every
comparison is done starting from first row and first column.
Given image contain salt and pepper noise whichisremoved
in enhancement stage. The image shown in figure 2 contains
the noise and in the subsequent image noise being removed
and better appearance is achieved. This pre-processing step
make the segmentation process more effective as the better
contrast and improvement can be easily observedinthetwo
images. The threshold value is represented by T.
Fig – 1: Research Methodology of the Proposed System
3. IMAGE DENOISING
3.1 Median Filter
The median filter could be in style nonlinear digital filtering
technique, usually accustomed take away noise. Such noise
reduction could be a typical pre-processing step toboost the
results of later process (for example, edge detection on
associate degree image). Median filtering is incredibly wide
employed in digital image process as a result of below sure
conditions; it preserves edges whereas removing noise[24].
Typically called a rank filter, this abstraction filter
suppresses isolated noise by substitution every pixel’s
intensity by the median of the intensities of the pixels in its
neighborhood. It’s wide employed in de-noising and image
smoothing applications. Median filters exhibit edge-
preserving characteristics (cf. linear strategies like average
filtering tends to blur edges), that is incredibly fascinating
for several image process applications as edges contain
necessary data for segmenting, labelling and protective
detail in pictures. This filter may be represented by Eq (2).
(2)
Where 𝑤𝐹 = 𝑤 𝑥 𝑤 Filter window with pixel (𝑢,𝑣) as its
middle
3.2 Gaussian Filter
Gaussian filter could be a filter whose impulse response is
Gaussian operate [25]. Gaussian filters square measure
designed to offer no overshoot to a step operate input
whereas minimizing the increaseandfall time.This behavior
is closely connected to the actual fact that the Gaussian filter
has the minimum attainable cluster delay. Mathematically,a
Gaussian filter modifies the signal by convolution with a
Gaussian function;thisdemodulationisadditionallyreferred
to as the Weierstrass transform. Smoothing is usually
undertaken mistreatment linear filters like the Gaussian
operate (the kernel relies on the conventional distribution
curve), that tends to provide smart ends up in reducing the
influence of noise with relation to the image. The 1D and 2D
Gaussian distributions with standard deviation for a data
point (x) and pixel (x,y), are given by Eq (3) and Eq (4),
respectively [26].
(3)
(4)
The kernel might be extended to more dimensions likewise.
For a picture, the 2nd statistical distribution is employed to
produce a point-spread; i.e.blurringoverneighboringpixels.
This is often enforced on each picture element within the
image exploitation the convolution operation. The degree of
blurring is controlled by the letterofthealphabetorblurring
constant, likewise because the size of the kernel used
(squares with associate odd variety of pixels; e.g. 3×3, 5×5
pixels, in order that the picture element being acted upon is
within the middle). The process is accelerated by
implementing the filtering within the frequency instead of
spatial domain, particularly for the slower convolution
operation (which is enforced because the quicker
multiplication operation within the frequency domain).
3.3 Wiener Filter
Wiener filters are a category of optimum linear filters that
involve linear estimation of a desired signal sequence from
another connected sequence. It’s not reconciling filter. The
wiener filter’s main purpose is to scale back the number of
noise gift in a picture by comparison with estimation of the
required quiet image. The Wiener filter may additionally be
used for smoothing. This filter is that the mean squares
error-optimal stationary linear filter for pictures degraded
by additive noise and blurring. It’ssometimesapplied within
the frequency domain (by taking the Fourier transform)
[26], owing to linear motion or unfocussed optics Wiener
filter is that the most significant technique for removal of
blur in pictures. From a symbol process stand. Every
component in an exceedingly digital illustration of the
photograph ought to represent the intensity of one
stationary purpose ahead of the camera. Sadly, if the shutter
speed is just too slow and therefore the camera is in motion,
a given component is amalgram of intensitiesfrompoints on
the road of the camera's motion.
The goal of the Wiener filter is to filtrate noise that has
corrupted a symbol. It’s supported an applied math
approach. Typical filters are designed for a desired
frequency response. The Wiener filter approaches filtering
from a unique angle. One is assumed to own data of the

spectral properties of the first signal andthereforethenoise,
and one seeks the LTI filter whose output would return as
near the first signal as doable [27]. Wiener filters are
characterized by the following:
1. Assumption: signal and (additive) noise are stationary
linear random processes with known spectral
characteristics.
2. Requirement: the filter must be physically realizable, i.e.
causal (this requirement can be dropped, resulting in a non-
causal solution).
3. Performance criteria: minimum mean-square error.
Wiener Filter in the Fourier Domain as in Eq (5).
(5)
Where
(𝑢,𝑣) = Fourier transform of the point spread function
𝑃(𝑢,𝑣) = Power spectrum of the signal process, obtained by
taking the Fourier transform of the signal autocorrelation
𝑃(𝑢,𝑣) = Power spectrum of the noise process, obtained by
taking the Fourier transform of the noise autocorrelation
It should be noted that there are some known limitations to
Wiener filters. They are able to suppress frequency
components that have been degraded by noise but do not
reconstruct them. Wiener filters are also unable to undo
blurring caused by band limiting of (𝑢,𝑣), which is a
phenomenon in real-world imaging systems.
4. FEATUREEXTRACTION: GREYLEVELCO-OCCURRENCE
MATRICES (GLCM)
The textural choices supported gray-tone abstraction
dependencies have a general connection in image
classification. The three elementary pattern components
utilized in human interpretation of images square measure
spectral, textural and discourse choices. Spectral choices
describe the everyday tonal variations in varied bandsof the
visible and/or infrared portion of a spectrum. Textural
choices contain information concerning the abstraction
distribution of tonal variations at intervals a band. The
fourteen textural decisions contain data regarding image
texture characteristics like homogeneity, gray-tone linear
dependencies, contrast, choice and nature of boundaries gift
and therefore the quality of the image. Discourse decisions
contain data derived from blocks of pictorial information
encompassing the realm being analyzed.Haralicketall initial
introduced the utilization of co-occurrence prospects
victimization GLCM for extracting varied texture decisions.
GLCM is additionally referred to as grey level Dependency
Matrix. It’s printed as “A two dimensional chartofgraylevels
for an attempt of pixels that square measure separated by a
tough and quick abstraction relationship.”GLCMofanimage
is computed using a displacement vector d, printed by its
radius δ and orientation θ. ponder a 4x4 image pictured by
figure 1a with four gray-tone values zero through 3. A
generalized GLCM for that image is shown in table1 where
stands for style of times gray eight tones i and j
neighbors satisfying the condition declaredbydisplacement
vector d.
Texture options from GLCM variety of texture optionsisalso
extracted from the GLCM [20]. Wehavea tendencytousethe
subsequent notation: Here G stands for variety ofgreylevels
taken. µ is parameter stands for the normalization of
P. , , and are the parameters and customary
deviations of and . (i) is that the ith entry within the
marginal-probability matrix obtained by summing the rows
of [21]
H (6)
(7)
(8)
(9)
(10)
(11)
and
(12)
for k=0,1,…..2(G-1).
(13)
for k=0,1,…..2(G-1).
Homogeneity, Angular Second Moment (ASM) [22]
(14)
Angular SecondMomentisthemeasurementofhomogeneity
of an image. A homogeneous picture contained some gray
levels, provide a grey level co-occurrence matrix with a
relatively larger but few values of P(i, j). Thus larger values
will be obtained after summing of squares.
Contrast [22]
(15)
The contrast measurement or variation in image will
contribute from P(i, j) away from the diagonal, i.e. .
Local Homogeneity, Inverse Difference Moment (IDM) [22]:
(16)
Inverse Difference Moment is a parameter that isaffectedby
the homogeneity term of the image. Due to weighting factor
Inverse Difference Moment will lead little
contributions from nonhomogeneous zone ( ).
Energy [23]
(17)
Entropy [22]
(18)
Nonhomogeneous images have little first order entropy,
while a homogeneous picture has larger entropy.
Auto Correlation [23]:
(19)
Correlation [22]:

(20)
Correlation is the measurement of gray level linear
dependence between the pixels at the specified positions
relative to each other.
Sum of Squares, Variance [22]
(21)
This parameter is most effective element which provide
relatively high weights on the image elements that is
different from the average value of H(i, j).
Sum Average [22]
(22)
5. SVM BASED CLASSIFICATION
SVM is a classification algorithm for high-dimensional data
analysis which is proposed by Vapnik to solve the
classification problems of two issues [28]. SVM has been
widely used in the fields of medical image processing, text
analysis, image retrieval and so on. SVM is based on the
principle that the data in the original input space can be
linear separable in a higher-dimensional feature space after
a certain mapping. The feature space after mapping is a
complete Hilbert space, in which the inner product of data
can be calculated as the equivalent function value of the
input vectors by introducing the corresponding kernel
function. The inner product is the measurement of the
distance, which can be expressed as the degree of similarity
between two vectors in a certain extent. In general, the
closer the distance between two vectors is, the higher the
similarity of them is. Due to the kernel function, the
computation of distance in the feature space is transformed
to the input space without the need to solve the mapping of
spatial transformation, therefore, it reduces the
computational difficulty to a certain extent.
SVM has many extensions, such as one-classification SVM, in
which all the interested data will be considered as one class
and all the other data as the other class. In the classification,
the training is only fulfilled on the data of the first class. This
is a simple application of two-class SVM classification [29].
Another important application of SVM is the multi-kernel
SVM [30-31]. It is especially suitable for dealing with high-
dimensional data of different sources and different types.
The classification algorithm does not use a uniform kernel
function in order to be able to dig the implicit information in
the data and to take full use of the characteristics in thedata.
In general, the multi-kernel SVM has two typical forms:
1. For different types of input data from different sources,
use different kernel functions or different parameterswitha
same kernel function to design different classifiers for each
input respectively, select the optimal parameters to classify
each input one by one to obtain the optimal results.Thefinal
result is the integration of all the optimal results of the input
data;
2. First, some strategy is adopted to fuse all the input data,
and then use different kernel functions or different
parameters of a same kernel function to carry out the
classification; finally using the selected fusion strategy to
obtain the integrated results.
The two forms of multi-kernel SVM have very different
characteristics. The former takes full use of the propertiesof
their own in different types of data, which are independent.
The use of different kernel functions can mine a variety of
important and implied characteristics of data and the
ultimate classification result is the integration of all the
optimal results. The latter fuses the data first to use the
implicit information, characteristics or relationship in
various data, although different kernel functions are applied
to all the parts of the data. The more important point in this
processing mode is to use the correlation and mutual
influence of data to achieve the classification. In addition to
adapt to multiple types of input data, multi-kernel SVM can
also extended to the multi-class classification problem.
At present, the multi-class classification problems on brain
tissues’ detection mainly depends on the fuzzy theory or the
probability models, both of which are not very accurate in
general. Some algorithms can just segment edema and
normal tissues separately, which indicates the lack of ability
to deal with brain tumors.
There is no specific multi-class classification algorithm in
traditional SVM. The existing multi-class classification
strategies are the simple expansions of two-class
classification algorithms, such as a commonly used one
versus all strategy, in which it is necessary to construct a
two-class classifier for each classification of different data.
The interested data will be considered as one class and all of
the other data as the other class. Each class has its special
classifier to make the margin between this class of samples
and the other samples the largest. Then the multi-class
classification can be achieved by determine each category
one by one.
6. RESULTS & DISCUSSION
In this project, method consists of three stages:
Step 1. Preprocessing (including feature extraction and
feature reduction);
Step 2. Training the kernel SVM;
Step 3. Submit new MRI brains to the trained kernel SVM,
and output the prediction.
In this dissertation two types of MRI images are taken into
consideration, one is benign MRIandother ismalignantMRI.
A benign tumor is not a malignant tumor, which is cancer. It
does not invade nearby tissue or spread to other parts of the
body the way cancer can. In most cases, the outlook
with benign tumors is very good. But benign tumors can be
serious if they press on vital structures such asbloodvessels

or nerves. Database of benign MRI images are shown in fig -
2.
Fig – 2: Benign MRI Images
Malignant tumors are cancerous and are made up of cells
that grow out of control. Cells in these tumors can invade
nearby tissues and spread to other parts of the body.
Sometimes cells move away from the original (primary)
cancer site and spread to other organsandboneswherethey
can continue to grow and form another tumor at that site.
This is known as metastasis or secondarycancer.Metastases
keep the name of the original cancer location. eg. pancreatic
cancer that has spread to the liver is still called pancreatic
cancer.
Database of malignant MRI images are shown in fig - 3.
Fig – 3: Malignant MRI Images [32]
Testing of MRI image is done using MATLAB software.
In second step segmentation of selected image is done by
clicking on segmented image tab. In third step classification
is done suing KSVM technique and result is shown in tumer
type box.
Selected image is shown in fig -4 (a) and segmented image is
shown in fig - 4(b).

Fig – 4: Test 1 (a) Original MRI Image (b) Segmented Image
Original image is inserted to system and after its
segmentation support vector machine start and check the
features and finding matchingwithdata basefinallyevaluate
the type of tumor, which is Benign type tumor resulted for
above image. And accuracyisalsocalculated,RBF accuracyis
80%, linear accuracy is 80%, polygonal accuracy is 80%and
quadratic accuracy is 90%.
Another image is taken for classification as shown original
MRI image in fig. 5(a) and its segmented image is shown in
fig. 5(b)
Fig – 5: Test 2 (a) Original MRI Image (b) Segmented Image
This tested image is resulted malignant type tumor. And
accuracy is also calculated, RBF accuracy is 80%, linear
accuracy is 90%, polygonal accuracy is 80% and quadratic
accuracy is 80%.
7. CONCLUSION
In this research paper an introduction to different type of
MRI image classification techniques are introduced and
special focus on support vector machine. And how
classification accuracy can be achieved depending on the
changes and other neural approaches. Neural when
combined with fuzzy logic also gives an efficient classifier.
The classifiers described in this thesis categorizes the image
as normal and abnormal and present the location of tumour
via clustering.Themanual procedurethatpathologistchoose
for diagnosis is microscopic detection which is often time
consuming and causes fatigue to them, hence this proposed
system is quite beneficial . But there are several forms in
which a tumour is categorized depending upon the size and
location of tissues inside the brain.
REFERENCES
[1] A. K. Jain, 1989, “Fundamentals of digital image
processing,” Prentice-Hall Inc.,EnglewoodCliffs,New Jersey.
[2] J. S. Lee, 1980, “Digital image enhancement and noise
filtering by use of local statistics,” IEEE Trans. on PAMI,
Vol.2, No.2, pp.165- 168.
[3] S. G. Chang, B.Yu and M. Vetterli,2000, “Spatiallyadaptive
wavelet thresholding with context modeling for image
denoising,” IEEE Trans. on Image Processing, Vol.9,No.9, pp.
1522-1531.
[4] M. K. Mihcak, I. Kozintsev, and K. Ramchandran, 1999,
“Spatially Adaptive Statistical Modeling of Wavelet Image
Coefficients and its Application to Denoising,” Proc.IEEEInt.
Conf. Acoust., Speech, Signal Processing, Vol.6, pp. 3253–
3256.
[5] M. Sezgin, B. Sankur. Survey over Image Thresholding
Techniques and Quantitative Performance Evaluation [J].
Journal of Electronic Imaging, 2004, 13 (1): 146–165.
[6] Y.J. Zhang. Image Processing and Analysis [M]. Beijing:
Tsinghua University Press, 1999.
[7] N. Otsu. A Threshold Selection Method from Gray-Level
Histograms [J]. IEEE Transactions on System, Man and
Cybernetics, 1979, 9 (1): 62-66.
[8] R.C. Gonzalez. Digital Image Processing [M]. Peking:
Publishing House of Electronics Industry, 2003.
[9] M.R. Kaus, S.K. Warfield, P.M. Black, et al. Automated
Segmentation of MR Images of Brain Tumors [J]. Radiology,
2001, 218: 586-591.
[10] M. Prastawa, E. Bullitt, S. Ho, et al. A Brain Tumor
Segmentation Work based on Outlier Detection [J]. Medical
Image Analysis, 2004, 8: 275-283.
[11] M. Prastawa, E. Bullitt, N. Moon, et al. Automatic Brain
Tumor Segmentation by Subject Specific Modification of
Atlas Priors [J]. Medical Image Computing, 2003, 10: 1341-
1348.
[12] M.B. Cuadra, C. Pollo, A. Bardera, et al. Atlas-based
Segmentation of Pathological MR Brain Images Using a
Model of Lesion Growth [J]. IEEE Transactions on Medical
Imaging, 2004, 10: 1301-1313.
[13] P.Y. Lau, F.C.T. Voon, and S. Ozawa. The Detection and
Visualization of Brain Tumors on T2-weighted MRI Images
Using Multiparameter Feature Blocks [C]. 27th IEEE Annual
Conference of the Engineering in Medicine and Biology
(EMBS), 2005: 5104-5107.
[14] R.A. Lerski, K. Straughan, L.R. Schad, et al. MR Image
Texture Analysis: AnApproachtoTissueCharacterization [J].
Magnetic Resonance Imaging, 1993, 11: 873-887.
[15] A. Zizzari, U. Seiffert and B. Michaelis. Detection of
Tumor in Digital Images of the Brain [C]. Poceedings of the
IASTED International Conference on Signal Processing,

Pattern Recognition and Applications (ICSPPRA),2001:132-
137.
[16] R. Nezafat and H. Soltanian-Zadeh. Multiwavelet-based
Feature Extraction for MRI Segmentation [C]. Processing of
SPIE Wavelet Applications in Signal and ImagingProcessing,
1998, 3458: 182-191.
[17] K.M. Ifterkharuddin, M.A. Islam, J. Shaik,etal.Automatic
Brain Tumor Detection in MRI: Methodology and Statistical
Validation [C]. Proceedings of Medical Imaging, 2005, 5747:
2012-2022.
[18] S. Herlidow-Meme, J.M. Constans, B. Carsin, et al. MRI
Texture Analysis on Texture Test Objects, Normal Brain and
Intracranial Tumors [J]. Magnetic Resonance Imaging,2001,
21: 989-993.
[19] Z.Q. Bian and X.G. Zhang. Pattern Recognition [M].
Beijing: Tsinghua University Press, 2000.
[20] R. M. Haralick, K. Shanmugam and I. Dinstein “Textural
features for Image Classification”, IEEE Transactions on
Systems, Man andCybernetics,Vol.3,pp.610-621,November
1973.
[21] JC-M. Wu, and Y-C. Chen, “Statistical Feature Matrix for
Texture Analysis”, Computer Vision, Graphics, and Image
Processing; Graphical Models and Image Processing,Vol.54,
pp. 407-419, 1992.
[22] M.M. Trivedi, R.M. Haralick, R.W. Conners, and S. Goh,
”Object Detection based on Gray Level Co-ocurrence”,
Computer Vision, Graphics, and Image Processing, Vol. 28,
pp. 199-219, 1984.
[23] Sushila Shidnal, “A texture feature extraction of crop
field images using GLCM approach”, IJSEAT, Vol 2, Issue 12,
December – 2014.
[24] H. Raymond Chan, Chung-Wa Ho,andM.Nikolova,“Salt-
and-Pepper noise removal by median-type noise detectors
and detail-preserving regularization”. IEEE Trans Image
Process 2005; 14: 1479-1485.
[25] P. Hsiao1, S. Chou, and F. Huang, “Generic 2-D Gaussian
Smoothing Filter for Noisy Image Processing”. National
University of Kaohsiung, Kaohsiung, Taiwan, 2007.
[26] S. Kumar, P. Kumar, M. Gupta and A. K. Nagawat,
“Performance Comparison of Median and Wiener Filter in
Image De-noising,” International Journal of Computer
Applications 12(4):27–31, December 2010.
[27] Kazubek, “Wavelet domain image de-noising by
thresholding and Wiener filtering”. M. Signal Processing
Letters IEEE, Volume: 10, Issue: 11, Nov. 2003 265 Vol.3.
[28] V. Vapnik. The Natureof Statistical Learning Theory[M].
London: Springer, 1995.
[29] J. Zhou, K.L. Chan, V.F.H. Chong, et al. ExtractionofBrain
Tumor from MR Images Using One-Class Support Vector
Machine [C.] 27th Annual International Conference on
Engineering Medicine and Biology Society (EMBS), 2005:
6411-6414.
[30] Z. Wang, S.C. Chen and T.K. Sun. MultiK-MHKS: A Novel
Multiple Kernel LearningAlgorithm[J].IEEE Transactions on
Pattern Analysis andMachineIntelligence,2008(2),30:348-
353.

Performance Analysis of SVM Classifier for Classification of MRI Image

More Related Content

What's hot (20)

Similar to Performance Analysis of SVM Classifier for Classification of MRI Image (20)

More from IRJET Journal (20)

Recently uploaded (20)

Performance Analysis of SVM Classifier for Classification of MRI Image