Classification of Abnormalities in Brain MRI Images Using PCA and SVM

Ajala F. A. et al. Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 5, Issue 7, ( Part - 1) July 2015, pp.56-62
www.ijera.com 56|P a g e
Classification of Abnormalities in Brain MRI Images Using PCA
and SVM
Ajala Funmilola Aa
.,Fenwa Olusayo Db
.,Amusan Elizabeth A.c
a b c
Department of Computer Science and Engineering, LAUTECH, P.M.B 4000, Ogbomoso, Nigeria
(*
Corresponding authors: fenwadeborah@yahoo.com,eaadewusi@lautech.edu.ng , faajala@lautech.edu.ng)
Abstract
The impact of digital image processing is increasing by the day for its use in the medical and research areas.
Medical image classification scheme has been on the increase in order to help physicians and medical
practitioners in their evaluation and analysis of diseases. Several classification schemes such as Artificial Neural
Network (ANN), Bayes Classification, Support Vector Machine (SVM) and K-Means Nearest Neighbor have
been used. In this paper, we evaluate and compared the performance of SVM and PCA by analyzing diseased
image of the brain (Alzheimer) and normal (MRI) brain. The results show that Principal Components Analysis
outperforms the Support Vector Machine in terms of training time and recognition time.
Keywords: Alzheimer, Support Vector Machine, Principal Components Analysis, Medical Image
I. Introduction
In recent decades, the need for computers
inassisting the processing and analysis of medical
imageshas become inevitable with the mounting size
andnumber of medical images [1].Brain tumors are
abnormal and uncontrolled proliferations of cells.
Some originate in the brain itself, in which case they
are termed primary. Others spread to this location
from somewhere else in the body through
metastasis, and are termed secondary. Primary brain
tumors do not spread to other body sites, and can be
malignant. Secondary brain tumors are always
malignant. Both types are potentially disabling and
life threatening,this is because the space inside the
skull is limited, their growth increases intracranial
pressure, and may cause edema, reduced blood flow,
displacement with consequent degeneration of
healthy tissue that controls vital functions [2]. Brain
tumors are, in fact, the second leading cause of
cancer related deaths in children and young adults.
According to the Central Brain Tumor Registry of
the United States (CBTRUS), there were 64,530
new cases of primary brain and central nervous
system tumors diagnosed by the end of 2011. In
general, more than 600,000 people currently live
with the disease (www.cewebsource.com).
A major challenge facing MRI brain image
segmentation is that the brain tissue is often divided
into white matter (WM), graymatter (GM) and
cerebrospinal fluid (CSF). The precise measurement
of WM, GM and CSF is important for quantitative
pathological analyses and so it becomes a goal of
lots of methods for segmenting MRI brain image
data.Because of the imperfections of imaging
scanner and imaging techniques, obtained medical
images will inevitably be affected by some
corruption factors such as random additive noises,
partial volume effect and intensity bias field. For
improving segmentation performance, many
different strategies, for example, Mercer Kernel
techniques, Filtering techniques and so on, can be
adopted [3].However, Alzheimer‟s Disease (AD) is
a condition where the brain slowly goes down, as
well as a serious loss of thinking ability in a person
and cognitive impairment. PET scan image is
mainly used for the Alzheimer‟s neurological
disorder treatment and other kind of dementia. In
2006, there were 26.6million sufferers worldwide.
Alzheimer's is predicted to affect 1 in 85 people
globally by 2050.
II. Related Works
A number of approaches have been used to
segment and predict the grade and volume of the
brain tumor. [4], in their work proposed a Fuzzy
Cognitive Map (FCM) to find the grade value of
tumor. Authors used the soft computing method of
FCM to represent and model expert‟s knowledge.
FCM grading model achieved a diagnostic output
accuracy of 90.26% and 93.22 % of brain tumors of
low grade and high grade respectively. They
proposed the technique only for characterization and
accurate determination of grade. [5] in their work
proposed an implementation of evaluation method
known as Image Mosaicking in evaluating the MRI
brain abnormalities segmentation study. 57 mosaic
images are formed by cutting various shapes and
size of abnormalities and pasting it onto normal
brain tissue. PSO, ANFIS, FCM are used to segment
the mosaic images formed. Statistical analysis
method of Receiver Operating Characteristic (ROC)
was used to calculate the accuracy.
RESEARCH ARTICLE OPEN ACCESS

[6], in their work proposed detection of tumor
growth by advanced diameter technique using MRI
data. To find the volume of brain tumor they
proposed diameter and graph based methods. The
result showed tumor growth and volume.[8]
proposed a system that automatically segments and
labels tumor in MRI of the human brain. They
proposed a system which integrates knowledge
based techniques with multispectral analysis. The
results of the system generally correspond well to
ground truth, both on a per state basis and more
importantly in tracking total volume during
treatment over time.
In this work,classification system for
abnormalities of MRI brain images using principal
component analysis, and support vector machine
was proposed.
III. METHODOLOGY
The block diagram of the proposed system
showing the stages of the system development is as
displayed in figure 3.1. The stages are: Medical
Image Acquisition; Image Preprocessing; Feature
Extraction; Classification (using SVM and ANN);
and Performance Evaluation.
Figure 3.1 Model Arhcitecture of the Proposed System
Extract ion of intensity
based features using
PCA
Image
Classification using
PCA
Image Classification
using SVM
Performance Evaluation
of PCA and SVM

3.1 Image (Data) Acquisition
The image used for this research work was
obtained from an online database of brain MRI
images. Typical data samples consisting of
Alzheimer diseased brain and normal brainare as
shown in figures 3.2. and 3.3. The database provides
a repository of these images which can be
downloaded and regenerated in the MatLab
environment. Some of these images are stored for
research purposes, and for other image processing
analaysis. After the images were gotten from an
online source, a database that contains both images
was created in the MatLab environment. The image
was called from the database using Matlab algorithm.
Figure3.2: A typical image of the Alhzeimer Disease
(Source : www.mednet.com)
Figure 3.3: Normal MRI Brain(Source:
www.mribrain.com)
3.2 Image Pre-processing
Pre-processing of image is necessary before any
image analysis can be carried out. It involves
filtering, selecting, randomizing, conversion to gray-
scale, resizing and removal of objects that could
affect the proper processing of the images. In analysis
of the medical images used in this paper, we try to
avoid image pre-processing because image
preprocessing can decrease image information
content.
3.2.1 Pre-processing to gray scale
A major pre-processing is conversion to
grayscale. Most images obtained are always in
colored form, and the only way to process such
image is by conversion to gray scale. An RGB image
is a 3 by 3 image matrix consisting of rows, columns,
and index type. The value of an image is also
identified by its class, such as uint8, uint16, double.
For a grayscale image, it is made to be 2 by 2 matrix,
depending on the image size, and also retain its class.
3.3 Feature Extraction
The purpose of feature extraction is to reduce the
original data set by measuring certain properties of
features that distinguish on input pattern from
another.
3.3.1 Texture features
Texture is a very useful characterization for a
wide range of image. It is generally believed that
human visual systems use texture for recognition and
interpretation. In general, color is usually a pixel
property while texture can only be measured from a
group of pixels. A large number of techniques have
been proposed to extract texture features. Based on
the domain from which the texture feature is
extracted, they can be broadly classified into spatial
texture feature extraction methods and spectral
texture feature extraction methods. For the former
approach, texture features are extracted by computing
the pixel statistics or finding the local pixel structures
in original image domain, whereas the latter
transforms an image into frequency domain and then
calculates feature from the transformed image.
3.3.2 Shape Features
Shape is known as an important cue for human
beings to identify and recognize the real-world
objects, whose purpose is to encode simple
geometrical forms such as straight lines in different
directions. Shape feature extraction techniques can be
broadly classified into two groups, viz., contour
based and region based methods. The former
calculates shape features only from the boundary of
the shape, while the latter method extracts features
from the entire region.
3.4. Feature Selection
Feature selection (also known as subset
selection) is a process commonly used in machine
learning, wherein a subset of the features available
from the data is selected for application of a learning
algorithm. The best subset contains the least number
of dimensions that contributes to high accuracy; we
discard the remaining, unimportant dimensions.
3.4.1. Forward Selection
This selection process starts with no variables
and adds them one by one, at each step adding the
one that decreases the error the most, until any
further addition does not significantly decrease the
error. We use a simple ranking based feature
selection criterion, a two –tailed t-test, which
measures the significance of a difference of means
between two distributions, and therefore evaluates the
discriminative power of each individual feature in
separating two classes.The features are assumed to

come from normal distributions with unknown, but
equal variances. Since the correlation among features
has been completely ignored in this feature ranking
method, redundant features can be inevitably
selected, which ultimately affects the classification
results. Therefore, we use this feature ranking method
to select the more discriminative feature, e.g.by
applying a cut-off ratio (p value<0.1), and then apply
a feature subset selection method on the reduced
feature space.
3.4.2. Backward Selection
This selection process starts with all the
variables and removes them one by one, at each step
removing the one that decreases the error the most (or
increases it only slightly), until any further removal
increases the error significantly. To reduce over
fitting, the error referred to above is the error on a
validation set that is distinct from the training set.
The support vector machine recursive feature
elimination algorithm is applied to find a subset of
features that optimizes the performance of the
classifier. This algorithm determines the ranking of
the features based on a backward sequential selection
method that remove one feature at a time. At each
time, the removed feature makes the variation of
SVM based leave-one-out error bound smallest,
compared to removing other features.
3.5 Principal Components Analysis
Principal components are the projection of
the original features onto the eigenvectors and
correspond to the largest Eigenvalues of the
covariance matrix of the original feature set. PCA
provides linear representation of the original data
using the least number of components with the mean
squared error minimized. PCA can be used to
approximate the original data with lower dimensional
feature vectors. The basic approach is to compute the
eigenvectors of the covariance matrix of the original
data, and approximate it by a linear combination of
the leading eigenvectors. By using PCA procedure,
the test image can be identified by first projecting the
image onto the Eigen-space to obtain the
corresponding set of weights, and then comparing
with the set of weights of the images in the training
set.
3.5.1 Algorithm for Principal Component
Analysis
Data: Intensity and label training images
Result: Intensity, label, and intensity/label coefficient
PCA subspaces.
Step 1:Register the training images group-wise
Step 2: Apply transformations from the registration to
the respective label data
Step 3: Compute the PCA basis for the registered
intensity training data
Step 4: Compute the binary images from the
registered training labels
Step 5: Compute the PCA basis for the binary images
Step 6: Project the binary label and intensity training
images onto their respective bases
Step 7 Compute the PCA basis of the coefficients of
the training data from theprojection
3.6 Support Vector Machine
Support Vector Machines are a state of the
art pattern recognition technique grown up from
statistical learning theory. SVM is capable of
producing a hyper plane which linearly separates two
distinct classes, or two different types of images.
SVM uses an optimum linear separating hyper plane
to separate two set of data in feature space as shown
in figure 3.4. This optimum hyper plane is produced
by maximizing minimum margin between the two
sets. Therefore the resulting hyper plane will only be
dependent on border training patterns called support
vectors. The standard SVM is a linear classifier
which is composed of a set of given support vectors z
and a set of weights w.
Figure 3.4: Linear separation in feature space
The computation for the output of a given SVM with
N support vectors z1, z2, ....,zN and weights w1, w2,
...., wN is then given by:
F(x) = wi zi, x + bN
i=1
(1)
3.6.1 Algorithm for Image Classification using
Support Vector Machine
Step 1: Input image into Mat lab
Step 2: Perform Pre-Processing, and Normalization
Step 3: Perform feature extraction by extracting
surface texture and roughness
Step 4: For 100 image datasets, label first 50 normal
image as 1, and the last 50 images as 2.

Step 5: Create train datasets and test datasets
Step 6: In test datasets, for another 100 image
datasets, label first 50 normal images as 1,
and the last 50 abnormal images as 2.
Step 7: Perform SVM classification by calling the
command svmstruct in Mat lab.
Step 8: Check classification by calling the variable
group
IV. Results and Discussion
4.1. System Design and Implementation
This involves loading the image anywhere in the
computer to get the image. The first stage of any
image processing is getting the image. After the
image has been gotten, the various tasks which
involve processing the image, analyzing the image
was now performed. For this purpose, an interactive
interface was developed which was used to carry out
the process.
4.2 Mat lab Implementation
Any image which comes into the Mat lab
environment has a definite value, which is identified
by Mat lab. In image processing Mat lab enhances
the image and generates its value using the im-read
function. This enables further processing. Image
components are also gotten in the work environment,
after its value has been gotten. This is very important
because it helps to enhance image processing. These
components are identifiable and are attached to each
image, this implies that for every image coming into
the Mat lab environment. It has a definite component,
which makes it very useful in analysis.
Figure 4.1: User Interface for Analysis
4.3 Training Data
In this work, a total number of 50 images containing both Normal MRI Brain and Alhzeimerdisease were used.
The features of the images were extracted, and passed into training using principal component analysis, while
another set of 50 images were used for testing. The same set of images was passed into the Support Vector
Machine, and another set of these images were tested and evaluated. Figures3.2 and 3.3 show typical images of
the Alzheimer Disease and Normal MRI Brain, from which features were extracted.
SVM Results
SVM maps input vectors to a higher dimensional vector space where an optimal hyper plane is constructed. The
data with linear severability may be analyzed with a hyper plane, and the linearly non separable data are
analyzed with kernel functions such as GaussianRBF. The output of an SVM is a linear combination of the
training examples projected onto a high dimensional feature space through the use of kernel function. For this
work SVM with kernel function linear and RBF (Radial Basis Function) is used for classification of images into
two classes namely “Alhzeimer” and “Normal MRI Brain”. The labels for these classes are using „1‟ and „2‟
for “Normal” and “Abnormal” respectively.

A total of 50 images were passed into the database for training and testing. The resulting classification shows
the SVM result generating a confusion matrix as shown in table 4.1 below. Confusion matrix is a table matrix
which shows correct classification, and misclassification. Correct Classification occurs when Normal Brain is
classified as Normal and Diseased Brain is classified as Diseased Brain.
Table 4.1: Confusion Matrix for SVM Classification
Normal Brain Diseased Brain
(Alzheimer)
Normal Brain 19 6
Diseased Brain
(Alhzeimer)
6 19
PCA Results
The principal components of the images were extracted, and trained while another set of test images were used
for testing, the results gotten was poor this is because principal components of an image are not enough for
classification. The result is shown in table 4.2 below:
Table 4.2: Results for PCA and SVM
Correct
Classification
Incorrect Classification
PCA 26% 74%
SVM 76% 24%
PCA picks components based on Eigen Values and Eigen Vectors.These components are used for
classifications; however, they don‟t give accurate result during testing, due to redundancy in image.SVM results
seem better as compared to the PCA results.
4.4 Training time
The training time is the time required by the system to train and understand the behavior of the images in
question, during training patterns are recognized, studied and evaluated until after which the carefully studied
pattern are extracted, and used for testing. Training time is of significant importance, as it is used for evaluation
of speed and performance. The training and recognition time for both classifiers is shown in table 4.3 below.
4.5 Recognition Time
Recognition of images occurs when the classifier is able to correctly identify and acquaint itself with what it has
been trained with. Recognition time plays an important role in classification of medical images because higher
recognition time could lead to memory consumption which could affect corresponding results. In the case of a
lower recognition time, it yields low memory which is useful and better, and does not affect corresponding
results.
Table 4.3: Training and Recognition Time as for both classifiers
In table 4.3 above, the training and recognition time is seen and observed for the PCA and SVM. The training
time and recognition time of PCA seem smaller and shorter than that of the SVM, hence PCA performs better in
terms of training and recognition.
Number of
Images
Training Time
(in seconds)
Recognition Time
(in Seconds)
PCA 50 6.4847 0.006
SVM 50 37.406 0.0589

4.6. Performance Evaluation Metrics
The performance evaluation metrics used
are:Accuracy, Specificity, Sensitivity, True Positive,
True Negative, False Positive and False Negative.
Accuracy- is defined as the accurate value of
classification which equals the total sum of correct
classification over the total sum of correct and
incorrect classification multiplied by 100.
Specificity- is defined as the total division of true
positive against the total of true positive and false
negative, while sensitivity is defined as true negative
against true negative and false positive.
True positive (Tp)- occurs when the correct
classification for the right image is done i.e. when a
normal brain is classified as normal, while false
negative (Fn) occurs when a normal brain is
incorrectly classified as diseased.
True negative (Tn)- occurs when the correct
classification for the right image is done i.e. when a
diseased image is classified as being diseased, while
false positive (Fp) occurs when a diseased image is
incorrectly classified to be normal.
Specificity =
𝑇𝑝
𝑇𝑝+𝐹𝑛
∗ 100; Sensitivity =
𝑇𝑛
𝑇𝑛+𝐹𝑝
∗ 100;
Accuracy =
𝑇𝑛+𝑇𝑝
𝑇𝑝+𝑇𝑛+𝐹𝑛+𝐹𝑝
∗ 100
where, Tp is true positive (i.e. Normal Brain
classified correctly as Normal Brain), Fn is false
negative (i.e. Normal Brain is incorrectly classified
as Diseased), Tn is true negative (i.e. Diseased Brain
correctly classified as Diseased), Fp is false positive
(i.e. Diseased Brain incorrectly classified as Normal).
From the confusion matrix above, it is seen that our
Tp is 19, while Fn is 6, and Tn is 19, while Fp is 6.
Hence it is discovered to be
Specificity =
19
19+6
𝑥 100= 76%
Sensitivity =
19
19+6
𝑥 100 = 76%
Accuracy =
19+19
19+19+6+6
𝑥 10 0 = 76%
V. Conclusion and Recommendation
5.1 Conclusion
In this work, a classification scheme between PCA
and SVM were carried out on diseased image of the
brain (Alzheimer) and normal (MRI) brain. The
results show that PCA outperforms the SVM.
5.2 Recommendation
The following recommendations are suggested:
 A mapping feature of principal components
and support vector machine should be
carried out, and results compared.
 The database containing the images should
be increased to as much as 400image
samples, and subsequent results evaluated.
 Other classification schemes such as Bayes,
KNN, could also be evaluated with the
corresponding image disease.
 Other diseases of the brain could also be
analyzed, and compared with normal brain.
 Students of Higher Institution should be
exposed to MatLab, as a core course of
study.
References
[1] Pham D., Xu C., and Prince J. (2000): “A
Survey ofCurrent Methods in Medical
ImageSegmentation,” Annual Review of
BiomedicalEngineering, 2(3): 315-337, 2000.
[2] GladisPushparathi V. P. and Palani S(2012):
“A novel approach for feature extraction
andSelection on MRI images for brain Tumor
classification “,Proceeding CCSEA,
SEA,CLOUD, DKMP, CS & IT-CSCP
2012,New Delhi: 225–234.
[3] Neha J. and Karaulia D.S. (2014): „‟A
Comparative Analysis of Filters on Brain MRI
Images‟‟, International Journal of Advanced
Research in Computer Science and Software
Engineering, 4(11):893-897.
[4] Papageorgiou E.I., and Spyridonos P.P.
(2008):“ BrainTumor characterization using
thesoft computing technique of fuzzy cognitive
maps”Applied soft computing8: 820-828.
[5] Shafaf Ibrahim, Noor Elaiza Abdul Khalid
(2011):” Image Mosaicing for evaluation of
MRI Brain Tissue abnormalities segmentation
study”,Int.J.Biology and Biomedical
Engineering, 4(5): 181-189
[6] Karpagam S., and Gowri S. (2011):”Detection
of tumor growth by advanced diameter
technique using MRI data”,Proc.The world
congress of Engineering, 1WEC 2011,
London, U.K
[7] Matthew C.Clark, Lawrence O.Hall
(1998):”Automatic tumor segmentation
usingKnowledge based techniques”, IEEE
Transactions on Medical Imaging, 17(2).
[8] Carlos A Parra, Khan Iftekharuddin(2003): ”
Automated brain data segmentation and Pattern
recognition using ANN” Proc. CIRAS, 2003

Classification of Abnormalities in Brain MRI Images Using PCA and SVM

Related slideshows

More Related Content

Classification of Abnormalities in Brain MRI Images Using PCA and SVM