A Survey Paper on Magnetic
Resonance Image (MRI) Classifier
Department of ECE
Mepco Schlenk Engineering College, Sivakasi.
Abstract - Magnetic resonance imaging (MRI) is a
technique boon to the medical field, which is used in radiology
to diagnosis the anatomy and physiological processes of the
body. It can be used in detecting brain tumour, Liver Cancer,
spinal column and also in blood vessels malfunctioning. Mostly
MRI image is used compared to compute topography (CT) scan
image because MRI give clear differences between normal and
abnormal tissue and it also has better contrast of image. It
provides High Resolution Image.
It is very important to provide a clear Image of
diseased organ so only physician can correctly diagnose and
provide an appropriate treatment to the patient. The recent
technique used for MR image classification is explained.
Many methods use a supervised classification framework.
Supervised classification includes a) training images with
known labels are aligned to a common space, b) distinguished
feature extraction, c) extracted features calculated from
regions of interest (ROIs). These ROIs are either predefined
or learned from the data.
The classifier-training algorithm uses pre-classified examples
to determine the set of parameters required for proper
discrimination.The algorithm then encodes these parameters
into a model called a classifier.
Image Classifier can be divided into three types are as
A. Supervised Classification Framework
This Framework can be established by using concepts of
1) Ensemble Learning
It involves training multiple classifiers on different
feature sets. It gives better performance than individual
2) Feature Ranking
It mainly focuses on ordering features based on their
relevance to classification.In supervised classifier has higher
accuracy than unsupervised classifier.
B. Semi- Supervised Classification Framework
It is the combination of both supervised and unsupervised
C. Unsupervised Classification Framework
Unsupervised techniques are self organizing method.
III. TYPES OF CLASSIFIER
There are different types of classifier are discussed in this
section as follows.
A. Neural Network Classifier
Neural network is a set of connected input/output units
and connection has its unique weight. This networks have the
remarkable ability to derive meaning from complicated or
blurred image and can be used to extract original information
and to detect image that are too complex to be noticed by
either physician or other computer techniques. These are well
suited for continuous valued inputs and outputs the learning
phase, network learns by adjusting weights so as to be able to
predict the correct information. The neural network is trained
for either continuous or discrete valued features to analyze the
efficiency of the network for varying inputs to identify the
tumor detection In Medical Resonance images noise gets
added due to operator performance. This may further leads to
inaccuracies categorization which is very severe. Artificial
intelligent methods with neural networks have revealed
excellent potential for experimentation .Neural Networks
generalize better on unseen data and is easier to train and
doesn’t have any local optima.
Neural network training algorithm is judged by a set of
conflicting requirements such as simplicity, flexibility, and
B. Support Vector Machine (SVM) Classifier
Support vector machine (SVM) is widely used classifier
in Medical Image and it gives more accuracy comparable to
neural-network. It also used in object detection & recognition,
content-based image retrieval, text recognition, biometrics,
1) Support Vector Machine (SVM) Algorithm
Step 1: Choose a kernel function
Step2.: Choose a value for C
Step3: Solve the quadratic programming problem (many
software packages available)
Step4:Construct the discriminate function from the
support vectors .
2) Approaches of SVM Classifier
a) Nonlinear Separable Case
SVM model using a sigmoid kernel function is
equivalent to a two-layer, feed-forward neural network.
b) Linear Separable Case
In most of the real-world applications of SVM we
combine the kernel trick and the soft margin and use them
c) Kernal Trick
A kernal function is defined as a function that
corresponds to a dot product of two feature vectors in some
expanded feature space
SVMs classifier techniques suitable for binary
classification takes as input labeled data from two classes and
outputs a model file for classifying new unlabeled/labeled
data into one of two classes. It is proved that this kennel
technique helps to get more accurate result.
do not have to be defined. Do not use abbreviations in the title
unless they are unavoidable.
SVM have high approximation capability and much
SVM does not provide accurate results for a large
data due to the training complexity of SVM.
It is highly dependent on the size of data.
C. Hybrid Classifier
The term hybrid refers to combination of any two classifier
which gives better efficiency than the individual classifier
Adaptive neuro-fuzzy inference systems (ANFIS technology)
it includes the advantages of both the ANN and the fuzzy
logic systems the classification of different brain images using
hybrid classifier .The hybrid systems have wider scope, dual
advantages this technique increases accuracy.
It does not require prior human expertise.
It uses membership functions and desired dataset to
It provides greater choice of membership functions.
D. Naïve Bayes Classifier
All naive Bayes classifiers has the value of a particular
feature is independent of the value of any other feature, given
the class variable.For example, in a brain if tumor is detected
a normal it may be round, and about15 cm in diameter
depends upon its above features. A naive Bayes classifier
considers each of these features to contribute independently to
the probability that this tumour is an, regardless of any
possible correlation between the color, roundness, and
Naive Bayes requires a small number of training data to
estimate the parameters which is necessary for classification
E. k-means Clustering
k–Means clustering classifier was proposed to segregate
different intensity pixels in group of pixels having similar
pattern or characteristics.
The basic K-Means clustering algorithm is given below:
Step 1: K cluster centers are selected based on some
heuristic or random.
Step 2: Assigning pixels to those cluster center having
minimum distance from that center based on
Step 3: Recomposing the cluster centers by averaging
the pixels in the clusters.
Step 4: Repeat the last two steps till no more new
center is formed. The segregation of pixels is
done on the basis of intensity, size colour,
texture, location etc.
K-MEANS clustering algorithm is simple
It has relatively low computational complexity
Sometimes it fails to give accurate results
F. K- Nearest Neignbour(KNN)
KNN is a non parametric classifier. This classifier
provides output in the form of class membership and the
object is related to the class with maximum resemblance. In
the training stage of the KNN classifier the input dataset is
divided into k classes, with each class containing only the
inputs belonging to its class. This is done by finding the
spearman distance between all inputs and distributing them
into k classes.Different distance measures can be adopted as
the reference measure for classification.
1) KNN Algorithm
Step 1: Training set includes classes.
Step 2: Examine K items near item to be classified.
Step 3: New item placed in class with the most number of
Step 4: O (q) for each tuple to be classified (Here q is the
size of the training set.).
The K-nearest neighbour classification is performed by
using a training data set which contains both the input and
the target variables and then by comparing the test data which
contain only the input variables to that reference set the
distance of the unknown to K nearest neighbours determines
its class assignment by either averaging the class numbers of
the K nearest neighbour points
It is simplest technique conceptually
It provide better accuracy
G. Radial Basis Function (RBF)
RBF network is trained by supervised manner. The
generalization property of RBF is to train a network on a set
of input vectors and get the result without even training the
network on all the possible input and output pairs. The Radial
Basis Function (RBF) network and Back Propagation network
performs similar function mapping. BP network is the global
network where RBF network is the local network.
The Radial Basis Function (RBF) is used for adjusting
the weights and minimizes the errors and gives the
H. Artifical Neural Network (ANN)
Artificial neural networks have Proven themselves as
proficient classifiers and are particularly well suited for tumor
classification.ANN models widely used in early detection of
cancer. An ANN is a computational model consisting of a
number of highly interconnected processing elements called
These neurons are organized in layers. The neurons are
connected with each other through links (connections). Each
link is assigned a weight. A neuron also has an associated
bias. The output of a neuron is the output of the activation
function; argument of this function is the sum of incoming
signals multiplied with respective weights, plus the bias.
Activation function is usually sigmoid. The geometry and
functionality of neural network resemble the human brain.
The basic form of ANN is the Multilayer Perceptron (MLP)
which is a neural network that updates the weights through
back propagation. The neural network has been trained to
adjust the connection weights and biases in order to produce
the desired mapping.
It accurately segments tumorous and normal brain
It is reliable, fast, automatic and robust diagnosis
a. Neuro Fuzzy Classifier
Neuro-fuzzy classifier is used to find whether the input
image is tumor image or not. The Neuro fuzzy and the Feed
Forward Neural Network (FFNN) and the obtained results are
analyzed in terms of sensitivety and accuracy.Due to the
fuzzy and deterministic characteristics of MRI brain image,
various tissues are complex and heterogeneous.
Fuzzy C-means algorithm (FCMA) is a nonlinear
iteration optimization method based on objective functions
I. Decision Tree
This classifier divides datasets into smaller and more
uniform datasets. The subsets are broken down into further
subsets via different variables and attributes. The data is
partitioned based on maximum reduction in deviance over all
splits of all nodes to choose the next split
It requires very little computational load.
does not require extensive network training.
It is best suited for non-parametric data
This survey gives an information regarding MRI Image
Classifier which is used in medical field. Various Classifiers
are discussed in the above survey sections. Each classifier has
its own merits as well as demerits .Depending upon the
application, suitable classifier is used to get the better results
in diagnosis purpose.
 Amer AI-Badarneh, Hassan Najadat, and Ali M AIraziqi. \"A
classifier to detect tumor disease in MRI brain images\". In: Advances in
Social Networks Analysis and Mining (ASONAM), 2012 IEEEIACM
1nternational Conference on. IEEE. 2012, pp. 784787.
 Ketan Machhale et al. \"MRI brain cancer classification using hybrid
classifier (SVM-KNN)\". In: Industrial Instrumentation and Control (1CIC),
2015 1nternational Conference on. IEEE. 2015, pp. 6065.
 Yudong Zhang et al. \"A hybrid method for MRI brain image classification\".
In: Expert Systems with Applications 38.8 (2011), pp. 10049-10053
 .Jayachandran .A & R.Dhanasekaran ,”Brain tumor Detection and
Classification of MRI Using Texture Feature and Fuzzy SVM Classifiers”,
Research Journal of Applied Sciences, Engg and Tech 6(12):2264-2269,
 C. Ramalakshmi, A.Jaya Chandran,”Automatic Brain Tumor Detection in
MR Images Using Neural Network Based Classification”, int Journal of
science research and management (IJSRM) volume Issue 2IIpages 109113II2013
 Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo
M, Parkin DM, Forman D, Bray, F. GLOBOCAN v1.1, 2012
 Matei Mancas, Bernard Gosselin, and Benoît Macq, “Tumor Detection using
Airways Asymmetry”, Engineering in Medicine and Biology Society, 2005.
IEEEEMBS 2005, 27th Annual International Conference, 2005
 V. Ulagamuthalvi, D. Sridharan; “Automatic Identification of Ultrasound
Liver Cancer Tumor Using Support Vector Machine”; International
Conference on Emerging Trends in Computer and Electronics Engineering,
 Vanitha. L. and Venmathi. A.R, “Classification of Medical Images Using
Support Vector Machine”, International Conference on Information and
Network Technology IACSIT Press, Singapore,vol.4,pp 63-67,2011
 S. N. Sulaiman and N. A. Mat Isa, \"Adaptive fuzzy-K-means clustering
algorithm for image segmentation,\" in IEEE Transactions on Electronics,
vol. 56, no. 4, pp. 2661-2668, November 2010.
...(download the rest of the essay above)