International Journal of Recent Technology and Engineering (IJRTE)

ISSN: 2277-3878, Volume-8, Issue-1S4, June 2019

Brain Tumor MRI Segmentation and

Classification Using Ensemble Classifier
Parasuraman Kumar, B. VijayKumar 

Abstract---Brain tumor is a group of tissue that is prearranged brain tumor using image processing techniques. The
by a slow addition of irregular cells. It occurs when cell get detection of a brain tumor at an early stage is a key issue for
abnormal formation within the brain. Recently it is becoming a providing improved treatment. Once a brain tumor is
major cause of death of many people. The seriousness of brain
clinically suspected, radiological evaluation is required to
tumor is very big among all the variety of cancers, so to save a
life immediate detection and proper treatment to be done. determine its location, its size, and impact on the
Detection of these cells is a difficult problem, because of the surrounding areas. On the basis of this information the best
formation of the tumor cells. It is very essential to compare brain therapy, surgery, radiation, or chemotherapy, is decided. It
tumor from the MRI treatment. It is very difficult to have vision is evident that the chances of survival of a tumor-infected
about the abnormal structures of human brain using simple patient can be increased significantly if the tumor is detected
imaging techniques. Ensemble methods have been called the
accurately in its early stage [3]. As a result, the study of
most influential development in Data Mining and Machine
Learning in the past decade. They combine multiple models into brain tumors using imaging modalities has gained
one usually more accurate than the best of its importance in the radiology department. In this paper the
components. Ensemble methods combine the procedure of neural brain tumor identification is done by an image processing.
network, extreme learning machine (ELM) and support vector In this paper, there are four process are done to identify
machine classifiers. The proposed system consists of manifold the brain tumors. The first process is pre processing the
phases. Preprocessing, segmentation, feature extraction, and
image data from the collection of database using median
classification. At initially preprocessing is performed by using
filtering algorithm. Secondly segmentation is performed by using filtering, second stage is segmentation using Fuzzy C-means
clustering algorithm. Thirdly feature extraction is performed by Clustering Algorithm [4], third stage is feature extraction
Gray Level Co-Occurrence Matrix (GLCM). Automatic brain using Gray Level Co- Occurrence Matrix (GLCM), [5] and
tumor stage is performed by using ensemble classification. This the fourth stage is classification using ensemble classifiers is
phase classifies brain images into tumor and non-tumors using the combination of neural network, Extreme Learning
Feed Forwarded Artificial neural network based classifier.
Machine (ELM) and Support Vector Machine classifier
Experiments have exposed that the method was more robust to
initialization, faster and accurate. (SVM). This will be discussed briefly in this following
Keywords-- Ensemble classifiers, GLCM, ELM, SVM, Feed section.
Forward Artificial Neural Network and Fuzzy C-means
Clustering. II. LITERATURE REVIEW
Saleck et al [4] introduced a new approach using FCM
I. INTRODUCTION
algorithm, in order to extract the mass from region-of
In recent times, the introduction of information interested (ROI). The proposed method aims at avoiding
technology and e-health care system in the medical field problematic of the estimation of the cluster number in FCM
helps clinical experts to provide better health care to the by selecting as input data, the set of pixels which are able to
patient. Brain tumors affect the humans badly, because of provide us the information required to perform the mass
the abnormal growth of cells within the brain. It can disrupt segmentation by fixing two clusters only. The Gray Level
proper brain function and be life-threatening. Two types of Occurrence Matrix (GLCM) is used to extract the texture
brain tumors have been identified as benigntumors and features for getting the optimal threshold, which separate
malignanttumors. Benign tumors are less harmful than between selected set and the other sets of the pixels that
malignant tumors as malignant are fast developing and influences on the mass boundary accuracy. The performance
harmful while benign are slow growing and less harmful. of the proposed method is evaluated by specificity,
The various types of medical imaging technologies based on sensitivity and accuracy.
noninvasive approach like; MRI, CT scan, Ultrasound, Bhima and Jagan [6] demonstrated the superior accuracy
SPECT, PET and X-ray [1]. When compared to other for brain tumor detection in compared to the presented
medical imaging techniques, Magnetic Resonance Imaging methodologies. Also the major identified bottleneck of the
(MRI) is majorly used and it provides greater contrast recent research outcomes are limited to detection of brain
images of the brain and cancerous tissues. tumor and the overall analyses of internal structure of the
Therefore, brain tumor identification can be done through brain is mostly ignored being one of the most important
MRI images [2]. This paper focuses on the identification of factor for disorder detection.
Vrji and Jayakumari [7] improved brain tumor
Manuscript received June 10, 2019.
approximation after a manual segmentation procedure and
Parasuraman Kumar, Assistant Professor, Centre for Information
Technology and Engineering, Manonmaniam Sundaranar University, 2D & 3D visualization for surgical planning and assessing
Tirunalveli, Tamil Nadu, India.(e-mail: [email protected]) tumor.
B. VijayKumar, Research Scholar, Manonmaniam Sundaranar
University, Tirunalveli, Tamil Nadu, India.
(e-mail: [email protected])

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 244 & Sciences Publication
 Brain Tumor MRI Segmentation and Classification Using Ensemble Classifier

thetumor identification, the investigations has been made effectively extract the feature information of human viruses
for the potential use of MRI data for improving brain tumor (HV) microscopic images. Firstly, 20 pieces of microscopic
shape approximation. In Preprocessing and Enhancement images of human virus are obtained by using GLCM, and
stage, medical image is converted into standard formatted then the four texture feature parameters, entropy, energy
image. Segmentation subdivides an image into its inertia moment and correlation are extracted utilizing the
constituent regions or objects. GLCM, and then HV image recognition is carried out.
Rashid et al [8] investigated the chosen brain MRI image Parveen and Singh [14] proposed a new hybrid technique
and a method is targeted for more clear view of the location based on the support vector machine (SVM) and fuzzy c-
attacked by tumor. An MRI abnormal brain images as input means for brain tumor classification.
in the introduced method, Anisotropic filtering for noise The purposed algorithm is a combination of support
removal, SVM classifier for segmentation and vector machine (SVM) and fuzzy c-means, a hybrid
morphological operations for separating the affected area technique for prediction of brain tumor. In this algorithm the
from normal one are the key stages if the presented method. image is enhanced using enhancement techniques such as
Attaining clear MRI images of the brain are the base of this contrast improvement, and mid-range stretch. Double
method. The classification of the intensities of the pixels on thresholding and morphological operations are used for
the filtered image identifies the tumor. skull striping. Fuzzy c-means (FCM) clustering is used for
Sudharani et al [9] the present paper proposed the the segmentation of the image to detect the suspicious
c1assification and identification scores of brain tumor by region in brain MRI image.
using k-NN algorithm which is based on training of k. In Grey level run length matrix (GLRLM) is used for
this work Manhattan metric has applied and calculated the extraction of feature from the brain image, after which SVM
distance ofthe c1assifier. The algorithm has been technique is applied to classify the brain MRI images, which
implemented using the Lab View. provide accurate and more effective result for classification
Vidyarthi, A., &Mittal[10] proposed a hybrid model of brain MRI images.
which identifies the region of interest using fused results of AmasyaliandErsoy [15] proposed ensemble classifier in
threshold segmentation and morphological operations. order to improve the accuracy and execution time.
Initially, an abnormal brain MR image is processed with Classification accuracy and execution time are two
Otsu threshold based segmentation and morphological important parameters in the selection of classification
operations like erosion. Further, both the segmented algorithms.
resultant images are fused with the original MR image to In these experiments, 12 different ensemble algorithms,
preserve the background and correctly identification of the and 11 single classifiers are compared according to their
tumor region. accuracies and train/test time over 36 datasets. The results
Li et al [11] proposed framework employs local binary show that Rotation Forest has the highest accuracy.
patterns (LBPs) to extract local image features, such as However, when accuracy and execution time are considered
edges, corners, and spots. Two levels of fusion (i.e., feature- together, Random Forest and Random Committees can be
level fusion and decision-level fusion) are applied to the the best choices.
extracted LBP features along with global Gabor features and
original spectral features, where feature-level fusion III. PROPOSED METHODOLOGY
involves concatenation of multiple features before the The proposed work is mainly focused on the
pattern classification process while decision-level fusion identification of brain tumor to reduce the death rate. The
performs on probability outputs of each individual identification of brain tumor is done by MRI segmentation
classification pipeline and soft-decision fusion rule is and by using Ensemble Classifiers. The proposed
adopted to merge results from the classifier ensemble. methodology consists of four stages.
Moreover, the efficient extreme learning machine with a The first stage is pre-processing by using filtering
very simple structure is employed as the classifier. algorithm, the second stage is Segmentation is done by
Dhanaseely et al [12] presented and investigated two clustering algorithm, third process is feature extraction
different architectures in this work. The cascade architecture which is done by Gray – Level Co-occurrence
(CASNN) and feed forward neural architecture (FFNN) are Matrix(GLCM) [40] and the fourth stage of work is
investigated. The feature extraction is performed using Classification by using Ensemble classifier which is a
principal component analysis (PCA) as it reduces the combiner process of neural network, Extreme Learning
computational burden. Machine (ELM) and Support Vector Machine (SVM) and
For a given database the features are extracted using here an automatic brain tumor stage by ensemble classifier
PCA. The Olivetti Research Lab (ORL) database is by forwarded Artificial Neural Network.
used.The extracted features are divided into training set and Image segmentation is an essential preprocessing trend in
testing set. The training data set is used to train both the a complicated and composite image dealing algorithm in
neural network architectures. Both are tested extensively Brain magnetic resonance imaging (MRI). Segmentation
using testing data. plays a fine role in the medical image segmentation. The
Liu and Liu [13] proposed an algorithm of HV overview of proposed methodology is shown in figure 1.
microscopic image feature extraction and recognition using
gray level co-occurrence matrix (GLCM) in order to

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 245 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1S4, June 2019

Pre – processing
Input brain
Median filtering
tumor
Algorithm
database

Segmentation
FCM Clustering
Algorithm

Feature Extraction
Gray Level Co-
occurrence Matrix

Ensemble classification
SVM Feed Forwarded Artificial ELM
Neural Network

Brain tumor detection

Figure 1:Overview of Proposed Methodology
3.1. Pre-Processing In our proposed work we are going to use median filter
for less computation complexity and better smoothing of
It is very difficult to process an image. Before any image
images. However, it is better in preserving useful detail in
is processed, it is very significant to remove unnecessary
the image than the mean filter. Like the mean filter, the
items it may hold. After removing unnecessary artifacts, the
median filter considers each pixel in the image and replaces
image can be processed successfully. The initial step of
it with the median of the neighborhood pixel values. The
image processing is Image Pre-Processing [16].Pre-
median filter has two main advantages over the mean filter
Processing involves processes like conversion to grayscale
[20]:
image, noise removal and image reconstruction. Conversion
to grey scale image is the most common pre-processing  It is a more robust estimation than the mean. A
single unrepresentative pixel in a neighborhood will
practice [17]. After the image is converted to grayscale, then
not affect the median significantly.
remove excess noise using different filtering methods.
 It does not create new unrealistic pixel values, since
Median FilterThis most common technique which used
the median must actually be the value of one of the
for noise elimination. It is a „non-linear‟ filtering technique. pixels in the neighborhood.
This is used to eliminate „Salt and Pepper noise‟ form the
grayscale image [18]. Median filter is based on average 3.2. Segmentation
value of pixels. The advantages of median filter are efficient Segmentation of images is important as large numbers of
in reducing Salt and Pepper noise and Speckle noise. Also, images are generated during the scan and it is unlikely for
the edges and boundaries are preserved. The main clinical experts to manually divide these images in a
disadvantages are complexity and time consumption as reasonable time. Image segmentation refers to segregation
compared to mean filter. [19] of given image into multiple non-overlapping regions.
Segmentation represents the image into sets of pixels that
are more significant and easier for analysis. It is applied to
approximately locate the boundaries or objects in an image
and the resulting segments collectively cover the complete
image [21]. The segmentation algorithms works on one of
the two basic characteristics of image intensity; similarity
and discontinuity [22].

Figure 2: Median filter

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 246 & Sciences Publication
 Brain Tumor MRI Segmentation and Classification Using Ensemble Classifier

Segmentation has a significant part in clinical diagnosis 1

𝑢𝑖𝑗 = 2
and can be useful in pre-surgical planning and computer 𝑥 𝑖 −𝑐 𝑗 𝑚 −1
𝐶
assisted surgery. 𝑘−1 𝑥 𝑖 −𝑐 𝑘
Therefore, numerous segmentation techniques are
available which can be used widely, such as threshold based
segmentation, histogram based methods, region-based 3. If || U (k+1) - U(k)||< then STOP; otherwise return
(region growing, splitting and merging methods), edge- to step 2.
based and clustering methods (expectation maximization, k-
means, FCM and mean shift) [23]-[25]. 3.3. Feature Extraction
Clustering methods are most promising technique for Feature extraction [28] is an important step in the
processing the medical images. Cluster analysis can be set construction of any pattern classification and aims at the
out as a pre-processing stage for other methods, namely extraction of the relevant information that characterizes each
classifiers that would then run on selected clusters [26]. class. In this process relevant features are extracted from
Therefore in our system, we have used clustering objects/ alphabets to form feature vectors. These feature
segmentation techniques for diagnosis of tumor and vectors are then used by classifiers to recognize the input
calculating tumor area in MRI images. unit with target output unit.
Fuzzy C – means algorithm is used for proposed work It becomes easier for the classifier to classify between
which divide the set of pixels X= {x1, x2, . . .,xN} into C different classes by looking at these features as it allows
fuzzy clusters where each point has a degree of belonging to fairly easy to distinguish. Feature extraction is the process to
clusters. It allows a point to belong to more than one cluster retrieve the most important data from the raw data. In our
as per its membership value. It is an iterative process for work feature extraction is based on GLCM will be briefly
minimizing objective function, related to fuzzy membership discussed in this following section [40].
set U of cluster centers C: Gray-level co-occurrence matrix (GLCM) [5] is the
j= 𝑁 𝐶 𝑚
𝑖=1 𝑗 =𝑖 𝑢𝑖𝑗 (𝑥𝑖 − 𝑐𝑗 )
2 statistical method of examining the textures that considers
(1) the spatial relationship of the pixels. The GLCM functions
characterize the texture of an image by calculating how
Where, uij is the membership table, m is a cluster often pairs of pixel with specific values and in a specified
fuzziness factor and (xi – cj) is Euclidean distance. The data spatial relationship occur in an image, creating a GLCM,
points nearer to center of cluster have highest degree of and then extracting statistical measures from this matrix.
membership than the points on edge [27]. The gray co matrix function in MATLAB creates a gray-
FCM initially guess the cluster centers and assigns every level co-occurrence matrix (GLCM) by calculating how
point a membership grade for each clusters. Then, it moves often a pixel with the intensity (gray-level) value i occurs in
the cluster center to right location by iteratively updating the a specific spatial relationship to a pixel with the value j.
centers within a data set. By default, the spatial relationship is defined as the pixel
The membership defines the fuzziness of an image and of interest and the pixel to its immediate right (horizontally
also defines information contained in an image. Fuzzy adjacent), but you can specify other spatial relationships
partitioning is carried out through an iterative optimization between the two pixels. Each element (i, j) in the resultant
of the objective function shown above, with the update of GLCM is simply the sum of the number of times that the
membership uij and the cluster centers cj by: pixel with value i occurred in the specified spatial
1 relationship to a pixel with value j in the input image.
𝑢𝑖𝑗 = 2 (2)
𝐶
𝑥 𝑖 −𝑐 𝑗 𝑚 −1 Feature Extraction is helpful in identifying brain tumor
𝑘−1 𝑥 𝑖 −𝑐 𝑘 where is exactly located and helps in predicting next stage.
Transforming the input data into the set of features is called
𝑁 𝑢𝑚 𝑥
𝑐𝑗 =
𝑖−1 𝑖𝑗 𝑖
(3) feature extraction [29].
𝑁 𝑢𝑚
𝑖−1 𝑖𝑗 In this paper we‟re extracting some features by using
GLCM [27] and Gabor are: Contrast, Correlation,
(𝑘+1)
This iteration will stop when𝑚𝑎𝑥𝑖𝑗 𝑢𝑖𝑗 − 𝑢𝑖𝑗𝑘 <ε, Homogeneity, Entropy, Energy, Shape, Color, Texture and
where 𝛆 is a termination criterion between 0 and 1, Intensity.
whereas k are the iteration steps. This procedure converges Contrast: Contrast is defined as the separation between
to a local minimum or a saddle point of Jm. the darkest and brightest area
The algorithm is composed of the following steps: Contrast = 𝑛−1
𝑖𝑗 =0 𝑃𝑖𝑗 (𝑖 − 𝑗)
2
(4)

Correlation: Correlation is computed into what is known

1. Initialize U=[uij] matrix, U(0)
as the correlation coefficient, which ranges between -1 and
2. At k-step: calculate the centers vectors C(k)=[cj]
with U(k) +1.

𝑁 𝑚
𝑖−1 𝑢𝑖𝑗 𝑥𝑖
𝑐𝑗 = 𝑁 𝑚
𝑖−1 𝑢𝑖𝑗
Update U(k) , U(k+1)

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 247 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1S4, June 2019

reduce the error through averaging. Ensemble learning is

𝑛−1 (𝑖−𝜇 )(𝑗 −𝜇 ) primarily used to improve classification or prediction
Correlation = 𝑖𝑗 =0 𝑃𝑖𝑗 𝜎2
(5)
performance, where a single model does not have these
Homogeneity: Homogeneity is defined as the quality or capabilities, especially in dealing with multiclass problems.
state of being homogeneous In the work classification consist of three work they are
𝑛−1 𝑃 𝑖𝑗 support vector machine classifier, Extreme learning
Homogeneity = 𝑖𝑗 =0 1+(𝑖−𝑗 )2 (6)
Machine and Feed Forwarded Artificial Neural Network.
This phase classifies brain images into tumor and non-
Entropy: Entropy is a measure of the uncertainty in a tumors using Feed Forwarded Artificial neural network
random variable. based classifier.
Entropy = 𝑁−1
𝑖𝑗 =0 −ln⁡
(𝑃𝑖𝑗 )𝑃𝑖𝑗 (7)
Support Vector Machine (SVM)
Energy: It provides the sum of squared elements in the SVMs are set of related supervised learning methods used
GLCM .Also known as the uniformity or the angular second for classification and regression [37]. They belong to a
moment. family of generalized linear classification.
Energy = 𝑁−1 𝑖𝑗 =0 (𝑃𝑖𝑗 )
2
(8) A special property of SVM is, SVM simultaneously
Shape: The term shape is commonly used to refer to the minimize the empirical classification error and maximize
geometric properties of an object or its external boundary, as the geometric margin. So SVM called Maximum Margin
opposed to other properties such as color, texture, material Classifiers. SVM is based on the Structural risk
composition. Minimization (SRM). SVM map input vector to a higher
Color:Color is a component of light which is separated dimensional space where a maximal separating hyper plane
when it is reflected off of an object. Colors can be identified is constructed.
numerically by their coordinates. Two parallel hyper planes are constructed on each side of
Intensity: Intensity is a purity or strength of color. the hyper plane that separate the data. The separating hyper
Texture: It is the visual characteristic of a surface. For plane is the hyper plane that maximizes the distance
example, a surface can be rough or smooth. between the two parallel hyper planes.
An assumption is made that the larger the margin or
3.4. Classification
distance between these parallel hyper planes the better the
Classification is used to classify each item in a set of data generalization error of the classifier will be [37].We
into one of predefined set of classes or groups. In other consider data points of the form {(x1, y1), (x2, y2), (x3, y3),
words, classification is an important technique used widely (xn, yn)}.
to differentiate normal and tumor brain images. Where yn=1 / -1, a constant denoting the class to which
The data analysis task classification is where a model or that point xn belongs. n = number of sample. Each x n is p-
classifier is constructed to predict categorical labels (the dimensional real vector.
class label attributes). The scaling is important to guard against variable
Classification is a data mining function that assigns items (attributes) with larger variance. To view this Training data,
in a collection to target categories or classes. The goal of by means of the dividing (or separating) hyper plane,this
classification is to accurately predict the target class for each takes
case in the data. In this proposed work the classification is w.x+b=o (9)
done by the combined process of three different algorithms
is known as ensemble classification. Where b is scalar and w is p-dimensional Vector. The
vector w points perpendicular to the separating hyper plane.
Ensemble Classification
Adding the offset parameter b allows us to increase the
Ensemble based methods have recently enjoyed great margin. Absent of b, the hyper plane is forced to pass
attention [30] due to their reported superiority over single through the origin, restricting the solution.
method based system generalization performance [31], [32]. As with the interest in the maximum margin, we are
The aim of classification is to combine multiple models interested in SVM and the parallel hyper planes. Parallel
(classifiers or features) to solve particular problems [33]. hyper planes can be described by equation
Ensemble methods can be divided into a number of w.x + b = 1 (10)
categories, such as ensemble classifiers [34]; ensemble
features [35]; and ensemble feature and classifiers [36]. To w.x + b = -1 If the training data are linearly separable, we
demonstrate the full and practical importance of using a can select these hyper planes so that there are no points
multiple classifier system, an analogy can be made with between them and then try to maximize their distance. By
decision making in everyday life. geometry, we find the distance between the hyper planes is
When making an important decision, an expert is likely to 2 / │w│. So we want to minimize │w│.
ask opinions from several other experts before the final To excite data points, we need to ensure that for all I
decision is made. In such a situation, the final decision is either w. xi – b ≥ 1 or w. xi – b ≤ -1 this can be written as yi
made by combining the individual decisions of several ( w. xi – b) ≥1 , 1 ≤ i ≤ n (11)
experts.
The idea behind all ensemble based systems is that if
individual classifiers or features are diverse, then they can
make different errors, and combining these models can

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 248 & Sciences Publication
 Brain Tumor MRI Segmentation and Classification Using Ensemble Classifier

Feed Forwarded Neural Network

In more practical terms neural networks are nonlinear
statistical data modeling tools. They can be used to model
complex relationships between inputs and outputs or to find
patterns in Data. Using neural networks as a tool, data
warehousing firms are harvesting information from datasets
in the process known as data mining.
The difference between these data warehouses and an
ordinary database is that there is actual manipulation and
cross-fertilization of the data helping users makes more
informed decisions.
Feed forward Neural Network One of the simplest feed
forward neural networks (FFNN) [38], such as in Figure 5,
Figure 3: Maximum margin hyper planes for a SVM consists of three layers: an input layer, hidden layer and
trained with samples from two classes output layer. In each layer there are one or more Processing
SVMs fall into the intersection of two research areas: Elements (PEs). PEs is meant to simulate the neurons in the
kernel methods [39], and large margin classifiers. SVM has brain and this is why they are often referred to as neurons or
been applied to feature selection, time series analysis, nodes.
reconstruction of a chaotic system, and non-linear principal PE receives inputs from either the outside world or the
components. Further advances in these areas are to be previous layer. There are connections between the PEs in
expected in the near future. SVMs and related methods are each layer that have a weight (parameter) associated with
also being increasingly applied to real world data mining. them. This weight is adjusted during training. Information
Extreme Learning Machine only travels in the forward direction through the network -
there are no feedback loops.
Extreme Learning Machine (ELM) is a single hidden
layer feed forward neural network (SLFNN) which
randomly selects input weights and hidden neuron biases
without training. The outputs weights are analytically are
analytically determined using the norm least-square solution
and Moore-Penrose inverse of a general linear system, thus
allowing a significant training time reduction. The activation
function like sine, Gaussian, sigmoid etc., can be chosen for
hidden neuron layer and linear activation functions for the
output neurons .The SLFNN evaluated here uses additive
neuron design instead of kernel based [39], hence random
parameter selection. SLFNs are considered as a linear
system. The aim of the approach is to generate high
resolution images from inputs with low-resolution. In the
training process, the input was extracted from the image
features. Furthermore, the high frequency components that
were taken from the original images with high-resolution
Input Hidden Output
were utilized as the target values. Then, ELM learns a model
that is capable of mapping the interpolated image and Layer Figure 5: FIFOLayer
in Neural Networks Layer
imposing it on the high frequency components. Once
training is done, the learned model can predict the high- IV. RESULT AND DISCUSSION
frequency components using low resolution images [36]. In this section we are going to discuss the result of
various classification techniques. The comparison of
existing classifiers like Feed Forward Artificial Neural
Network (FFANN), Ensemble Learning Machine (ELM)
and Support Vector Machine (SVM) and proposed
Ensemble classifier results are discussed briefly in this
section. Comparison will be takes place with of accuracy,
precision, F1 Score and Sensitivity with the image pixel of
30,000.
A. Accuracy
Table 1 illustrates the comparison of various classifier
techniques with accuracy.
Figure 4: Overview of ELM

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 249 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1S4, June 2019

True Positive  True Negative

Accuracy = PRECISION(%)
Positive  Negative
96
Table1. Comparison between various techniqueswith
accuracy 94

Precision (%)
CLASSIFIER TECHNIQUE ACCURACY
(%) 92
Feed Forward Artificial Neural 84.33 90
Network
Extreme Learning Machine 86.00 88
Support Vector Machine 89.67 86
Ensemble Classifier 91.17
FFNN ELM SVM ENSEMBLE
It is very clear from table 1 and figure 6, the accuracy for
ensemble classifier is 91.17% whereas FFANN, ELM and Classifier Technique
SVM have an accuracy of 84.33%, 86% and 89.67%
respectively with the image pixel of 30,000. The proposed Figure 7: Comparison between various techniques with
Ensemble classifier has a higher accuracy when compared precision
with all other classifiers.
C. Sensitivity
Table 3 illustrates the comparison of various classifier
ACCURACY (%)
techniques with Sensitivity.
92 Sensitivity or True Positive Rate =
True Positive
90
True Positive  False Negative
Accuracy(%)

88
Table 3:Comparison between various techniques with
86 Sensitivity
CLASSIFIER TECHNIQUE SENSITIVITY
84
(%)
82 Feed Forward Artificial Neural 91.94
Network
80
Extreme Learning Machine 93.12
FFNN ELM SVM ENSEMBLE
Support Vector Machine 94.83
Classifier Technique Ensemble Classifier 95.47
It is very clear from table 3 and figure 8, the precision for
Figure 6: Comparison between various techniques with ensemble classifier is 95.47% whereas FFANN, ELM and
accuracy SVM have sensitivity of 91.94%, 93.12% and 94.83%
respectively with the image pixel of 30,000. The proposed
B. Precision Ensemble classifier has a higher sensitivity when compared
Table 2 illustrates the comparison of various classifier with all other classifiers.
techniques with precision.
Precision or Positive Predictive Value=
True Positive SENSITIVITY(%)

False Positive True Negative 96

Sensitivity (%)

95
Table 2.Comparison between various techniques with
94
precision 93
CLASSIFIER TECHNIQUE PRECISION 92
(%) 91
Feed Forward Artificial Neural 89.41 90
Network
Extreme Learning Machine 90.20
Support Vector Machine 92.99
Ensemble Classifier 94.17
It is very clear from table 2 and figure 7, the precision for
Classifier Technique
ensemble classifier is 94.17% whereas FFANN, ELM and
SVM have precision of 89.41%, 90.20% and 92.99%
Figure 8: Comparison between various techniques with
respectively with the image pixel of 30,000. The proposed
Sensitivity
Ensemble classifier has a higher precision when compared
with all other classifiers.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 250 & Sciences Publication
 Brain Tumor MRI Segmentation and Classification Using Ensemble Classifier

D. F1 Score results are shown in this paper and concluded that the
Table 4 illustrates the comparison of various classifier proposed ensemble classifier is best with various aspects.
techniques with F1 Score.
F1 Score = REFERENCES
2 True Positive 1. Borole, V. Y., Nimbhore, S. S., &Kawthekar, D. S. S.
(2015). Image Processing Techniques for Brain Tumor
2 True Positive  False Positive  False Negative Detection: A Review. International Journal of Emerging
Trends & Technology in Computer Science
Table 4.Comparison between various techniques with F1 (IJETTCS), 4(5), 2.
Score 2. Bahadure, N. B., Ray, A. K., &Thethi, H. P. (2017).
Image analysis for MRI based brain tumor detection and
CLASSIFIER TECHNIQUE F1 SCORE (%) feature extraction using biologically inspired BWT and
Feed Forward Artificial Neural Network 90.66 SVM. International journal of biomedical imaging, 2017
Extreme Learning Machine 91.63 3. Coatrieux, G., Huang, H., Shu, H., Luo, L., & Roux, C.
Support Vector Machine 93.90 (2013). A watermarking-based medical image integrity
Ensemble Classifier 94.81 control system and an image moment signature for
It is very clear from table 4 and figure 9, the F1 score for tampering characterization. IEEE journal of biomedical
and health informatics, 17(6), 1057-1067.
ensemble classifier is 94.81% whereas FFANN, ELM and
4. Saleck, M. M., ElMoutaouakkil, A., &Mouçouf, M.
SVM have F1 score of 90.99%, 91.63% and 93.90% (2017, May). Tumor Detection in Mammography Images
respectively with the image pixel of 30,000. The proposed Using Fuzzy C-means and GLCM Texture Features.
Ensemble classifier has an higher F1 score when compared In Computer Graphics, Imaging and Visualization, 2017
with all other classifiers. 14th International Conference on (pp. 122-125). IEEE.
5. Zulpe, N., &Pawar, V. (2012). GLCM textural features
SENSITIVITY(%) for brain tumor classification. International Journal of
Computer Science Issues (IJCSI), 9(3), 354.
96 6. Bhima, K., &Jagan, A. (2016, April). Analysis of MRI
95 based brain tumor identification using segmentation
Sensitivity (%)

94 technique. In Communication and Signal Processing

93 (ICCSP), 2016 International Conference on (pp. 2109-
92 2113). IEEE.
91 7. Vrji, K. A., &Jayakumari, J. (2011, July). Automatic
90 detection of brain tumor based on magnetic resonance
image using CAD System with watershed segmentation.
89
In Signal Processing, Communication, Computing and
88 Networking Technologies (ICSCCN), 2011 International
FFNN ELM SVM ENSEMBLE Conference on (pp. 145-150). IEEE.
8. Rashid, M. H. O., Mamun, M. A., Hossain, M. A.,
Classifier Technique &Uddin, M. P. (2018, February). Brain Tumor Detection
Using Anisotropic Filtering, SVM Classifier and
Figure 9: Comparison between various techniques with Morphological Operation from MR Images. In 2018
F1 Score International Conference on Computer, Communication,
We have a very detailed discussion in this section on Chemical, Material and Electronic Engineering
(IC4ME2) (pp. 1-4). IEEE.
various classifier techniques and came to know that our
9. Sudharani, K., Sarma, T. C., &Rasad, K. S. (2015,
proposed Ensemble classifier has best Accuracy, Precision, December). Intelligent Brain Tumor lesion classification
Sensitivity and F1 Score as compared with other classifier and identification from MRI images using k-NN
technique like FFANN, ELM and SVM. technique. In Control, Instrumentation, Communication
and Computational Technologies (ICCICCT), 2015
V. CONCLUSION International Conference on (pp. 777-780). IEEE.
10. Vidyarthi, A., & Mittal, N. (2013, December). A hybrid
In this work, brain tumor is identified by an image model for extraction of brain tumor in MR images.
processing techniques. This identification required various In Signal Processing and Communication (ICSC), 2013
processes like pre-processing using filter algorithm, International Conference on (pp. 202-206). IEEE.
segmentation using clustering algorithm, feature extraction 11. Li, W., Chen, C., Su, H., & Du, Q. (2015). Local Binary
Patterns and Extreme Learning Machine for
using Gray level co-occurrence matrix and ensemble
Hyperspectral Imagery Classification. IEEE Trans.
classification. Geoscience and Remote Sensing, 53(7), 3681-3693.
The ensemble classifier is finally classified the tumor and 12. Dhanaseely, A. J., Himavathi, S., &Srinivasan, E.
non-tumor region. Ensemble classifier plays an important (2012). Performance comparison of cascade and feed
role in our work. Ensemble classifier is the combination of forward neural network for face recognition system.
different individual or separate classifiers. In our work the 13. Liu, Q., & Liu, X. (2013, December). Feature Extraction
of Human Viruses Microscopic Images Using Gray
ensemble classifier is made up of combined classifiers of
Level Co-occurrence Matrix. In Computer Sciences and
feed forward artificial neural network, extreme learning Applications (CSA), 2013 International Conference
machine and support vector machine classifier. Here the on (pp. 619-622). IEEE.
ensemble have an high accuracy and less execution time and
it is very efficient when compared to all other classifier
techniques as we discussed in this paper. The comparison

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 251 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-1S4, June 2019

14. Singh, A. (2015, February). Detection of brain tumor in 31. Hsieh, S. L., Hsieh, S. H., Cheng, P. H., Chen, C. H.,
MRI images, using combination of fuzzy c-means and Hsu, K. P., Lee, I. S., ... & Lai, F. (2012). Design
SVM. In Signal Processing and Integrated Networks ensemble machine learning model for breast cancer
(SPIN), 2015 2ndInternationalConferenceon (pp. 98- diagnosis. Journal of medical systems, 36(5), 2841-2847.
102). IEEE. 32. Bouziane, H., Messabih, B., &Chouarfia, A. (2011).
15. Amasyali, M. F., &Ersoy, O. K. (2011, June). Profiles and majority voting-based ensemble method for
Comparison of single and ensemble classifiers in terms protein secondary structure prediction. Evolutionary
of accuracy and execution time. In Innovations in Bioinformatics, 7, EBO-S7931.
Intelligent Systems and Applications (INISTA), 2011 33. Dietterich, T. G. (2000). An experimental comparison of
International Symposium on (pp. 470-474). IEEE. three methods for constructing ensembles of decision
16. Hebli, A. P., & Gupta, S. (2016). Brain Tumor Detection trees: Bagging, boosting, and randomization. Machine
Using Image processing: A Survey. In Proceedings of learning, 40(2), 139-157.
65th IRF International Conference, 20th November. 34. Rokach, L. (2010). Ensemble-based classifiers. Artificial
17. Kapoor, L., & Thakur, S. (2017, January). A survey on Intelligence Review, 33(1-2), 1-39.
brain tumor detection using image processing techniques. 35. Jong, K., Mary, J., Cornuéjols, A., Marchiori, E.,
In Cloud Computing, Data Science & Engineering- &Sebag, M. (2004, September). Ensemble feature
Confluence, 2017 7th International Conference on (pp. ranking. In European Conference on Principles of Data
582-585). IEEE. Mining and Knowledge Discovery (pp. 267-278).
18. Rajesh C. patil, A.S. Bhalchandra, “Brain tumor Springer, Berlin, Heidelberg.
extraction from MRI images Using MAT Lab”, 36. Albadra, M. A. A., &Tiuna, S. (2017). Extreme Learning
IJECSCSE, ISSN: 2277-9477, Volume 2, issue1. Machine: A Review. International Journal of Applied
19. Gopal,N.N. Karnan, M. , ―Diagnose brain tumor Engineering Research, 12(14), 4610-4623.
through MRI using image processing clustering 37. Vapnik. The Nature of Statistical Learning Theory. NY:
algorithms such as Fuzzy C Means along with intelligent Springer- Verlag. 1995.
optimization techniques Page(s): 1 – 4, Computational 38. Sazli, M. H. (2006). A brief review of feed-forward
Intelligence and Computing Research (ICCIC), 2010 neural networks. Communications, Faculty of Science,
IEEE International Conference, 28-29 Dec. 2010. University of Ankara, 50(1), 11-17
20. Srivastava, H. B., Kumar, V., Verma, H. K., &Sundaram, 39. P. Kumar and B. Vijayakumar (2015). Brain Tumour Mr
S. S. (2009). Image Pre-processing Algorithms for Image Segmentation and Classification Using by PCA
Detection of Small/Point Airborne Targets. Defence and RBF Kernel Based Support Vector Machine.
Science Journal, 59(2). Middle-East Journal of Scientific Research, Vol. 23, No.
21. Janani V, Meena P, “Image segmentation for tumor 9, pp. 2106-2116.
detection using fuzzy inference system”, International 40. P. Kumar and B. Vijayakumar (2016). An Efficient Brain
Journal of Computer Science and Mobile Computing, Tumor MRI Segmentation and Classification Using
Vol. 2(5), 2013, pp. 244-248. GLCM Texture Features and Feed Forward Neural
22. Acharya J, Gadhiya S and Raviya, “Segmentation Networks. World Journal of Medical Sciences, Vol. 13,
techniques for image analysis: a review”, International No. 2, Pp. 85-92.
Journal of Computer Science and Management Research,
Vol. 2(4), 2013, pp. 1218-1221.
23. Naik D, Shah P, “A review on image segmentation
clustering algorithms”, International Journal of Computer
science and Information Technologies, Vol. 5(3), 2014,
pp. 3289-3293.
24. Christ SA, Malathy K, Kandaswamy A, “Improved
hybrid segmentation of brain MRI tissue and tumor using
statistical feature”, ICTACT Journal on Image and video
processing, Vol. 1(1), 2010, pp. 34-49.
25. Seerha GK, Kaur R, “Review on recent image
segmentation techniques”, International Journal of
Computer Science Engineering, Vol. 5(2), 2013, pp. 109-
112.
26. Panda M, Patra MR, “Some clustering algorithms to
enhance the performance of the network intrusion
detection system”, Journal of Theoretical and Applied
Information Technology, Vol. 4(8), 2008, pp.795-801.
27. Sachdeva.J, Kumar.V, Gupta.I, Khandelwal.N, Ahuj.C.K
, “Multiclass Brain Tumor Classification using GA-
SVM” 978-0-7695-4593-6/11 $26.00 © 2011 IEEE DOI
10.1109/DeSE.2011.31
28. Kumar, G., & Bhatia, P. K. (2014, February). A detailed
review of feature extraction in image processing systems.
In Advanced Computing & Communication
Technologies (ACCT), 2014 Fourth International
Conference on (pp. 5-12). IEEE.
29. Subashini. M.M, Sahoo. S.K, “Brain MR Image
Segmentation for Tumour Detection using Artificial
Neural Networks” ISSN: 0975-4024 Vol 5 No 2 Apr-
May 2013.
30. Bramer.M. Ensemble classification. Principles of Data
Mining, pages 209–220, 2013.

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: A10440681S419/19©BEIESP 252 & Sciences Publication