Automated Brain Tumor Detection Based on Feature Extraction from The MRI Brain Image Analysis

Abstract

The brain tumors are among the common deadly illness that requires early, reliable detection techniques, current identification, and imaging methods that depend on the decisions of neuro-specialists and radiologists who can make possible human error. This takes time to manually identify a brain tumor. This work aims to design an intelligent model capable of diagnosing and predicting the severity of magnetic resonance imaging (MRI) brain tumors to make an accurate decision. The main contribution is achieved by adopting a new multiclass classifier approach based on a collected real database with new proposed features that reflect the precise situation of the disease. In this work, two artificial neural networks (ANNs) methods namely, Feed Forward Back Propagation neural network (FFBPNN) and support vector machine (SVM), used to expectations the level of brain tumors. The results show that the prediction result by the (FFBPN) network will be better than the other method in time record to reach an automatic classification with classification accuracy was 97% for 3-class which is considered excellent accuracy. The software simulation and results of this work have been implemented via MATLAB (R2012b).

Full text

Received: 1 October 2020
Revised: 14 October 2020
Accepted: 16 October 2020
DOI: 10.37917/ijeee.16.2.6
Early Version | December 2020
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and
reproduction in any medium, provided the original work is properly cited.
© 2020 The Authors. Iraqi Journal for Electrical and Electronic Engineering by College of Engineering, University of Basrah.
https://doi.org/10.37917/ijeee.16.2.6 https://www.ijeee.edu.iq 1
Iraqi Journal for Electrical and Electronic Engineering
Original Article
 Open Access
Automated Brain Tumor Detection Based on Feature
Extraction from The MRI Brain Image Analysis
Ban Mohammed Abd Alreda*, Hussain Kareem Khalif, Thamir Rashed Saeid
Electrical Engineering Department, University of Technology, Iraq
*Ban Mohammed Abd Alreda
Electrical Engineering Department,
University of Technology, Iraq
Email: 31630@student.uotechnology.edu.iq
Abstract
The brain tumors are among the common deadly illness that requires early, reliable detection techniques, current identification,
and imaging methods that depend on the decisions of neuro-specialists and radiologists who can make possible human error.
This takes time to manually identify a brain tumor. This work aims to design an intelligent model capable of diagnosing and
predicting the severity of magnetic resonance imaging (MRI) brain tumors to make an accurate decision. The main contribution
is achieved by adopting a new multiclass classifier approach based on a collected real database with new proposed features that
reflect the precise situation of the disease. In this work, two artificial neural networks (ANNs) methods namely, Feed Forward
Back Propagation neural network (FFBPNN) and support vector machine (SVM), used to expectations the level of brain tumors.
The results show that the prediction result by the (FFBPN) network will be better than the other method in time record to reach
an automatic classification with classification accuracy was 97% for 3-class which is considered excellent accuracy. The
software simulation and results of this work have been implemented via MATLAB (R2012b).
KEYWORDS: Feed forward back propagation neural network, Support vector machine, Discrete wavelet transform,
Texture features by gray level co-occurrence matrix (GLCM).
I. INTRODUCTION
Research in the domain of biomedical image analysis has
been one of the most challenging and promising areas in
recent years due to an increase in brain tumors. Sadly, there
are many of these tumors are discovered late time when
symptoms of the disease are present and when the tumor has
become large, making it very difficult to treat or remove the
tumor and dangerous. While it is much safer and easier for
the removal of a limited tumor at an early stage.
Approximately 60 percent of glioblastomas (known as
glioblastoma multiform GBM) begin as lower-grade tumors
but become large tumors over time [1]. Computer-assisted
surgical preparation and modern image-guided systems have
been widely used in neuron surgery [1]. Therefore, when the
development of medical devices in general and medical
imaging devices in particular, many types and different
methods of medical imaging have appeared recently. So
there are many types of imaging are used in brain tumors
diagnoses such as magnetic resonance imaging (MRI),
radiography (like CT and x-ray), and ultrasound imaging [2].
Therefore, the diagnosis of the tumors in the brain are based
on image analysis of the MRI, and it is a saver method and
doesn't affect the human body because it doesn't use any
radiation anymore. It’s based on the magnetic area and radio
waves [2]. The primary reason for death in kids and adults is
the brain tumor [3,4]. The survival chances may be higher if
the tumors are correctly detected at an early stage [3,4].
II. LITERATURE SURVEY
The need for an automated and well-organized brain tumor
MR Image Classification and Diagnosis system has
increased with accurate results for proper treatment
directions (therapy and surgery planning). To this purpose, a
great several research has been submitted by different
researchers with good results but some researchers did not
use real data with data taken from the Internet, and no more
features of the brain images were extracted that would
therefore be used as inputs for the classifiers used. This
section will conclude with a summary discussion of
previous.
Naveena H. S. et al. [5] Utilize ANN ability to classify MRI
images as either cancerous or non-cancer tumors.
Segmentation is achieved by the K-means clustering
algorithm. The Gray Level Co-Occurrence Matrix (GLCM)
was then used to extracted features from the segmented
image. Finally, the classification of brain tumors was
performed by the Back Propagation Neural Network
(BPNN) and the Probabilistic Neural Network (PNN) with a
system accuracy of 79.02 % and 97.25 % respectively.
Ahmad Chaddad and Camel Tanougast [6] The proposed
algorithm uses a contrast enhancement image in

2 | Abd Alreda, Khalif & Saeid
preprocessing and segmentation using the proposed adaptive
skull stripping algorithm and detection tumor based entirely
on multi-threshold segmentation (MTS) to identify
characteristic pixel proximity by MTS algorithm threshold
detects.
Bahadure et. al. [7] Proposed BWT and SVM imaging
techniques for MRI-based brain tumor detection and
classification. In this method, 95 % accuracy was achieved
by SVM and segmentation by skull stripping, which
eliminated all non-brain tissues for detection purposes.
N. Varuna Shree and T. N. R. Kumar [8] region-based
segmentation images feature extraction 7 features from
GLCM. MRI images of 256×256, 512 × 512-pixel size on
dataset. dataset collected from Web sites www.diacom.com.
The accuracy of training 100% dataset because the statistical
textural features were extracted from LL and HL sub bands
wavelet decomposition and 95% of testing dataset.
Parasuraman Kumar and B. Vijay Kumar [9] FCM
Clustering Algorithm based segmentation and feature
extraction by using GLCM and Gabor are 9 features.
Preprocessing achieved by median filter. Ensemble methods
combine the procedure of feed forward neural network,
extreme learning machine (ELM) and support vector
machine classifiers to detected tumor. The accuracy for
ensemble classifier is 91.17% whereas FFANN, ELM and
SVM have an accuracy of 84.33%, 86% and 89.67%
respectively.
Ahsanullah Umary and Harpreet Kaur [10] The
suggested approach includes eliminating unnecessary noise
using filtering techniques, segmenting by using the threshold
to remove the skull from the image, utilizing GLCM to
extraction characteristics and classification by applying
SVM to tumor and non-tumor image. This proposed
approach was approximately 90 % accurate.
III. PROPOSED METHODOLOGY
The work proposed, contribute with the help of Non-
specialists to the understanding of the diagnostic mechanism
and will be contribute diagnosis decision whether the tumor
is present or absent along with the exact area of the tumor.
According to the proposed work stages as shown in Fig. 1.
3.1 Data acquisition stage
Harvard medical brain database most widely available on the
internet used by many researchers, will be used in this work
to test proposed algorithm in the classify the tumor of the
brain using actual data which are 83 cases for training and 37
cases for testing. Also, data from the cancer imaging archive
(TCIA) 59 casesfor training and 29 cases for testing, as well
through fieldwork to collecte 66 cases for training and 34
cases for testing from General Hospital of Baquba-Iraq of
MRI brain images. So, it becomes some of the data based in
this work 308 images using T1 weighted and T2-weighted
images. The MRI brain images data that taken from available
Harvard medical brain databases were available in the form
of images and were directly used for the following pre-
processing stage. While, the MRI brain images data taken
from hospitals was in the form of CD where was MRI brain
images in DCM format (real MRI brain images data) [11,12],
and by using an online website for image format conversion
and saved in the form of digital color images at format types
JPEG in the computer [13].
Fig. 1: Flowchart of the proposed methodology.
3.2 Pre-processing stage
The main task of pre-processing is to enhance the precision
of the MR images and also to get it suitable for further
processing by a human or machine vision system. The
grayscale image is quite important for many purposes, such
as segmentation of images, extraction of features and
classification of images using the (rgb2gray) function of the
MATLAB software used to convert the RGB image to the
Grayscale images [14]. The MATLAB (adjust) function has
been used to increase the Contrast some image by assigning
the values of the input intensity of the image to new values
such that the information is saturated at low and high input
data level [15].
Sometimes MRI images contain noise that must be discarded
and removed. The causes of this noise are due to the high
frequency of radio waves and the patient's movement during
the MRI [15]. But the noise removal should not destroy the
edges of the image and reduce the clarity and quality of the
image. Anisotropic filtering used anisotropic diffusion filter
is an image filtering method proposed by Persona and Malik
[16,17]. The anisotropic diffusion filter is the spearheading
work in partial derivatives equations (PDE) based de-noising
[17,18]. The algorithm is implemented on the MATLAB
DATA ACQUISITION
STAGE
TUMOR
CLASSIFICATIONS
STAGE
 Experimental data set by using
MRI images
 Texture Features by GLCM
 DWT Features
 Area Tumor
PRE-
PROCESSING
STAGE
 Image conversion
 Gray scale
 Image Enhancement. (Optional (
 Noise filtering by Anisotropic
Filter.
 Thresholding Segmentation.
 Morphological operation
 Tumor Outline
IMAGE
SEGMENTATION
STAGE
FEATURE
EXTRACTION
STAGE
 Artificial Neural Network
(FFBPNN)
 Support Vector Machine (SVM)
Classifier

Abd Alreda, Khalif & Saeid | 3
software used function (anisodiff.m) [19]. The remaining
steps of the methods are explained below.
The Algorithm
1) Load gray MRI images into the MATLAB
environment.
2) Convert into Double precision.
3) Add the random noise to the image.
4) Apply to open by reconstruction
5) Give the number of iterations (i=N)
6) i=1
7) Enter the following value
•Constant integration. (Usually, this parameter is set to
its maximum value due to numerical stability).
• The gradient modulus threshold that controls the
conductivity.
• Functions of the conduction coefficient (proposed by
Perona & Malik) Apply anisotropic diffusion filtering
process.
8) if (i< N) i=i+1 go to step no:7.
9) Display gray input image and filtered output.
3.3 Image segmentation stage
The segmentation technique plays a significant role in the
processing of images. Segmentation results will be used to
obtain quantitative information from images, including
clustering, thresholding, etc.
3.3.1 Threshold segmentation
Threshold segmentation technology Plays an essential part in
the processing of images. The threshold segmentation is used
to extract the various regions from the whole image
according to the difference in intensity. There are many
different methods to achieve thresholding methods, such as
the Otsu method [20]. Otsu's method is one of the most
effective methods used for image thresholding. It is based on
the criterion of minimizing class variance; however, the
search for the global optimal threshold is a fully
comprehensive algorithm. At first, the Otsu method needs to
calculate a gray-level histogram [21].




 Weights are The probability of the two classes divided by
the thresholds t and 2 variances of these classes.




Which is described in terms of class probabilities  and
class means. The class probability is calculated
from the histogram as t:



Whereas the class mean is:




In which () is the value at the center of the bin histogram.
Fig. 2 show steps of proposed algorithm in the work.
Fig. 2: Flowchart of proposed algorithm
3.3.2 Morphological operation
Typically, applied on the binary images (black & white
images) where the pixels value is between 0 and 1. This work
is used in post-processing to enhance the threshold
segmentation by removing noise and to filter out smaller
areas. The first step is very important to remove undesired
pixels as noise by filling the holes that can be described as a
distortion in the image. Where the small holes fill the white
pixel in the dark background by dark pixel and the holes dark
in the white region will convert to a white pixel. Filling
region algorithm based on set dilations, complements, and
intersections. In the second step, erosion operations are
intended to eliminate pixels from the boundary area of the
objects. The operation of adding or deleting pixels to or from
the boundary area of the object is focused on the structuring
elements of the selected image.

4 | Abd Alreda, Khalif & Saeid
3.3.3 Tumor Outline
Tumor outline is an additional step used to determine the
shape of the tumor and external limits. Image pixel
subtraction operators take two images as an input of the same
size and produce as output a third image; whose pixels values
are the values obtained by subtraction between the two
images. The pixel subtraction operation can be defined as
[22,23]:

Where  represented the result of subtracting the
two images, represented the result of the process
of segmentation of the image to BW by the threshold
segmentation algorith represented the resulting
image of the erosion process as show in Fig. 3.
Fig. 3: flowchart of tumor outline algorithm
3.4 Feature extraction stage
Extracting features is to extract and transform details for
input information into several features, called a feature
vector, by decreasing the pattern of data representation. The
components set will obtain the extracted from the input data
(image) to execute the classification task.
3.4.1 Discrete wavelet transform (DWT)
DWT has the benefit of extracting the most appropriate
features of various paths and dimensions as it provides local
signal time-frequency details utilizing cascaded high-pass
and low-pass filter banks to features extracted in a
hierarchical structure [24]. As a consequence, there are four
sub-band images at each point (LL, LH, HH, HL).
Approximate components of the image may represent the LL
sub-band, while the LH, HL, HH sub-bands may be
considered as the accurate components of the image [24,25].
Fig. 4 shown DWT 2-level decomposition of the image. The
different frequency components and each component were
studied with a resolution matched to its scale and expressed
as:
 
 
The coefficients di,j refers to the component attribute in
signal p(u) corresponding to the wavelet function, whereas
bi,j refers to the approximated components in the signal. The
functions h(u) and g(u) in the equation represent high-pass
and low-pass filter coefficients, respectively, while
parameters i and j refer to wavelet scale and translation
factors. In simulated images, the following wavelets were
used: Symlet 7 (sym7), Daubechies 4 (db4), Hear (haar), Bi-
orthogonal 3.5, and 3.7 (bior 3.5 and bior 3.7) respectively.
These wavelets were used to compare and find which level
of decomposition and which wavelet produces better results.
In this work, a 2-level decomposition of (sym) wavelet is
used to derive 12 features for each brain MRI. In this work,
various statistical features are calculated such as mean,
standard deviation, median, range, mode, max, min, etc. In
this step, the features extracted from the gray image are based
on a histogram using DWT with Symlet wavelet. The
proposed procedures by using Symlet wavelet
implementation to extract MRI images feature is described
as the following:
 Menu from command.
 Select the wavelet 2-D toolbox of MATLAB.
 Load MRI brain images.
 Decomposition MRI brain images Sym2.
 Set several levels (2-level ( .
 Set an analysis image.
 Detect the Histogram of MRI images.
 Load statistical analysis for the original image.
Fig. 4: Decomposition-tree-of-an-image-with-2D-DWT.
3.4.2 Texture features by gray level co-occurrence matrix
(GLCM(
The texture is an essential feature for the analysis of several
categories of images [26,27] because it gives a good
reference of image information [26], biomedical image
analysis, remote sensing, and automated inspection of
traditional texture analysis applications [28]. Texture
evaluates attempts to measure the intuitive properties defined
by terms such as coarse, smooth, velvety, or rough as a
feature of spatial variation in pixel intensity. Simple texture
features can be determined by calculating statistical
characteristics, such as mean and variance from the image's
gray level histogram. However, the quality of this kind of
first-order statistics is generally poor. Calculated second-
order gray - level statistics utilizing gray-level co-occurrence
matrices (GLCM) and described fourteen statistical
measurements for texture from images [28]. In this work, the
function used (graycomatrix) [29] in MATLAB generates
(GLCM) gives information on texture patterning and co-
occurrence used to compute textural characteristics of the
texture. The GLCM features describe the pixels of an image

Abd Alreda, Khalif & Saeid | 5
by computing how many pixels sets of particular values and
spatial configuration happen in an image, generating a
GLCM, and then generating statistical measurements from
this matrix [27].
Algorithm: For calculating GLCM measures for each pixel:
Step1: Read an image of the MRI input.
Step2: Convert the type of data to double,
Step3: Extract from the input image an 8×8 matrix image.
Step4: Calculate the amount of the texture that co-occurred.
Step5: Estimate the parameters of the texture for the image
obtained.
Step6: Repeat step 3,4and 5 by shifting the window until
the picture finish.
Step7: View parameters of different texture by normalizing
them.
The texture extract features from the GLCM matrix are
contrast, correlation, energy, smoothness, homogeneity, and
additional IDM. Added features from this matrix were also
derived which are: mean, standard deviation, entropy,
variance, kurtosis, skewness [27,28].
3.4.3 Area
The actual number of pixels returned as a scale in the region.
(This value may vary significantly from the value returned
by the (beware) function in MATLAB, which weights
different pixel patterns differently) [30].
3.5. Tumor classification stage
The system consists of two panels: the feature extraction
stage (based on preprocessing and post-processing
techniques) and the classification stage (based on artificial
intelligence algorithms and SVM).
3.5.1 Artificial neural network structure
An artificial neural network (ANN) is a machine learning or
a computer model that depends on biological neural systems
[31]. An ANN includes several basic groups called "neurons"
or artificial neurons that are interlinked and work in parallel
with processing information, known as parallel distributed
processing systems or connecting systems [32]. In this work,
the feed-forward backpropagation network (FFBPN) used,
which is one of the kinds of artificial neural networks due to
the ease of use, has a lot of learning algorithms that can use
in the training of the multi-layer neural network.
Setting the neural network as the following:
 Open (nn tool) package from command of MATLAB.
 Import the input layer consists of 13 neurons that are
reliant on categorical variables and the number features
extracted using a gray-level co-occurrence matrix
(GLCM) representing data input. (from the workspace)
 The import output layer consist of one neuron represents
the target data vector. (from the workspace)
 Set one hidden layer consist of 3 neurons by using Trial
and error
 Set the main Training parameters used in the network
for FFBPN.
 Set activation function tensing (sigmoid)
 Minimum gradient threshold of 0.1 ×10-10.
 The maximum number of epochs 1000.
 Sum-squared error goal of 0.001.
 Spread constant 1.
 Maximum time to train infinity.
 fail-max 10 times to validation check.
 Run to train network.
3.5.2 Support vector machine
Vladimir N. Vapnik contributed a support vector machine
algorithm and its modern model was designed by Cortes and
Vapnik in 1993 [33]. SVM is a supervised machine learning
algorithm used to classify data for different classes based on
a separating hyperplane. Although there two main types of
SVM classifications: linear and non-linear [34]. The input
vectors are assigned to the feature vector using the kernel
function that directly calculates the dot product in the feature
space. The hyper-plane is formed in a dimensional space
separated into two classes, which Optimizes the distance
between each other and closes the training sets. This hyper-
plane is used as a basis for classifying vectors of unknown
objects (testing objects).
The training sample is  =1 , where  is
represents the set of inputs for the instance and 
represents the output classes (target), which is represented by
= +1 or -1 and each refers to the linearly separable point
class, as shown in Fig. 5. The hyper-plane equation that
separates is [35]:

Where:
X: is an input vector.
W: is a modifiable weight vector.
b: is a bias.
 
 
Fig. 5. Support Vector Machine.
The case of the non-linear samples, the distribution of
samples is irregular, so that SVM can determine the
limitations of non-linear decisions by using kernel function
and using two phases. The first phase is the use of the kernel
equation that transforms the coming input into the higher
dimensional space, And the second phase of the new location
searches for a linear hyperplane to classify the samples [34].
There are several types of kernel function below [33] :
1. Radial basis function (RBF (

Where  and  are sample vectors and  is a
Euclidean distance between sample vectors.
2. Linear kernel
   

Where  and  are sample vectors where  is the linear
kernel which is a function of linear vectors.

6 | Abd Alreda, Khalif & Saeid
3. Polynomial kernel
 

Where  and  are sample vectors.
IV. RESULTS AND DISCUSSION
The experimental results provided by the work technique
depicted for the segmented outcome and the extracted tumor
region are given in Fig. 6. Concerning the variables with high
normalized importance such as kurtosis, contrast, entropy,
energy, correlation, and all other variables are not reflecting
a threshold of classification to categories of brain tumor
categories on their own. This is because other factors need to
be involved and added to the set of independent variables.
Laboratory information may add substantial information
about the type of brain tumor. Observation of the brain tumor
images suggests that tumors can be characterized by three
factors; these are site, size, and shape. Information about
these factors may also help to produce more reliable criteria
of classification and Fig. 7 shown the training network for
FFBPN. Lastly, 13 features are selected for classification.
Feature extraction using the GLCM method where
classification using GLCM-based features provides
classification accuracy. Also, The GLCM method is easy to
perform which has been shown to deliver very excellent
results in a variety of application fields [36]. The GLCM
method has a good processing time and complexity
performance [37] compared to the DWT method. Table 1
shows the texture and area features for eight MRI images
including cases (normal, benign, and malignant) for the
database, after which will be entered as inputs for the works
used for the classification of an image brain tumor after the
normalization process.
(a) (b) (c) (d) (e)
Fig. 6: Result of pre-processing AND segmentation (a)
Original MRI input image, (b) Anisotropic diffusion
filtering, (c) Proposed thresholding segmentation, (d)Eroded
and Morphological operation, and (e) Tumor outline.
The typical performance for the FFBPN pre-training using
various numbers of neurons in the hidden layer. The 2, 3, and
5 of neurons in the hidden layer has been chosen as the best
numbers after employing many training operations for the
FFBPN using trial and error method. The 3 neurons were
obtained as an optimal number used for the hidden layer. The
FFBPN gave 99.93% accuracy for the training dataset, while
99.99% and 76.23% accuracy was recorded for testing and
validation datasets respectively. So the unanimous accuracy
of the overall method becomes 96.25%. The given accuracy
is observed only for 31 epochs. Fig. 8and Fig. 9 shows the
system performance of our suggested methodology.
Fig. 7: Topology structure of FFBPN for training network.
In this work, using the linear kernel function for
transformation. The corresponding performance of the SVM
classification is used by the kernel function. The results are
shown in the command window of the MATLAB for brain
image it is benign or malignant (normal or abnormal) as
shown in Fig. 10.
V. PERFORMANCE ANALYSIS AND COMPARISON
The training dataset images for which the extracted features
have been practiced using the classification process classifier
(FFBPN & SVM), while the test dataset is not trained using
the FFBPN & SVM classifier, mostly the statistical and
textural features have been extracted. The precision of the
training and testing images were analyzed by comparison
focused on the classification of (normal, benign, and
malignant) tumors cells for FFBPN, while SVM was
centered on the classification of (benign and malignant)
tumors cells. The accuracy or classification performance rate
is the effectiveness of the relevant classification for the
overall amount of classification checks. The output of the
algorithm proposed can be computed using predictive values.
There are four predictive values: true positive values (TP),
true negative values (TN), false-positive values (FN), and
false-positive (FP). This was used to measure the efficiency
of the work results presented to the MRI images by
sensitivity, specificity, and accuracy as shown in equation
(13) (14) (15). The calculation for test images (images used
for test after the training process) shown in Table 2 for SVM
and Table 3 for FFBPN.

Abd Alreda, Khalif & Saeid | 7
TABLE 1
Textural features for eight of MRI images in addition to the area of the extracted tumor.
Features
Image 1
Image 2
Image 3
Image 4
Image 5
Image 6
Image 7
Image 8
Mean
0.0175
0.0653
0.0524
0.0107
0.1484
0.0061
0.0283
0.05338
Standard .dev
0.3486
0.6996
0.6428
0.2802
1.0703
0.2182
0.4631
0.64504
Skewness
18.9631
10.5919
11.4712
23.5019
6.8332
13.5967
15.0402
11.0665
Kurtosis
4.2E+02
1.21E+02
1.41E+02
6.4E+02
9.6E-01
4.1E+02
2.6E+02
1.3E+02
Energy
0.9504
0.9040
0.9259
0.9614
0.8519
0.9529
0.9393
0.908
Entropy
0.2979
0.5567
0.4597
0.2410
0.8074
0.3047
0.3361
0.5078
Contrast
0.3426
0.5319
0.3161
0.2544
0.7581
0.3512
0.4082
0.5755
Homogeneity
0.9845
0.9731
0.9815
0.9876
0.9637
0.9847
0.9812
0.9737
Variance
0.1150
0.40696
0.3722
0.0741
0.9009
0.0464
0.1932
0.3387
Correlation
0.4727
0.6474
0.7211
0.4262
0.7322
0.2812
0.5202
0.5796
smoothness
0.9902
0.9974
0.9967
0.9841
0.9988
0.9723
0.9939
0.9968
IDM
7.5726
53.7549
10.9712
2.1702
123.8235
2.2627
42.4959
27.1191
Area
774.75
2957
2487.5
501.25
6898.375
299.25
1318
2533.25
Fig. 8: Typical performance for the FFBPN training
Fig. 9: (Training and validation) and (Testing and overall)
regression depicting the relationship between targets and
outputs.
Fig. 10: The result of an SVM classifier
TABLE 2
The experiments dataset in SVM.
Cases
Benign
Malignant
Total
Benign
42TP
1FP
43
Malignant
6FN
51TN
57
Total
48
52
100

8 | Abd Alreda, Khalif & Saeid
TABLE 3
The experiments dataset in FFBPN.
Cases
Normal
Benign
Malignant
Total
Normal
36
0
0
36
Benign
1
29
2
32
Malignant
0
0
32
32
Total
37
29
34
100
 
 



 
The results of FFBPN and SVM classification performance
where SVM classifier gave accuracy 93% while, FFBPN
gave 97% using the same features as input in them. FFBPN
gave the best result for the classification MRI images for
brain tumors as shown in Table 4. In this section also,
presents a performance comparison of the existing method
with the results of the previous works in images processing
filed taking from MRI for classifying brain tumor based on
different algorithms so higher accuracy of brain tumor
classification is the plan of each researcher that tried to
stratify it as shown in Table 5.
TABLE 4
FFBPNN and SVM classification performance.
Accuracy
Specificity
Sensitivity
Classifier
type
93 %
98.08%
87.5%
SVM
97%
98.78
97.39%
FFBPN
VI. CONCLUSIONS
This work aimed to submit an Automated algorithm for
detecting brain tumors from MRI images by Artificial Neural
Networks and SVM. The data collected 308 images and
prepared by pre-processing and post-processing processes to
make it suitable for detection. Analysis of statistical features
has been used to extract features from images; attributes
calculated from Graycomatrix features dependent on the
(GLCM) of images. In the case of artificial neural networks,
the feed-forward backpropagation network with supervised
learning has been used to classify the images as normal,
benign, and malignant. And all the right features were used
as input variables for the FFBPN, then the network was
trained and its performance is measured. Finally; the
proposed algorithm, the FFBPN, gives the best results by
detecting and classifying brain tumors according to the
extraction feature achieved results with a precision of 97%
while the SVM has low accuracy. The proposed system
effectively classifies the images of the brain tumor of the
MRI.
TABLE 5
Comparison result with previous works of detection and
classification of brain tumors.
Authors
Features
Classification
Algorithm
Accuracy
2014 [38]
Extracted 4
textural
features by
GLCM
method
KNN
97.3%
2017 [7]
Extracted 4
textural
features by
GLCM
method
SVM-based
classification
95%
2018 [8]
Extracted 5
textural
features by
GLCM and
DWT
PNN classifier
Nearly 100%
was achieved
for trained
and95% for
tested
2019 [39]
Extracted 9
features by
DWT
technique
FF-ANN
95.48%
2019 [9]
9 features by
GLCM
method
SVM, FFNN
and ELM
92.99%,89.41%
and 90.20%
2020 [10]
Extracted 5
textural
features by
GLCM
SVM
90%
2020[40 ]
3 kinds of
features are
extracted as
shape
features-
intensity
features and
kurtosis
texture
features by
DWT
KNN and
SVM
85.2%,90.75%
with
smote sampling
disabled and
88.75%
,93.32%y with
smote sampling
enabled
Propose
Algorithm
Extracted 13
textural
features by
GLCM
method
SVM and
FFBPN
93% and 97%
VII. SUGGESTIONS FOR FUTURE WORK
This work can be extended in the future according to
expansion the data base to include the other type of MRI
images, developing the algorithm to detect and diagnoses
brain tumor to the 3D images MRI.Using the algorithm to
extend to the analysis of other medical images, such as CT
and PET images. Attempt to contact other laboratories on the
internet carrying out similar research to share data and work
together to establish to open test database.

Abd Alreda, Khalif & Saeid | 9
REFERENCE
[1] Young, RM; Jamshidi, A; Davis, G; Sherman, J. H.
"Current trends in the surgical management and
treatment of adult glioblastoma", Annals of
Translational Medicine, Vol. 3, Issue 9, June 2015.
[2] Umar, A. A., and S. M. Atabo. "A review of imaging
techniques in scientific research/clinical diagnosis",
MOJ Anat. & Physiol., Vol. 6, Issue 5, 2019, pp. 175-
183.
[3] Khodamoradi F., et al., “The incidence and mortality of
brain and central nervous system cancer and their
relationship with human development index in the
world”, World Cancer Res J, Vol. 4, 2017.
[4] Almutrafi Amna, et al., “The Epidemiology of Primary
Central Nervous System Tumors at the National
Neurologic Institute in Saudi Arabia: A Ten-Year
Single-Institution Study”, Journal of Cancer
Epidemiology, 2020.
[5] Naveena H., Shreedhara K., and Mohamed Rafi,
“Detection and Classification of Brain Tumor using
BPN and PNN Artificial Neural Network Algorithms”,
International Journal of Computer Science and Mobile
Computing. IJCSMC, Vol. 4, Issue. 4, April 2015,
pp.782 – 789.
[6] Chaddad Ahmad and Tanougast Camel “Quantitative
evaluation of robust skull stripping and tumor detection
applied to axial MR images”, Brain Informatics, Vol. 3,
Issue 1, 2016, pp. 53-61.
[7] Bahadure, Nilesh Bhaskarrao et al., “Image analysis for
MRI based brain tumor detection and feature extraction
using biologically inspired BWT and SVM”,
International journal of biomedical imaging, 2017.
[8] Shree, N. Varuna; Kumar, T. N. R., “Identification and
classification of brain tumor MRI images with feature
extraction using DWT and probabilistic neural
network”, Brain informatics, Vol. 5, Issue 1, 2018.
[9] Parasuraman Kumar, B. VijayKumar, “Brain Tumor
MRI Segmentation and Classification Using Ensemble
Classifier”, International Journal of Recent Technology
and Engineering (IJRTE), ISSN: 2277-3878, Vol. 8,
Issue 1S4, June 2019, pp. 244-152.
[10] Ahsanullah Umary, Harpreet Kaur., “Automatic Brain
Tumour Diagnosis and Segmentation: based on SVM
Algorithm”, International Journal of Innovative
Technology and Exploring Engineering (IJITEE)ISSN:
2278-3075, Volume-9 Issue-6, April 2020, pp. 1079-
1084
[11] Padmanaban, Sriramakrishnan & Thiruvenkadam,
Kalaiselvi & T., Padmapriya & KUMAR, RAM &
Nagarajan, Kalaichelvi. "An Medical Image File
Formats and Digital Image Conversion". International
Journal of Engineering and Advanced Technology. Vol-
9 Issue-1S4. 2020. 74-78.
[12] J. S. Ignacio, S. J. Buso, and W. A. Monteiro,
"processing and analysis of digital images: How to
ensure the quality of data captured?", International
Journal of Recent advances in Mechanical Engineering
(IJMECH), Vol. 2, No. 2, 2013.
[13] Larobina, Michele; Murino, Loredana. "Medical image
file formats". Journal of digital imaging, 2014, Vol. 27,
Issue 2, pp. 200-206.
[14] Convert RGB image or colormap to grayscale -
MATLAB ...
https://www.mathworks.com/help/matlab/ref/rgb2gray.
html
[15] Contrast Adjustment–MATLAB & Simulink – Math
works.
https://www.mathworks.com/discovery/image-
enhancement.html.
[16] K, S & Kalitkar, Kishan & Subba, G & Rao,. (2015)." A
Review on Noise Reduction Methods for Brain MRI
Images". SPACES-2015, Dept. of ECE, K L
UNIVERSITY. 10.1109/SPACES.2015.7058284.
[17] P. Perona and J. Malik. "Scale-space and edge detection
using anisotropic diffusion". IEEE Transactions on Pat-
tern Analysis and Machine Intelligence, Vol. 12, Issue
7, pp. 629–639, 1990.
[18] Reddy, Mummadi Gowthami; REDDY, Palagiri Veera
Narayana; REDDY, Patil Ramana. Medical Image
Fusion Using Integrated Guided Nonlinear Anisotropic
Filtering with Image Statistics. International Journal of
Intelligent Engineering and Systems, Vol.13, No.1,
2020
[19] Weickert, J. "Anisotropic diffusion in image
processing”. Stuttgart: Teubner, Vol. 1 (1998). pp. 59-
60.
[20] Zhan Yantong; Zhang Guoying. "An improved OTSU
algorithm using histogram accumulation moment for ore
segmentation". Symmetry, 2019, 11.3: 431.
[21] Cao, Xinhua, et al. A robust parameter-free thresholding
method for image segmentation. IEEE Access, 2018, 7:
3448-3458.
[22] Qidwai, Uvais, and Chi-hau Chen. "Digital image
processing: an algorithmic approach with MATLAB".
CRC press, 2009.
[23] Ilhan Umit; Ilhan Ahmet. "Brain tumor segmentation
based on a new threshold approach". Procedia computer
science, 2017, 120: 580-587.
[24] Icke, Guzide, Kamarthi, V. Sagar, “Feature extraction
through discrete wavelet transform coefficients”,
Intelligent Systems in Design and Manufacturing VI,
Proceedings of the SPIE, Volume 5999, pp. 27-35, 2005.
[25] Fouad, Amal; MOFTAH, Hossam M.; HEFNY,
Hesham A. Brain Diagnoses Detection Using Whale
Optimization Algorithm Based on Ensemble Learning
Classifier. International Journal of Intelligent
Engineering and Systems, Vol.13, No.2, 2020
[26] Mahmood, Faleh H.; Abbas, Wafaa A. Texture Features
Analysis using Gray Level Co-occurrence Matrix for
Abnormality Detection in Chest CT Images. Iraqi
Journal of Science, Vol. 57, Issue 1A, pp. 279-288,
2016.
[27] Armi Laleh Fekri-Ershad Shervan, “Texture image
analysis and texture classification methods-A review”,
arXiv preprint arXiv:1904.06554, 2019.

10 | Abd Alreda, Khalif & Saeid
[28] Zulpe N., Pawar V., “GLCM Textural Features for Brain
Tumor Classification”, International Journal of
Computer Science Issues, Vol. 9, Issue 3, No 3, 2012.
[29] Texture Analysis Using the Gray-Level Co–
MathWorks.
https://www.mathworks.com/help/images/texture-
analysis-using-the-gray-level-co-occurrence-matrix-
glcm.html.
[30] Area of objects in binary image - MATLAB bwarea.
https://www.mathworks.com/help/images/ref/bwarea.ht
ml
[31] Dr. Adel M. Dakhil, "Short Term Load Forecasting
Based Artificial Neural Network", Iraq Journal for
Electrical and Electronic Engineering, Vol. 10, No. 1,
2014.
[32] K. L. Du, M. N. Swamy, “Neural Network in Soft
Computing”, Springer, 2006.
[33] Singh, Khushboo; Verma, Satya. Detecting Brain Mri
Anomalies By Using Svm Classification. International
Journal of Engineering Research and Applications
(IJERA) Vol. 2, pp. 724-726, 2012.
[34] M.S. Uzer, N. Yilmaz and O. Inan,"Feature Selection
Method Based on Artificial Bee Colony Algorithm and
Support Vector Machines F=for Medical Datasets
Classification". The Scientific World Journal, 2013.
[35] F. Melgani And L. Bruzzone, "Classification of Hyper
spectral Remote Sensing Images With Support Vector
Machines", IEEE Transactions on Geoscience and
Remote Sensing, Vol. 42, Issue 8, pp.1778-1790, 2004.
[36] P. Mohanaiah, P. Sathyanarayana, and L.
GuruKumar."Image texture feature extraction using
GLCM approach". International Journal of Scientific
and Research Publication., vol. 3, no. 5, p. 1, 2013.
[37] M. Mehri, M., Héroux, P. Gomez-Krämer, and R.
Mullot."Texture feature benchmarking and evaluation
for historical document image analysis". Int. J.
Document Anal. Recognit., vol. 20, no. 1, pp. 1–35,
Mar. 2017
[38] Chitari Amruta, BAG V. V., “Detection of brain tumour
using classification algorithm”, Int. J. Invent. Comput.
Sci. Eng, Vol. 1, pp. 2348-3539, 2014.
[39] Zahid Ullah, Su-Hyun Lee, Muhammad Fayaz,
“Enhanced feature extraction technique for brain MRI
classification based on Haar wavelet and statistical
moments”, International Journal of Advanced and
Applied Sciences, Vol. 6, Issue 7, 2019, pp. 89-98.
[40] Latha, r. S., et al., “Brain Tumor Classification Using
SVM and KNN Models For Smote Based MRI Images”,
Journal of Critical Reviews, Vol. 7, Issue 12, pp. 1-4,
2020.