125
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
HYBRID CLASSIFIER FOR BRAIN ABNORMALITY
DETECTION IN BRAIN MRI
Adhi Lakshmi
Associate Professor, Electronics and Communication Engineering, Kalasalingam
Academy of Research and Education, Tamilnadu, (India).
E-mail: lakshmi@klu.ac.in
ORCID: http://orcid.org/0000-0002-6744-7048
Thangadurai Arivoli
Professor/Principal, Electronics and Communication Engineering,
Vicram College of Engineering, Tamilnadu, (India).
E-mail: t.arivoli@gmail.com
ORCID: http://orcid.org/0000-0003-1612-1029
M. Pallikonda Rajasekaran
Professor/ Director of Controller of Examination,Electronics and communication
EngineeringKalasalingam Academy Of Research and Education, Tamilnadu, (India).
E-mail: m.p.raja80@gmail.com
ORCID: https://orcid.org/0000-0001-6942-4512
N. Bhuvaneshwary
Assistant professor, Electronics and Communication Engineering, Kalasalingam
Academy of Research and Education, Krishnankoil, Tamilnadu, (India).
E-mail: bhuvaneswary.n@klu.ac.in
ORCID: http://orcid.org/0000-0001-6400-6602
S. Sathya
PG student,Eectronics and Communication Engineering, Kalasalingam Academy Of
Research and Education, Krishnanloil, Tamilnadu, (India).
E-mail: ssathya08nov@gmail.com
ORCID: https://orcid.org/0000-0002-1440-663X
Recepción:
25/10/2019
Aceptación:
20/10/2020
Publicación:
30/11/2021
Citación sugerida:
Lakshmi, A., Arivoli, T., Rajasekaran, M., Bhuvaneshwary, N., y Sathya, S. (2021). Hybrid classier for
brain abnormality detection in brain MRI. 3C Tecnología. Glosas de innovación aplicadas a la pyme, Edición
Especial, (noviembre, 2021), 125-153. https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
126
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ABSTRACT
Brain abnormality detection is very essential now a days. Brain abnormality includes both
massive abnormal growth of cells called tumour and blood lump in a veins or artery. If The
abnormalities are detected at the earlier stages it would help to save life. The symptoms for
the abnormalities are same and vary with patient aging and severity of the abnormality.
The algorithm which works for tumor can not be used for the detection of abnormality
caused in blood lumps of veins and artery. Normally the tumor is the massive growth of the
cell and the features can be extracted and applied for the classier to identify the severity
of the abnormality. But the detection of abnormality caused in vein or artery due to blood
lumps are very dicult to identify and feature extraction is also dicult. A sophisticated
algorithm should be used for identifying the blood lumps. This paper deals with hybrid
classier (SVM and ANFIS) for detecting the abnormalities such as tumour as well as
stroke. Till now separate algorithms are used for detecting tumour or stroke from brain MR
image. In our proposed work it is possible to identify stroke or tumour with same algorithm
by using dierent hybrid classiers. The proposed system helps the physicians to diagnose
human brain stroke. Accuracy of 0.999, sensitivity 0.38, specicity 0.86, PPV 0.91, NPV
0.99 is obtained by ANFIS classier. Three quantitative events to calculate brain tumor
of average segmentation results: Similarity Index (SI), Overlap fraction (OF), and Extra
Fraction (EF) is 0.776194, 0.775198, 0.222213 is obtained by SVM classier.
KEYWORDS
SVM, ANFIS, Brain Stroke, Tumour.
127
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1. INTRODUCTION
Brain Stroke
Stroke otherwise called as “Brain Attack” that happens in the brain rather than the heart.
A stroke or “brain assault” occurs:
• Due to blood lump in a veins or artery.
• Due to blood vein breakage, interfering with blood stream to a region of the brain.
As it happens, brain cells start to expire. When brain cells die during a stroke, functions such
as speech, movement and memory guarded by that part of the brain are vanished. The
lost of function depends on the location of the stroke and on its severity i.e., the amount
of brain cell loss. Several persons get well entirely from fewer severe strokes, while other
strokes can be considered life threatening. Opposite treatments are required for both strokes
namely ischemic stroke and hemorrhagic stroke.
Figure 1. Brain Attack. a) Ischemic, b) Hemorrhagic Stroke.
Suorce: own elaboration.
Brain Tumour
A tumour is unusual tissue that develops by unrestrained cell splitting up. Healing methods
vary depending on the following factors such as tumour type, size, location, age and medical
health of the person. Curing method may be therapeutic or attention on discharging
indications.
Figure 2. MRI Brain Tumours.
Suorce: own elaboration.
128
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Primary brain tumours may be benign (non-cancerous) or malignant (cancerous). A benign
brain tumour develops gradually and malignant brain tumour grows rapidly, has unequal
boundaries, and spreads to close by brain areas. Sometimes they are called brain cancer.
Whether a brain tumour is benign, malignant are potentially life-threatening. Grade means
the way tumour cells look under the microscope and is a suggestion of aggressiveness. If the
tumours are detected at near the beginning stage there is the opportunity of saving the life.
Cerebrovascular accident (CVA) is a damage caused because of functionality of brain due
to lack of ow of blood (Adams et al., 2005). This smash up is due to either ischemic or
hemorrhage. Blockage of blood in blood vessels inside the brain is known as Ischemic
stroke. Ischemicstroke results in about 80% deaths caused by all strokes. Bleeding of blood
in blood vessels inside the brain is called Hemorrhagic stroke (Udupa & Samarasekera,
1996). In any case, the damaged area of the brain cannot function in a normal way. When
brain cells expire, the tasks controlled by the region of the brain are vanished.
Kharat and Kulkarni (2012) presented the classication of brain tumors using neural
networks. Due to complexity and variance, the MR image classication was a dicult task.
Generally two neural network techniques were used.
The taken out features are categorized using Multi Layer Perceptron (MLP) (Amutha &
Rajagopalan, 2013). Feature lessening is expertise by grade of the features using Information
gain. Multilayer Perceptron (MLP) is used to categorize the extracted features.
Mid line of brain helps in measuring the brain’s amount of equilibrium. Here, the brain’s
amount of equilibrium was determined using MLS technique (Dzialowski et al., 2012). It
can detect early identication of ischemic stroke to improvise eectiveness and correctness
of medical practice.
Hema and Bhavani (2013) suggested SVM, K-NN, ANN and decision tree classication
with accuracy of 98%, 97%, 96% and 92%. This system helped the physicians for enhanced
detect of human brain stroke, for advance action. The brain area aected by abrasion can
be accurately detached from the brain image. It improves the exactness in recognition of
ischemic stroke.
129
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
An improved watershed image segmentation technique was existing by Bala (2012).
Watershed Transformation based an edge detection algorithm was used for image
segmentation and makes use of gradient operators in mathematical morphology. But
the drawback is over segmentation. To avoid over segmentation, Prewitt’s edge detection
operator with noise removal techniques and image enhancement wereestablished and the
trouble of over segmentation was also reduced.
Mahmoud-Ghoneim et al. (2003) developed in intensity-based analysis. Features of dierent
types can be calculated for texture analysis, such as co-occurrence matrix.
Udupa and Samarasekera (1996) presented the purpose fuzzy connectedness procedure has
been applied. The grades of work have 0.750, 0.706, and 0.107 SI, OF, and EF respectively.
Our proposed work has higher performance than the work by Udupa and Samarasekera
(1996).
From the literature survey it was understood that fully automated system to detect stroke or
tumour will assist the doctor very much. Till now separate algorithms are used for detecting
tumour or stroke from brain MR image. In our proposed work it is possible to identify
stroke or tumour with same algorithm by using dierent hybrid classiers.
2. MATERIALS AND METHODS
The main objective is to detect the presence of Brain Abnormalities in MRI by Hybrid
(SVM and ANFIS) classiers. The Brain Abnormalities such as stroke and tumor images
are classied by Image Intensity Threshold Level. After the classication, images are
preprocessed. After preprocessing, the features are extracted and the Hybrid classiers are
used to segmentation and detection of Brain Abnormalities and also performance analysis
if performed.
130
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 3. Proposed Data ow diagram.
Suorce: own elaboration.
2.1. PROPOSED METHOD FOR IDENTIFYING TUMOUR
If the tested MRI image intensity level is above 200, it will detect tumour otherwise to
detect stroke as shown in the block diagram in Figure 4 the following steps will be carried
in tumor segmentation:
• Pre-processing-Noise removal using Anisotropic lter.
• Canny Edge detection.
• Feature Extraction using GLCM matrix and LBP.
• Tumor detection using SVM classier.
131
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.1.1. PRE-PROCESSING-NOISE REMOVAL USING ANISOTROPIC FILTER
It is necessary to reduce image noise exclusive of removing noteworthy elements of the image
content such as edges, lines or other elements those are signicant for the understanding of
the image.
The dataset of numerous diverse images of the brain tumour have studied and through
testing threshold appropriate for those images are identied. Then, for same intensity
ranges in all the images, the maximum and the smallest intensities are restricted to the
range [0,255]. Then anisotropic diusion is dened as,
(1)
where ∆ means the Laplacian,
signies the gradient, div (…) is the divergence operator
and c(x,y,t) is the diusion coecient. C(x,y,t) organizes the degree of diusion and is
regularly preferred as a the image gradient so as to conserve boundaries in the image.
Pietro Perona and Jitendra Malik lead the way the knowledge of anisotropic diusion in
1990 and projected two functions for the diusion coecient:
(2)
(3)
Where K manages the sensitivity to boundaries. The noise removed brain MR image
byanisotropic ltering is shown in Figure 4.
132
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 4. Noise Removed brain MR image using Anisotropic lter.
Suorce: own elaboration.
The noise ltered image, the intensity gradient of the image is found out. This is done by
applying pair of convolution mask along the x and y directions. Then gradient strength and
direction is found out using the formula.
(4)
133
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
The direction was rounded o to any possible angle values namely (0, 45, 90 and 135).
By doing this operation pixel which belongs to edges of theimage will be maintained and
the pixel which are not belong to edges of the image are removed. Canny provides two
thresholds namely upper and lower. If the pixel gradient is larger than upper threshold
value, that pixel is established as edge. If the pixel gradient is lesser than the lower threshold
value, then it is discarded. The Canny Edge detected brain MRI is shown in Figure 5.
Figure 5. Canny Edge Detection.
Suorce: own elaboration.
134
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.1.2 FEATURE EXTRACTION USING GLCM MATRIX AND LBP
The replicating pattern of local dissimilarity in image intensity is called consistency. This
technique can be in consistency studies co-occurrence matrix. The information of pixel
strength was expected from image that can be done using co-occurrence matrix. This
pattern will compare by excellent consistency than by common consistency. The gray level
co-occurrence matrix (GLCM) technique is a method of removing second order values
stability elements. Though, the performance of a certain GLCM based on aspect, on
top of the position the consistency characteristics; hinge on on the amount of gray levels
applied. The subsequent representations are: mean value of P is represented as µ. were the
means and standard deviations of Px and Py are denoted as µx, µy, σx and σy. The size of
the occurrence matrix is represented as G. At this time the sum of rows and columns of
co-occurrence matrix is identical. The subsequent GLCM characteristics are used in our
study attempt: angular second moment, contrast, entropy, and correlation, sum of squares,
variation entropy, inverse distinction moment, inertia, cluster importance, cluster shade,
energy, homogeneity, distinction and diversity in variance. They are clear in following
Equations.
1) Angular second moment (ASM):
(5)
ASM is a determined similarity of the image. A standardized image will enclose simply
some gray intensities, GLCM provides simply a rare except comparatively elevated ideals
of P (i, j).
2) Contrast:
(6)
It is determined for the limited deviation in an image.
135
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
3) Inverse distinction moment (IDM):
(7)
IDM as well as inclined with the similarity of the image.
4) Entropy:
(8)
It assesses the disarray or density of an image.
5) Correlation:
(9)
It is calculated for gray level linear reliance
6) Sum of squares, variance:
(10)
It is calculated for gray level dissimilarities at an assured distance, d.
7) Variation entropy:
(11)
Variation entropy is an evaluate, of histogram satised and rational rate among two images.
136
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
8) Inertia:
(12)
The inertia species the portion of gray scales in the image.
9) Cluster shade:
(13)
The image is not symmetric while darkness is elevated.
10) Cluster prominence:
(14)
Cluster Prominence species the image is not symmetric while prominences are elevated.
11) Energy:
(15)
The energy of a surface explains the evenness of the surface. Energy is one for a stable
image.
12) Homogeneity:
(16)
Homogeneity proceeds a worth that events the nearness of the allocation of essentials in
the GLCM to the GLCM slanting. It is 1 for a slanting GLCM. A standardized image will
eect in a co-occurrence matrix with a mixture of elevated and short P [i, j]’s. A varied
image will eect in an even extend of P [i, j]’s.
137
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
13) Distinction:
(17)
It is determined to nd constancy between two groups.
14) Diversity in variance:
(18)
It is the summation of divergence among greatness of the middle pixel and its locality. The
GLCM feature is tabulated in Figure 6.
Figure 6. GLCM Features performance for Tumour.
Suorce: own elaboration.
Local Binary Patterns: the LBP feature vector is fashioned in the following manner:
• The inspected space is segregated into chambers (e.g. 16x16 pixels for each
chamber).
• For every pixel in a chamber, evaluate the pixel to each one of its 8 neighbors (on
its left-top, left-middle, left-bottom, right-top, etc.) either in clockwise or counter-
clockwise.
138
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
• If the middle pixel’s value is superior to the nearby value then replace with “1”.
If not, replace with “0”. Thus 8-digit binary number (which is generally changed
to decimal).
• Histogram is computed over the cell (i.e., arrangement of pixels are lesser or larger
than the midpoint) and feature vector of the box is obtained.
• The feature vector can be trained using the Support Vector Machine (SVM) and
ANFIS to classify images.
2.1.3 TUMOR DETECTION USING SVM CLASSIFIER
SVM fundamentally tries to split the given data into decision plane. Decision plane is a
hyper plane which splits the data into two classes. Training points obtained by the feature
vectors are the supporting vector which describes the hyper plane. The training data x
contains of n data trials each of m dimensions and tting to class y, is expressed as:
(19)
Table 1. GLCM Matrixfor Tumor Results.
PARAMETERS RESULTS
ASM 2.49
Contrast 5.47
IDM 9.28
Entropy 4.52
Correlation 2.04
Variance 6.63
Variance Entropy 8.38
Inertia 8.15
Cluster Shade 2.59
139
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Cluster
Prominence
6.26
Energy 3.79
Homogeneities 4.25
Distinction 6.25
Difference
Variance
5.47
Source: own elaboration.
SVM assigns data (xi ,y) into an innitedimensional hyper plane (xi ,y) by using Gaussian
kernel function and describes its rule decision as sign (f(x)). The discriminant function f(x)
produces the optimal hyper plane decision border by using weight vector w* and bias b*.
(20)
The optimal values of w* and b are assessed by explaining the following optimization
problem.
(21)
where, C is regularization constraint, which lastly produces optimum level for weight vector
w* and bias b*. Figure 7 represents the malignant brain tumour identied image by using
SVM classier.
140
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 7. Tumour Identied Image.
Suorce: own elaboration.
2.2. PROPOSED FOR IDENTIFYING STROKE
If the tested MRI image intensity level is below 200, it will detect stroke otherwise to detect
tumour. The following steps will be carried in stroke segmentation as shown in Figure 4. All
process are same except the classier.
• Noise and skull removal in pre processing.
• K-means segmentation.
• Feature Extraction using GLCM.
• Stroke detection using ANFIS classier.
2.2.1. PRE-PROCESSING-NOISE AND SKULL REMOVAL
The same preprocessing steps of brain tumor MRI is carried over here The noisy image
removed from noise and skull is shown in Figure 8.
141
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 8. (a) Noise affected image. (b) noise removed image. (c) Skull removed image.
Suorce: own elaboration.
2.2.2 K-MEANS SEGMENTATION
K-means clustering targets to divide n interpretations into k clusters in which each reection
belongs to the cluster with the near mean, helping as a model of the cluster.
The subsequent steps of K-means:The total number of clusters k with primary cluster
centroid was chosennvi=1,2,…k (22)
Separation of the input statistics points into k clusters with allocating everystatistics element
xj to the neighbouring cluster centroid vi by nominated distance extent. Euclidean distance
equation was given as,
(23)
Where X = {x1, x2, . . .xn} is the input data set.
Cluster transfer matrix Ui calculated which is the separation of the data points with the
binary bias value of the jth data point to the ith cluster such that U = |ui, j| and u i, j in
{0,1}for all i, j
(24)
Centroid is rearranged using the membership values by
142
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
(25)
The iteration will be carried over by step 2, until there is no change in the centroid of the
cluster from the previous iteration. The k-means clustering process improves the sum-of-
squared-error-based objective function Jw(u, v) then
(26)
The projected method K-means clustering is exact hopeful algorithm for clustering
formation from Ischemic stroke MR image. To test our projected technique a Magnetic
Resonance Imaging (MRI) image of human brain was taken. The k-mean segmented
image is given in the Figure 9.
Figure 9. k-means clustering image.
Suorce: own elaboration.
2.2.3 FEATURE EXTRACTION USING GLCM
The feature extraction is carried as in brain tumour MRI. The features are tabulated in
Table 2. In Figure 10 GLCM features analysis of stroke graph is provided.
143
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 10. GLCM Features performance for stroke.
Suorce: own elaboration
Table 2. GLCM Matrix for Stroke Results.
PARAMETERS RESULTS
ASM 5.51
Contrast 4.02
IDM 9.63
Entropy 4.52
Correlation 1.62
Variance 6.63
Variance Entropy 4.83
Inertia 1.81
Cluster Shade 5.64
Cluster
Prominence
3.43
Energy 4.35
Homogeneities 4.55
Distinction 5.33
Difference
Variance
4.02
Suorce: own elaboration.
144
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Stroke detection using ANFIS classier
It uses hybrid knowledge technique that proposes ANFIS to build an input-output plotting
model. In the simulation, the ANFIS architecture uses IF THEN rules.
ANFIS Architecture
Assume - two inputs X and Y and one output Z
Rule 1: If x is A1 and y is B1,
then f1=p1x+q1y+r1 (27)
Rule 2:If x is A2 and y is B2,
then f2 = p2x + q2y +r2 (28)
Layer 1
Figure 11. Layer Architecture of ANFIS.
Suorce: own elaboration.
Every node i in this layer is an adaptive node with a node function, O1,i = mAi (x), for I =
1,2, or O1,i = mBi-2 (y), for I = 3,4 (29)
145
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Layer 2
Figure 12. 2nd layer Architecture of ANFIS.
Suorce: own elaboration.
Every node i in 2nd layer is a xed node labeled ∏, whose yield is the multiplication of all
the incoming signals:
O2,i = Wi = min{mAi (x) , mBi (y)}, i = 1,2 (30)
Each node output characterizes the ring asset of an instruction.
Layer 3
Figure 13. 3rd layer Architecture of ANFIS.
Suorce: own elaboration.
Every node in 3rd layer is a xed node labeled N.
O3,i = Wi = Wi /(W1+W2) , i =1,2 (31)
146
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
The ith node estimates the proportion of the ith instruction’s ring asset to the sum of all
rules’ringstregths.
Layer 4
Figure 14-15. 4th layer Architecture of ANFIS.
Suorce: own elaboration.
Every node i in this layer is an adaptive node with a node function,
O 4,i = w̅i = w̅i (pix + qiy +ri) …Consequent parameters. (32)
Layer 5
Figure 15. 5th layer Architecture of ANFIS.
Suorce: own elaboration.
The nodes in this layer is a stable node labeled S, used to calculate the complete yield as the
summation of all received signals.
O5,1=Siw̅if̅i. (33)
By using ANFIS classier, stroke can be detected and is shown in Figure 16.
147
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 16. Stroke Identied Image.
Suorce: own elaboration.
3. RESULTS
3.1. PERFORMANCE ANALYSIS FOR STROKE USING ANFIS CLASSIFIER
The classier namely ANFIS had established to ischemic brain and normal brain. The
performance of the classier in provisions of sensitivity, specicity and accuracy was
estimated. The performances are tabulated in Table 3. The accuracy obtained in 0.99
using ANFIS classier.
Human brain stroke can be diagnose using this proposed system. Accuracy of 0.998428
isobtained by ANFIS classier. In Figure 17 performance analysis graph is given using
ANFIS classier.
Figure 17. Performance Evaluation Graph in Stroke.
Suorce: own elaboration
148
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
3.2. PERFORMANCE ANALYSIS OF TUMOUR
The SVM Classiers had established to classify benign brain tumour and malignant brain
tumour. The performance of the classiers in provisions of similarity index, overlap fraction
and Extra fraction was estimated.
Three quantitative events to calculate in the region of the segmentation result are Similarity
Index (SI), Overlap fraction (OF), and Extra Fraction (EF).
Figure 18. Comparison between the various performance parameter by proposed segmented regions and
manual segmentation.
Suorce: own elaboration.
Figure 18 shows that TP is the amount of true-positive pixels identied by manual
segmentation and proposed method, FP is the amount of false-positive pixels noticed by
manual segmentation and proposed method, and FN is the amount of false-negative pixels
comparative to the tumourarea with manual segmentation but it is not identied by the
proposed method.
(34)
(35)
(36)
In a worthy segmentation Similarity Index (SI), Overlap Fraction (OF) which are achieved
by equations (36), (34) should be high and Extra fraction (EF) obtained by equation (35)
149
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
should be low. The performances are tabulated in Table 3. In Figure 17 performance
analysis graphs are given.
Table 3. Segmentation.
PARAMETERS
SIMILARITY INDEX
(SI)
OVERLAP FRACTION
(OF)
EXTRA FRACTION
(EF)
Tumour Image set 1 0.785531 0.783523 0.211364
Tumour Image set 2 0.766866 0.766866 0.233134
Average 0.776198 0.775194 0.222249
Suorce: own elaboration.
4. DISCUSSION
Our projected method is executed in MATLAB. The projected method has been tried on
dataset of real brain MR images containing of normal and tumor brain images. Hybrid
classier for detecting abnormalities in brain has been proposed. The abnormality may be
stroke or brain tumour. Separate algorithm will be used for detecting brain tumour and
stroke. In the projected method, the abnormality region is segmented and it is treated as test
image. Based on the intensity level of the abnormalities, either it will detect stroke or tumour.
If the threshold value is less than 200 then it will be considered as stroke brain MRI. Then
pre-processing and skull removal of image is carried out. After which the segmentation
is passed out by using K-means algorithm. Then from the segmented result, features are
extracted using GLCM matrix in order to train the ANFIS classier and performance
analysis is carried out. If the threshold value is above 200 then it will be considered as
tumour brain MRI. Then pre-processing and edge detection of image is carried out by
using anisotropic lter and canny edge detections method. After which the segmentation
is passed out by K-means algorithm. Features are extracted using GLCM matrix and LBP
in order to train the SVM classier and performance analysis is carried out in terms of
Similarity Index (SI), Overlap Fraction (OF), and Extra Fraction (EF).
150
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
5. CONCLUSIONS
The proposal of image segmentation and classication appears to be an interesting task,
as the medical images are exposed to noise. In particular anisotropic lter based image de-
noising scheme proves to be an ecient method for noise removal. Computer aided detection
system helps in detecting ischemic stroke and malignant brain tumor has been developed.
This system detects to generate accurate quantitative results. The proposed system helps the
physicians to diagnose human brain stroke. Accuracy of 0.999, sensitivity 0.38, specicity
0.86, PPV 0.91, NPV 0.99 is obtained by ANFIS classier. Three quantitative events to
calculate brain tumor of average segmentation results: Similarity Index (SI), Overlap
fraction (OF), and Extra Fraction (EF) is 0.776194, 0.775198, 0.222213 is obtained by
SVM classier. It usually aided to progress the accuracy in detection of malignant brain
tumor.
ACKNOWLEDGEMENT
We would like to thank our institution Kalasalingam Academy of Research and Education
for support.
REFERENCES
Adams, H., Adams, R., Zoppo, G. del, & Goldstein, L. B. (2005). Guidelines
for the early management of patients with ischemicstroke, guidelines update, a
scientic statement from the Stroke Council of the American Heart Association/
American Stroke Association. Stroke, 36(2), 916–921. https://doi.org/10.1161/01.
STR.0000163257.66207.2d
Amutha, C., & Rajagopalan, S. P. (2013). Brain stroke classication based on MultiLayer
Perceptron using Watershed segmentation. Journal of Theoretical and Applied Information
Technology, 56(2), 410-416. https://www.researchgate.net/publication/285940599_
Brain_stroke_classication_based_on_multi-layer_perceptron_using_watershed_
segmentation_and_gabor_lter
151
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Bala, A. (2012). An Improved Watershed Image Segmentation Technique using MATLAB.
International Journal of Scientic & Engineering Research, 3(6), 5-8. https://www.ijser.
org/researchpaper/an-improved-watershed-image-segmentation-technique-using-
matlab.pdf
Bian, W., Hess, C. P., Chang, S. M., Nelson, S. J., & Lupo, J. M. (2013). Computer-
aided detection of radiation-induced cerebral micro bleeds on susceptibility-
weighted MR images. NeuroImage: Clinical, 2, 282–290. https://pubmed.ncbi.nlm.
nih.gov/24179783/
Dzialowski, I., Hill, M. D., Coutts, S. B., Demchuk, A. M., Kent, D. M.,
Wunderlich, O., & von Kummer, R. (2012). Extent of Early Ischemic Changes
on Computed Tomography (CT) Before Thrombolysis Prognostic, Value of the
Alberta Stroke Program Early CT Score in ECASS II. Stroke, 37(4),973-978. https://
doi.org/10.1161/01.STR.0000206215.62441.56
Ford, A. L., An, H., Vo, K. D., Lin, W., & Lee, J.-M. (2012). Dening the Ischemic
penumbra using hyper active neuroimaging: driving quantitative Ischemic
Thresholds. Translational Stroke Research, 3(2), 198-204. https://pubmed.ncbi.nlm.nih.
gov/24323775/
Grady, L. (2006). Random Walks for Image Segmentation. IEEE Transactionson Pattern
Analysis and Machine Intelligence, 28(11), 1768–1783. https://ieeexplore.ieee.org/
document/1704833
Grady, L., & Schwartz, E. L. (2006). Isoperimetric Graph Partitioning for Image
Segmentation. IEEE Transactionson Pattern Analysis and Machine Intelligence, 28(3), 469–
475. https://ieeexplore.ieee.org/document/1580491
Hema, N., & Bhavani, R. (2013). Computer aided detection of ischemic stroke using
segmentation and texture features. Measurement, (46), 1865-1874. https://www.infona.
pl/resource/bwmeta1.element.elsevier-6cea82e0-3e95-37c1-81e7-b48eb7905e72
Kharat, K. D., & Kulkarni, P. P. (2012). Brain Tumor Classication Using Neural
Network Based Methods. International Journal of Computer Science and Informatics
152
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ISSN (PRINT), 1(4), 2231–5292. https://www.interscience.in/cgi/viewcontent.
cgi?article=1075&context=ijcsi
Mahmoud-Ghoneim, D., Toussaint, G., Constans, J.-M., & Certaines, J. D. de.
(2003). Three dimensional texture analysis in MRI: a preliminary evaluation in
gliomas. Magnetic Resonance Imaging, 21(9), 983-987. https://doi.org/10.1016/S0730-
725X(03)00201-7
Rekik, I., Allassonnière, S., Carpenter, T. K., & Wardlaw, J. M. (2012). Medical
image analysis methods in MR/CT-imaged acute-subacute ischemic stroke lesion:
Segmentation, prediction and insights into dynamic evolution simulation models.
NeuroImage: Clinical, 1(1), 164–178. https://doi.org/10.1016/j.nicl.2012.10.003
Runchey, S., & McGee, S. (2010). Does This Patient Have a Hemorrhagic Stroke?.
JAMA: the journal of the American Medical Association, 303(22), 2280-2286. https://doi.
org/10.1001/jama.2010.754
Shi, J., & Malik, J. (2000). Normalized Cuts and Image Segmentation. IEEE
Transactionsonpatternanalysis and machine intelligence, 22(8), 888-905. https://ieeexplore.
ieee.org/document/868688
Thom, T., Haase, N., Rosamond, W., Howard, V. J., Rumsfeld, J., Manolio, T.,
Zheng, Z.-J., Flegal, K., O’Donnell, C., Kittner, S., Lloyd-Jones, D., Go,
D. D., Hong, J., Adams, R., Friday, G., Furie, K., Gorelick, P., Kissela,
B., Marler, J., … & Wolf, P. (2006). Heartdisease and stroke statistics update: a
report from the American Heart Association Statistics Committee and Stroke
Statistics Subcommittee. Circulation, 113(6), 85–151. https://pubmed.ncbi.nlm.nih.
gov/16407573/
Toni, D., Iweins, F., von Kummer, R., Busse, O., Bogousslavsky, J., Falcou, A.
E., Lesare, E., & Lenzi, G. L. (2006). Identication of lacunar in farcts before
thrombolysis in the ECASS 1 study. Neurology, 54(2),684–688. https://pubmed.ncbi.
nlm.nih.gov/10680804/
153
https://doi.org/10.17993/3ctecno.2021.specialissue8.125-153
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Udupa, J., & Samarasekera, S. (1996). Fuzzy Connectedness and Object Denition:
Theory, Algorithms, and Applications in Image Segmentation. CVGIP Graph.
Model. Image Process., 58, 246-261. https://www.semanticscholar.org/paper/Fuzzy-
Connectedness-and-Object-Denition%3A-Theory%2C-Udupa-Samarasekera/4a
3e9fcfc88d53cdf0b7633fbc914f98d1e92225
Von Kummer, R. (2005). The impact of CT on acute stroke treatment. In P. Lyden (ed.),
Thrombolytic Therapy for Stroke. Humana Press, pp. 249–278.
