Measurement: Sensors 24 (2022) 100440

Contents lists available at ScienceDirect

Measurement: Sensors
journal homepage: www.sciencedirect.com/journal/measurement-sensors

Brain tumor segmentation of MR images using SVM and fuzzy classifier in

machine learning
Ramdas Vankdothu *, Mohd Abdul Hameed
Department of Computer Science & Engineering, University College of Engineering(A), Osmania University Hyderabad, India

A R T I C L E I N F O A B S T R A C T

Keywords: Medical image processing is a rapidly growing and concentrating topic today. Medical image analysis techniques
Brain tumour are used to diagnose and cure illnesses. One such fundamental and potentially fatal illness is brain tumor, which
Segmentation is an abnormal growth of brain cells within the brain. Due to the complexity of the brain’s anatomy. To improve
SVM
efficiency and reduce the complexity of the picture segmentation process, this work investigated computer to­
Fuzzy classifier
Magnetic resonance imaging(MRI)
mography (CT)-based brain tumor segmentation. CT scans are often used to diagnose head traumas, malig­
nancies, and skull fractures. The images from the brain tumor database are evaluated in this study effort, and a
preprocessing approach called adaptive median filter is used to increase the image’s clarity. The preprocessing
stage eliminates noise and high-frequency artifacts from the pictures. The median filter is a type of nonlinear
digital filter commonly used to reduce noise in a picture or signal. Regardless of the preprocessing technique
used, feature extraction techniques are updated, and then classification procedures such as the Adaptive Neuro-
Fuzzy Inference System (ANFIS) and Support Vector Machine (SVM) classifier are applied to the picture to
classify it as normal or abnormal. Following classification, aberrant images are tracked and selected for seg­
mentation using the Fuzzy C-Means (FCM) clustering technique and associated optimization techniques. In the
suggested technique, centroid optimizations such as Grey Wolf Optimization (GWO) and Social Spider Optimi­
zation (SSO) combined with a Genetic Algorithm (GA) are utilized to improve the accuracy of the FCM centroid.
The suggested work produces the most extreme execution in tumour picture segmentation evaluation appears
differently from other works. The conclusion indicates that the hybrid technique (SSO-GA) obtains the highest
accuracy of 99.24% compared to other individual algorithms. MATLAB 2014 is utilized to implement the brain
tumor classification and segmentation algorithms in this research effort.

1. Introduction problem, increasing the amount of useful information available to the

medical specialist who interprets it (Antoine Maintz and Max Viergever
Medical image processing is a cutting-edge technological innovation 1998). An MRI scan uses a strong magnetic field to detect radio fre­
in today’s research period. Noninvasive solutions can be determined quency pulses and produce comprehensive pictures of internal organs,
with the use of medical imaging. Medical imaging technologies that soft tissues, bone, and other structures [4,5].
employ a noninvasive manner include PET, CT scan [1,2], Ultrasound, Image capture, preprocessing, image processing (which includes
SPECT, MRI scan, and X-ray. In the field of medical diagnostics, mag­ classification and segmentation), and presentation make up an image
netic resonance imaging (MRI) produces superior outcomes than processing system (Ao Qian, Chen Rongguan, Ning Ning, Sun Guozhi,
computed tomography. Imaging improves the contrast between Zhang Xiaochen, Zheng Jie, and Kong Deyu2012). Image processing is a
different soft tissues of the human body [3]. technique for converting an image to digital form and performing de-
The most basic duty of digital image processing in medicine is source noising operations on it to create a better image or extract important
image augmentation. Images from various medical devices (X-ray, MRI, information. It’s a signal distribution method in which the input is an
X-Ray based CT) are typically blurry and contain noise; this noise is image, such as a video frame or a picture, and the output is an image or
filtered using low-level image processing algorithms (denoising, sharp­ image-related feature. In most Image Processing systems, pictures are
ening, edge detection) with parameter values tweaked for the specific treated as two-dimensional signals and signal processing algorithms are

* Corresponding author.
E-mail address: [email protected] (R. Vankdothu).

https://doi.org/10.1016/j.measen.2022.100440
Received 31 July 2022; Accepted 30 August 2022
Available online 28 September 2022
2665-9174/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

applied to them [6,7]. or spread to other sections of the body as cancer does. The prognosis for
The three phases that comprise image processing are as follows: benign tumours is generally favorable. A brain tumor is the growth of
abnormal cells in the brain’s tissues. Brain tumours may be benign (no
i. Optical scanning or digital photography to import the picture cancer cells) or malignant (fast-growing cancer cells). Some of them are
ii. Picture analysis and manipulation, which includes data primary brain tumours, meaning they begin in the brain. Other cancers
compression and image augmentation, as well as denoising al­ are secondary or metastatic, meaning they begin elsewhere in the body
gorithms for finding patterns that are not obvious to the naked and spread to the brain. Although the causes of brain cancer are difficult
eye. to determine, it is recommended that substances connected to cancer
iii. Output is the last step, when the picture or report based on image formation be avoided. Sleepiness, convulsions, disorientation, and
analysis may be updated. behavioral abnormalities are all possible indications of brain cancer cells
[12].

1.1. Purpose of image processing 1.3.1. Primary brain tumours

Primary brain tumours arise in the brain or nearby tissues, such as
Image processing may be classified into five categories. To develop the meninges (brain covering membranes), cranial nerves, pituitary
better images, a visualization technique is first used to observe items gland, or pineal gland. Primary brain cancers develop when normal cells
that are not visible, followed by an image sharpening and restoration accumulate DNA mistakes (mutations).
procedure. The goal of image retrieval is to locate the picture of interest.
Pattern measurement is used to determine the size of distinct items in an 1.3.2. Secondary brain tumours
image. The goal of image recognition is to identify the items in an image Secondary (metastatic) brain tumours are tumours that develop
[8]. when cancer spreads (metastasizes) from another part of the body to the
brain. Malignant tumours are the other name for it.
1.2. Methods of image processing
1.4. MRI scans
There are two methods are available in image processing.
MRI is the most accurate method of imaging for diagnosing most
1.2.1. Analog image processing types of brain cancers. These scans provide comprehensive images of the
Analog image processing is the technique of altering an image via brain using magnetic fields and radio waves rather than X-rays and
electrical methods. The television image is the most frequent example. computers. Josefina Perlo, Christoph Mülder, Ernesto Danieli, Christian
The television signal is a voltage level that fluctuates in amplitude to Hopmann, Bernhard Blümich, and Federico Casanova (MRIs) display
indicate visual brightness. The look of the displayed image is changed by visual “slices" of the brain that may be joined to form a three-
electrically changing the signal. The brightness and contrast settings of a dimensional representation of the tumour [13]. To improve the scan
television set govern the amplitude and reference of the video signal, pictures, a contrast dye may be employed. In an MRI scan, one of the
causing the brightness range of the shown picture to brighten, darken, or intense magnetic field components is used to resolve radio frequency
change [9]. pulses and provide comprehensive images of organs, soft tissues, bone,
and other inside human body structures. MRI is a useful technique that
1.2.2. Digital image processing offers extensive information on the architecture of the targeted brain
Digital computers are used to process images in digital image pro­ tumor, allowing for accurate diagnosis, therapy, and monitoring of the
cessing. A scanner or digitizer will be used to convert the picture to illness. Its methods have been refined to offer measurements of change
digital format, which will then be processed. It’s described as putting inside and surrounding primary and metastatic brain tumours, such as
numerical meanings of things through a sequence of procedures to get edoema, volume deformation, and anatomic characteristics within tu­
the desired outcome. It begins with a single picture and generates a mours, among other things (Dina Aboul Dahab, Samy Ghoniemy and
changed version of it. As a result, it is a method of transforming one Gamal Selim 2012). A computerized representation of tissue charac­
picture into another. Digital image processing is the process of a digital teristics may be produced in any tissue plane using MRI. MRI scans may
computer processing a two-dimensional picture. It refers to the digital be used to complete the brain mass detection process. For medical image
processing of any two-dimensional data in a larger sense. A group of real processing, image processing and image enhancement technologies are
numbers is represented by a limited amount of bits in a digital picture. employed to increase picture quality. For emphasising the characteris­
The reproducibility, adaptability, and retention of original data accu­ tics of MRI pictures, contrast adjustment and threshold approaches are
racy are the key benefits of digital image processing systems [10]. applied [19,20].

1.3. Brain tumours 1.5. Image classification techniques

Medical science is the discipline of science that studies the inside 1.5.1. SVM
workings of the human body. For the examination of illness within the SVM is a two-class classification algorithm. One vs rest classifiers are
human body, doctors employ several imaging techniques. Each imaging employed to create the multi-class SVM, and the one that separates the
modality has its own set of capabilities and limitations when it comes to test data by a significant margin is chosen. The characteristics extracted
human anatomy. Brain tumours are one of the most life-threatening from each fingerprint picture are sent into the multi-class SVM trainer,
disorders among various complicated diseases [11]. A tumour is a which builds an SVM structure. The structural data is then passed into
term used in medicine to describe abnormal cell proliferation inside the the SVM classifier, together with the test datum, to provide a categorized
human body. Tumours are classified as benign or malignant based on fingerprint result. In general, an error-correcting output codes classifier
their pace of growth and other clinical characteristics. (2104) Vidyarthi is meant to assist in the reduction of numerous, binary classifiers,
and Mittal Malignant tumours are malignant because they are the most making it beneficial for multiclass learning, and SVM is one of them. So,
likely to cause death in humans. This form of tumour spreads quickly first and foremost, the aforementioned classifier must be trained using
throughout the human body, affecting vital organs. A tumour is a whatever training data is available. The trained classifiers, which are
non-functional abnormal development of cells. A benign tumour is not made up of the training data, are then employed in the classification
the same as a cancerous tumour. It doesn’t infiltrate neighboring tissue process to determine the new test datum’s class. Furthermore, our error-

2
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

correcting output code classifier simplifies multi-class classification order and size of anatomy areas. The proposed method was tested for
problems to simple binary classifier [16]. whole-body (torso) and partial-body MRI tests and found to be effective
[15].
2. Literature survey The super resolution techniques for medical image processing were
explained by (Jithin Saji Isaac and Ramesh Kulkarni 2015). Medical
A medical image processing technique for cancer detection (Priya imaging is a critical diagnostic tool for determining the existence of
Darshini Velusamy and Porkumaran Karandharaj 2014). This research specific disorders. Due to imaging surroundings, physical imaging sys­
delves into the specifics of many methodologies that show how hybrid tem restrictions, and quality limiting variables such as noise and blur,
intelligence approaches were applied to various cancer diagnosis and obtaining an image at the appropriate resolution proved difficult. Super
therapy categories. The three major types of cancer concerns, such as resolution techniques can be used to process such photos to avoid this.
breast, liver, and brain tumours, were thoroughly discussed using cancer As a result, increasing picture resolution should greatly increase the
databases. Field Programmable Gate Array (FPGA) based 3-D Ultra­ ability to diagnose and prescribe corrective treatment. Furthermore, a
sound Computer Tomography with image reconstruction algorithms higher resolution could significantly improve automatic detection and
was employed to diagnose malignancies. Finally, it demonstrated that recognition [17].
an overview of various states of cancer diagnosis and detection tech­
nologies is available [14]. 3. The proposed methodology
The brain, according to B.Srikanth et al., [2021], is a collection of
white masses of cells that make up the human body’s central nervous Image processing has turned into a basic component in many fields of
system (CNS). A brain tumor is a sort of abnormal cell development that biomedical research as well as clinical practice. Medical image pro­
happens when cells multiply rapidly and may lead to death if not cessing has achieved the gigantic place in diagnosing and segmenting
detected early. The most common types of brain tumours include me­ brain tumours after the revelation of Computed Tomography (CT). The
ningioma, glioma, and pituitary tumours [22]. Physicians and clinical segmentation of brain tumours from CT image is a testing and trouble­
experts have a challenging challenge in manually recognizing and some assignment as a result of the assortment of their conceivable
classifying brain tumours, and their accuracy is depending on their shapes, areas, and image intensities. This section explores the brain
expertise. Computer-assisted technologies must be used to overcome tumor image investigation because of feature extraction, and classifi­
these limitations. As medical standards improve, artificial intelligence is cation along with segmentation strategies with the enlivened optimi­
becoming more effective in detecting and classifying brain tissues. Re­ zation techniques.
searchers are quite interested in this. It’s being constructed right now
using a range of machine learning and deep learning architectures. 3.1. Brain tumor analysis methodology
Several machine learning algorithms have been employed in the past to
detect tumours and categorize them using MR images, including SVM, A brain tumor is an exceptionally unsafe illness for individuals. It is
KNN, and CNN. These algorithms, however, do not offer the best results. an intracranial mass made up by abnormal development of tissue in the
As a result, we’ve proposed a VGG-16 network built on a DNN to brain or else in the region of the brain. A powerful method is used to
improve the accuracy of multi-classification of brain tumor MR images. characterize images as nontumor brain or tumor brain. The segmented
Finally, after 20 epochs of model training, our recommended model tumor part from the CT scan Brain images is used in this strategy. This
gives the best results, with an accuracy of 98% [18]. methodology includes.
To identify brain cancer from MRI scans, Raheleh Hashemzehi et al.,
[2020] suggested a hybrid paradigm consisting of a Convolutional 1 Pre-processing
Neural Network (CNN) and a Neural Autoregressive Distribution Esti­ 2 Feature extraction
mation (NADE). They automatically extracted the attributes and esti­ 3 Classification
mated the data distribution. The approach looked at 3064 CE-MRI 4 Segmentation
pictures from 233 people, including 1426 glioma photos, 708 menin­
gioma images, and 930 pituitary tumour images. They proved that their At first, the pre-processing procedure will be established over the CT
proposed method has a classification accuracy of 95%. 3) More discus­ brain images for upgrading the precision of the image. From that point
sions of the results should be provided [23]. onward, the extraction of features from the improved image is trailed by
Brain illnesses such as brain tumours and multiple sclerosis, which the ideal elements utilized to prepare the classifier. The reason for the
may be diagnosed using the MRI technique, have been advocated by classification procedure is to classify the normal and abnormal images
Halimeh et al., [2019] as an important strategy to identify it. Brain from the CT brain image in light of the threshold esteem.
tumor sickness occurs when abnormal brain cells grow out of control. M. After the classification steps, the brain tumor part comes under
S. is a long-term illness that disrupts the brain’s neurological system. segmentation. The clustering process alongside the enhancement strat­
Magnetic resonance imaging (MRI) is the most effective way to identify egy is utilized. These enhancement methods are utilized to enhance the
and diagnose multiple sclerosis (MS) and malignancies (MRI). Because segmentation exactness furthermore optimal centroid is found to get the
cancers and multiple sclerosis are so similar, scientific mistakes in exact tumor part of the brain image.
diagnosis might result in patient harm or even death. A convolutional
neural network was used by the authors to diagnose a brain tumor and 3.1.1. Pre-processing
multiple sclerosis at the same time. As a consequence, scientists were Medical images are inclined to unimportant and inconsistent bois­
able to diagnose cancers and M.S. with 96% accuracy [24]. terous information. De-noising and image determination improvement
Márton Tóth, László Ruskó, and Balázs Csébfalvi (Márton Tóth, strategies assume a fundamental part in the accomplishment of high
László Ruskó, and Balázs Csébfalvi 2016) proposed an approach for image quality (Meenakshi Sundaram et al., 2014). The adaptive median
detecting anatomical regions in three-dimensional medical images. Each filter is used to improve the filter’s adjustability to convert its size as
axial slice of the image is labeled with the anatomy region to which it needs are discovered using the local noise density measure. The adaptive
corresponds using this procedure. The head (and neck), chest, belly, median filter is based on a trans-conductance comparator, in which the
pelvis, and legs were the areas that were discovered. To establish an drenching current may be controlled as a local weight operator (Vicky
initial slice-by-slice categorization, standard image processing tech­ Ambule et al., 2013).
niques and machine learning tools were used. To avoid incorrect clas­ The median filter is one of the well-known strategies to be utilized to
sifications, the method’s post-processing section combines the expected diminish motivation noise levels from adulterated images. The ordinary

3
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 1. Block diagram of Segmentation Process.

Fig. 2. Brain tumor analysis based on Fuzzy methodology.

median filter is not able to separate between noise free and noisy pixels. 3.2. Grey-level run-length matrix (GLRLM)
If the image is boisterous and is the estimated pixels then the neigh­
boring pixel values in and around the image are someplace around 0’s The Grey Level Run Length technique is a method for extricating
and 255’s in this filter, then this work undermines pixel view with the higher request measurable texture data. The quantity of dark levels G, in
adaptive median filter. the image is frequently lessened by re-quantization preceding the ag­
gregation of the matrix (Wang Xinli, 1994). A matrix form of quality
3.1.2. Feature extraction features that might be extracted for texture procedure is called as
The feature extraction approach is used to extract specific features GLRLM. A texture comprehension is a holder of grey force pixel in a
from CT brain images after image pre-processing methods have been particular heading from the reference pixels. The amount of the close-by
applied. In pattern recognition and image processing, feature extraction pixels that have a similar grey force in a particular heading is named as
is a unique sort of dimensionality reduction. When an algorithm’s input Run length. Grey level run length matrix is a two dimensional matrix
data is too large to be prepared in any manner and is suspected of being where each constituent z(u, v/Ɵ) is the number of elements v by means of
notoriously repetitious, the input data is transformed into a reduced the intensity u in the directionƟ. The Grey Level Run Length matrix is
representation set of (features vector) (Sudipta Roy, 2016). The brain CT produced as, GLRLM achieves more data by utilizing the grey intensity
scans are examined as part of the feature extraction technique. The level value co-props equity. These statistics give knowledge about the
feature extraction is passed on to the ghostly inspection strategy. Feature texture part of an image. These are as follows,
extraction infers the adjustment of input image values into a set of
features. In this investigation, the deliberated features are. ● Short Run Emphasis (SRE)
● Long Run Emphasis (LRE)
● Grey-Level Non uniformity (GLN)

4
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 3. Brain tumor analysis based on SVM with fuzzy methodology.

● Run Length Non uniformity (RLN) 4.1. Brain tumor analysis based on fuzzy methodology
● Run Percentage (RP)
The block diagram of the proposed techniques is provided in Fig. 2
4. Classification and Fig. 1. To begin with, the pre-processing step where the adaptive
median filter is applied over the CT brain images to remove the errors
After the above feature extraction procedure, the inventive classifi­ and enhance the purity of the input brain image. The feature extraction
cation approach is used to recognize the tumor part. Image classification might be completed if the extracted features from the image are used to
sorts recognized images into predefined classes by utilizing a reasonable train by the ANFIS classifier which classifies the normal and abnormal
technique that contrasts the image designs and the objective examples. image (Minakshi Sharma et al., 2012).
After extricating features, the classification approach is associated to
distinguish the normal and abnormal images. Here, consider two brain 4.1.1. Classification using adaptive neuro-fuzzy inference system (ANFIS)
tumor investigation philosophies because of various classification and classifier
segmentation methods. The systems are. Adaptive Neuro-Fuzzy Inference System (ANFIS) is an extremely well
known strategy that incorporates advantages of both methods, they are
A. Brain tumor analysis based on Fuzzy methodology named underneath:
B. Brain tumor analysis based on SVM with fuzzy methodology
● Artificial Neural Network
● Fuzzy Inference System

5
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

4.1.1.1. Artificial neural network (ANN). An ANN is a computational 4.1.1.3. Adaptive Neuro-Fuzzy Inference System (ANFIS) classifier.
structure that is roused by an observed process in common systems of ANFIS is an adaptive network. An adaptive network is a network of
natural neurons in the brain. Neural systems are normally organized in nodes with directional connections. It is named adaptive because a few
three layers in which various interconnected nodes contain an ‘activa­ or the majority of the nodes have parameters that influence the output of
tion function’ (Deepak et al., 2009). Every neuron applies an activation the node. These networks are taking in the connection amongst inputs in
function to its net contribution to decide its output signal. The artificial addition to outputs. The ANFIS approach takes in the standards and
neural network has three layers, for example, the Input layer, the Hidden membership functions from the information. The classification is made
layer and the Output layer. The architecture of ANN Input patterns are in two essential stages, to be specific, the training stage and the testing
presented to the system through the input layer, which conveys to at stage. In the training stage, the classifier is prepared with the extricated
least one hidden layer where the real processing is done using an features of the training information.
arrangement of weighted associations. The hidden layers are related to
an output layer where the required output is acknowledged. 4.2. Brain tumor analysis based on support vector machine (SVM) with
ANN can recognize and learn associated patterns amongst input and fuzzy methodology
compare target values. ANN emulates the learning procedure of the
human brain and can handle issues including non-straight and complex The block diagram of the research work is given in Fig. 3. For better
information regardless of the possibility that the information is impre­ classification results we have to move on to the SVM classifier with a
cise and noisy. fuzzy methodology. Once the feature extraction process is over the
collected features are utilized to train by the SVM classifier.
4.1.1.2. Fuzzy Inference System. A fuzzy neural network or neuro-fuzzy
framework is a learning machine that finds the constraints of a fuzzy 4.2.1. SVM
framework (i.e., fuzzy sets, fuzzy standards) by misusing estimation It is a supervised learning technique from the field of machine
procedures from neural networks. Fuzzy inference is the way toward learning material to both classification and regression. The target of the
detailing an input fuzzy set guide to an output fuzzy set utilizing fuzzy SVM approach is to discover a hyperplane that boosts the geometric
rationale. Gradient descent and Backpropagation calculations are uti­ edge and limits the classification mistake when given a two-class line­
lized to alter the parameters of membership functions (fuzzy sets) along arly distinguishable issue (Jing Xia 2010). SVM is viewed as a valuable
with the weights of defuzzification (neural networks) for fuzzy neural system for nonlinear classification. With a specific end goal to play out
networks (Anant Bhardwaj et al., 2013). the non-direct process, the kernel functions are initiated in the SVM
The essential structure of a fuzzy inference framework is promoted in classification. The two key unimportant stages of the SVM procedure are
Fig. 3.4. The framework changes over the crisp input to a linguistic the training stage and the testing stage. Given an arrangement of pre­
variable utilizing the membership functions put away in the fuzzy in­ paring illustrations, each set apart as having a place with one of two
formation database. It contains three stages that progress the framework classes, an SVM training algorithm manufactures a model that relegates
inputs to the fitting framework outputs. These steps are. new cases into one classification or the other.

1. Fuzzification, 4.2.2. Image segmentation using fuzzy C-means clustering

2. Membership and Rule generation Image segmentation is a significant instrument for decreasing the
3. Defuzzification data in image processing. The goal of image segmentation is to isolate
significant data from irrelevant data to accomplish the information
Database: A database that characterizes the membership functions diminishment expected to enhance the interpretation procedure (Tito
of the fuzzy sets utilized as a part of the fuzzy guidelines. Amaral1998). By and large, the image segmentation methodologies can
Fuzzification: The way toward changing crisp input values into be separated into four classifications: thresholding, clustering, edge
linguistic qualities is called fuzzification and it includes two procedures. detection, and region extraction. Here, a clustering based Fuzzy C-Means
To start with, the input qualities are converted into linguistic ideas (FCM) strategy will be considered for image segmentation and the
spoken to by fuzzy sets. Linguistic variables are the input or output relating segmentation process is exposed.
variables of the framework whose qualities are from a characteristic
dialect, rather than numerical qualities. At that point membership 4.2.3. Fuzzy C-means (FCM) clustering
functions are connected to the estimations and the level of truth in each In image segmentation, the FCM system is an unconfirmed method of
introduction is resolved (Burcin Cem Arabacioglu et al., 2010). clustering. The FCM is here and largely used for clustering where the
Membership and rule generation: Membership functions are utilized experiment of the FCM depends upon the assurance of beginning cluster
as a part of the fuzzification and defuzzification ventures of a FIS, to center or enrollment incentive to the tumor image pixel values (Khali­
outline non-fuzzy input qualities to fuzzy linguistic terms and the other quzzaman 2016). The FCM calculation starts with a game plan of
way around. A membership function is utilized to evaluate a linguistic initialized cluster centers (or) arbitrary membership values in the given
term. The most well known sorts of membership functions are trian­ image. To upgrade the cluster center of the tumor part of the image using
gular, trapezoidal along with Gaussian shapes. an optimization strategy [21]. In the segmentation technique, the
For considering rule generation in a FIS, a rule base is assembled to distinguished tumor image is allocated as the contribution to the seg­
control the output variable. A fuzzy rule is a simple IF-THEN rule using a mentation procedure to constrain the tumor part of the image, which is
condition as well as a conclusion. The estimations of the fuzzy rules and appeared. The primary target of iterative clustering and fuzzy c-Means
the permutation of the consequences of the individual regulations are calculation is to limit the weight age inside the gathering entirety of the
performed with fuzzy set operations. squared error objective function (Benson et al., 2016).
Defuzzification: On the off chance that a crisp estimation of the
framework is required, the last fuzzy output must be defuzzified. This is 4.2.4. Optimization algorithm
the motivation behind the defuzzifier segment of a FLS. Defuzzification Optimization alludes to the way toward choosing the best arrange­
is performed by the membership function of the output variable. This ment from the arrangement of every single conceivable answer for
can be used by different techniques like gravity, the bisector of area, maximizing or minimizing the cost of the issue. This procedure is
mean of maximum, smallest of maximum, and largest of maximum. conceded to accomplish the segmented image and to accomplish the
centroid optimization of the tumor image with the most extreme pre­
cision. The proposed optimization procedures are.

6
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 4. Segmentation process.

Fig. 5. Flowchart of GWO

Fig. 6. Sample CT Brian image database.

1. Grey Wolf Optimization 4.2.4.1. Grey Wolf Optimization. Grey Wolf Optimization (GWO) is a
2. Social Spider Optimization (SSO) technique with Genetic Algorithm recently presented developmental calculation, which recommends that
(GA) the grey wolves have a flourishing generation more than hunting in the
pack.
The grey wolf begins the relocation operation by picking one pack

7
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 7. Comparative analysis for Validation 1.

that has the best fitness esteem, called the best pack. Alternate packs are solve constrained optimization problems also to attain the maximum
moved to the best pack. In each pack (aside from the best pack), the accuracy of the segmented tumor part of the image.
number of wolves that have moved ought to be unequal. Grey wolves
that have the fitness esteem superior to anything the normal fitness 4.2.6. Optimal centroid to the FCM
estimation of their pack would move to the best pack. After the migra­
tion operation, the best pack would have more individual wolves so the 1. In the segmentation process, the best values obtained from different
determination operation is performed to lessened wolf measures of the runs of FCM turn to be optimal centroid of FCM.
best pack. In the choice operation, the fitness estimation of every indi­ 2. Finally utilizing the optimal centroid leads to extracting tumor part
vidual is assessed and sorted by ascending (minimize function). The of the brain images in CT scan.
chosen individual is browsed the grouping of fitness esteem. At that 3. The optimal centroid of FCM achieves maximum accuracy of the
point alternate packs would haphazardly produce new individuals for proposed work.
supplanting to the resettlement wolf. The accompanying Fig. 5 demon­
strates the graphical perspective of GWO (see Fig. 6). 5. Result and discussion

4.2.5. Limitations of first work (ANFIS classifier, FCM with GWO This area talks about the trial consequences of the classification
optimization) process done on the brain CT images. In the CT images, the tumor and
the non-tumor parts are grouped by using the Adaptive Neuro-Fuzzy
1. In the training process, time consumption is the major disadvantage Interface System (ANFIS) and Support Vector Machine (SVM) classi­
of ANFIS while choosing the type of network architecture. fier. The proposed technique helps the best classification of the tumor
2. SVMs convey a unique solution since the optimality issue is curved. and the non-tumor parts. Once the classification is done, the threatening
This is a favorable position contrasted with Neural Networks, which brain images are additionally portioned for extraction of tumor area
have numerous solutions related to local minima and hence may not from these brain CT images.
be vigorous over various examples. The execution of every classifier is measured as far as sensitivity,
3. The standard Grey Wolf Optimization (GWO) algorithm has some specificity, and accuracy. Sensitivity is a measure that decides the
drawbacks of low precision, slow convergence, along with bad local probability of the outcomes that are truly constructive to such an extent
searching ability. that an individual has the tumor. Specificity is a measure that decides
4. To overcome these disadvantages of GWO, Social Spider Optimiza­ the probability of the outcomes that are true negative with the end goal
tion (SSO) technique with Genetic Algorithm (GA) is proposed to that the individual does not have the tumor. Accuracy is a measure that
decides the probability that how many outcomes are precisely classified.

8
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 8. Comparative analysis for Validation 2.

5.1. Database description TP: tumor part correctly marked as tumor TN: normal area correctly
unmarked as tumor.
This dataset included various slice thicknesses, noise levels, and in­ FP: normal area wrongly marked as tumor FN: tumor area wrongly
tensity non-uniformity levels. Weighted modality pictures with 1 mm unmarked as tumor.
slice thickness, 3% noise, and 20% intensity non-uniformity were used
in our research. The most recent dataset compiled by professional ra­ 1. Classification Results for SVM Algorithm
diologists included pictures from all modalities. This dataset included
ground truth photographs that compared the results of our method
against radiologists’ manual inspection. The data for this suggested Image TP TN FP FN Sensitivity Specificity Accuracy
tumour detection approach came from publicly available sources such as 1 10 3 7 7 0.59 0.30 0.48
(M/s Aarthi checks, Tirunelveli, India). This image collection comprises 2 5 3 9 1 0.333333 0.25 0.30
brain CT pictures, which include tumour images (150 numbers) as well 3 9 4 8 6 0.6 0.33 0.48
as non-tumor images (60 numbers), as shown in Fig. 4.1. The data set of 4 15 6 4 2 0.882353 0.60 0.78
5 7 5 7 8 0.466667 0.42 0.44
brain pictures is partitioned into two portions, example, the training
6 4 8 7 8 0.333333 0.53 0.44
dataset and the testing dataset, and all test images are produced on 512 7 7 5 7 8 0.466667 0.42 0.44
× 512 matrixes with 256 unique grey levels. 8 19 3 3 2 0.90 0.50 0.81
9 8 7 9 3 0.73 0.44 0.56
10 7 3 15 2 0.78 0.17 0.37
5.2. Classification validation results 11 5 10 12 0 1.00 0.45 0.56
12 6 6 11 1 0.86 0.35 0.50
The Adaptive Neuro-Fuzzy Interface System (ANFIS) and Support 13 9 5 8 5 0.64 0.38 0.52
Vector Machine (SVM) strategies are used to assess the CT Brian tumour 14 9 5 5 8 0.53 0.50 0.52
15 10 3 11 3 0.77 0.21 0.48
categorization procedure. The complete amount of data is divided into
subgroups using this method. One subset is expired in each phase, and
the classifier is ready to use the remains. To approve the analysis, the The SVM algorithm’s rule is to use a function called the SVM’s kernel
classifier is then applied to the left out subgroup. This technique is function to convert a nonlinear separation goal into a linear trans­
repeated until each subgroup has been exhausted. They are divided into formation. The nonlinear specimens may be transformed into a high-
True Positive (TP), True Negative (TN), False Positive (FP), and False dimensional future space using a kernel function, where the division
Negative (FN) categories (FN). of nonlinear samples or data is conceivable, resulting in a classification
Here.

9
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 9. Comparative analysis for Validation 3.

that is supporting. SVM is the default choice for brain tumor classifica­ to hard segmentation by allowing pixels to be enrolled in clusters with
tion after features selection with kernel class separability. The accuracy, the highest extreme estimate of membership coefficients. The goal of
sensitivity, and specificity of the SVM algorithm may all be measured. clustering is to create decision boundaries based on unlabeled training
The above table shows the confusion matrix for the words TP, TN, FP, data. In multidimensional feature space, clustering is the process of
and FN from the normal and ground truth results for the calculation of identifying regular gathering groupings. Because clusters of varied
accuracy, sensitivity, and specificity. forms and sizes may occur in multidimensional feature space, it is
problematic.
2. Classification Results for Fuzzy Classifier Algorithm

5.3. Validation 1
Image TP TN FP FN Sensitivity Specificity Accuracy
In the classification process, primarily every classifier is tried for
1 1985 62237 1094 220 0.90 0.98 0.98 utilizing the first order feature set, mainly the performance of every
2 0 57472 1441 6623 0 0.98 0.88
classifier is measured for this feature set. In this chart, the execution of
3 0 52622 6262 6652 0 0.89 0.80
4 0 61495 1186 2855 0 0.98 0.94 every classifier is indicated when the first order histogram base feature
5 128 61495 1252 2661 0.045895 0.98 0.94 set is utilized for classification. Fig. 7 demonstrates the validation 1
6 1852 59643 1186 2855 0.393457 0.98 0.94 process and it shows the classification images and performance mea­
7 1244 60253 1185 2854 0.303563 0.98 0.94
sures sensitivity, specificity, and accuracy for various procedures like
8 0 61495 2458 1583 0.00 0.96 0.94
9 0 61495 1288 2753 0.00 0.98 0.94
ANFIS, Support Vector Machine(SVM), Backpropagation Neural
10 9643 249699 2589 213 0.98 0.99 0.99 Network (BNN) and Radial Bias Forward Neural.
11 17364 241943 2585 252 0.99 0.99 0.99 For that, the SVM is contrasted with another strategy the accuracy
12 6745 253632 1542 225 0.97 0.99 0.99 varies as 7.2–10% and other techniques likewise get some distinction as
13 6152 254002 1676 314 0.95 0.99 0.99
5.26%. These underlying validations process the three images as tu­
14 6521 253535 1776 312 0.95 0.99 0.99
15 5841 254213 1876 214 0.96 0.99 0.99 mours and stay two as non-tumor images.
The classifier BNN is more fruitful in delivering an accuracy rate that
is considerably more prominent than the current classifiers, for example,
The performance data for FCM without optimization are shown RBFN and ANFIS. The accuracy increases with the expansion in the
above. Fuzzy C-mean provides soft segmentation that may be converted number of features. This procedure advances with a high accuracy rate
in correlation with other accessible classification techniques. The

10
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 10. Comparative analysis for validation.

sensitivity increases with the expansion in the number of features. This parameter which likewise assumes a significant part in ANFIS classifier,
procedure turns out with more prominent sensitivity rates in contrast and a suitable smoothing parameter is frequently data subordinate. The
with other classification strategies. The principal validation process test execution of the SVM classifier is dictated by the calculation of the
achieved 100% accuracy when contrasted with different strategies. measurable parameters, for example, sensitivity, specificity, and accu­
racy in the examination with divergent classifier strategies. Validation 3
accomplishes better outcomes contrasted with validation 3. The most
5.4. Validation 2
extreme accuracy was 95.26% and 96.23% attained in sensitivity and
specificity as well as relative esteem. Besides, higher estimations of ac­
curacy and sensitivity and a lower estimation of specificity demonstrate
better execution.
5.5. Validation 3
5.6. Validation 4
Classification models are practically comparable as far as discrimi­
native performance and just demonstrate a minor distinction by
methods for measurable evaluations. Also, most essential, none of the
two models are fundamentally unrelated and are in certainty corre­
sponding to each other. Figs. 8 and 9 demonstrate the performance 5.7. Validation 5
metrics comparison of validation 2 and 3. These two validations in the
subsequent model effectively grouped as Brian tests from tumor to non- Figs. 10 and 11 depict the comparative analysis for validation 4 and
tumor with fine accuracies of 95.4%. Notwithstanding the ANFIS pro­ 5. The classifier of progressive classification structure, a multiclass SVM
cess, a classification model because of the ideal structure is likewise built is utilized to classify images. To assess our feature selection algorithm
for examination and assessment. The training procedure is utilized to and SVM classifier we play out the classification with three classifiers
process the dataset and the testing procedure is utilized to check the ANFIS, BNN, and RBFN in the wake of utilizing the feature selection
accuracy and effectiveness of the trained network for the classification algorithm. In the lion’s share of the validation forms, high accuracy is
of brain tumours. The input brain CT scanning images are grouped by acquired just in the linear kernel with the SVM system. Of the total, five
back propagation neural network. These support vector machines are validation forms that are considered in every validation we use 5 im­
trained to utilize linear kernel function. ages. In the third validation process, the direct SVM produces 100%
The classification performance accuracy of the testing dataset range accuracy, and when it is contrasted with the ANFIS and BNN the accu­
is 92.3–99%. An outcome of an expansive system structure is that the racy of SVM is 80% higher than different procedures. On the off chance
classifier has a tendency to be oversensitive to the preparation infor­ that the performance of polynomial bit with RBFN and proposed SVM is
mation and is probably going to display poor speculation abilities to the contrasted and the performance of ANFIS, then the deviation in accuracy
concealed data. The second issue is identified with the smoothing is 80%.

11
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 11. Comparative analysis for validation 5.

classification performance of the classifiers on the diagnosis of the tumor

Table 1
from brain CT image with feature extraction for normal and abnormal
Classification performance analysis.
images.
Original Images Classified image ANFIS SVM

Tumor Tumor 85.62 90 6. Segmentation results analysis

Tumor Normal 86.25 88
Tumor Normal 95.45 95.68
In the segmentation procedure, the tumor part is segmented by the
Tumor Tumor 98.2 97.2
Tumor Tumor 94.5 94
proposed strategy i.e. Fuzzy C means with centroid optimization uti­
Normal Normal 96.2 96 lizing GWO and SSO with GA. The segmentation procedure includes the
Normal Normal 94.25 95.2 original image, the classified tumor image, and the segmented image
Normal Normal 75.25 78.25 alongside the performance measures, for example, Sensitivity, Speci­
Normal Normal 92.2 93.2
ficity, Accuracy, Random Index, Global Contingency Error (GCE), and
Normal Normal 88.2 90.25
Variation of Information (VI). For the testing procedure, a portion of the
images are considered and regarding these images, the previously
The outcomes affirm the significance of CT body scans in the diag­ mentioned parameters are dissected.
nosis and administration of patients with secondary brain tumours. Table 2 demonstrates the segmentation investigation for FCM with
Nevertheless, in patients thought to have an essential brain tumor our different parameters, for example, accuracy, Random Index (RI), Global
review shows that CT body scans may add little to their administration. Consistency Error (GCE) and Variation of Information (VI). For the most
Sometimes regulated image segmentation and classification techniques part, Brain images are mind-boggling and its inclination is inhomoge­
turn out to be extremely costly, troublesome, and even difficult to neous, changing from that inhomogeneous into homogeneous is
accurately choose and name the training data with its true category. extremely troublesome. Without image segmentation, it is extremely
Training is the fundamental prerequisite of numerous SVM and ANFIS hard to envision the structure of the image for neurologists or radiolo­
based algorithms where the classifiers should be prepared before it can gists. Subsequently, the FCM segmentation strategy is used to section the
be connected to segmentation and classification issues. Encouragement, tumor image.
for various informational collections, investigation of various images of Table 3 speaks to the segmentation examination for FCM with GWO.
various sorts and organization, the entire exertion of choosing training The review manages the idea of automatic brain tumor segmentation.
data collection and training is required to be revamped. Regularly the anatomy of the Brain can be seen by the MRI scan or CT
Table 1 demonstrates the classification accuracy analysis with the­ scan. In this review, the CT scanned image is taken as the input image for
original image and classified images for both procedures. It is critical to the whole procedure. From the input image, the ground truth image is
note that as a portion of the features doesn’t add to the classification, it is accomplished which alludes to the information given by direct percep­
around 94% in an Adaptive Fuzzy Inference System (ANFIS), 96.2% in tion. From that point forward, the brain tumor part is divided by using
SVM with feature extraction. The table demonstrates the accuracy of the Fuzzy C-means clustering and it is enhanced by the Grey Wolf

12
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Table 2
Segmentation analysis for FCM.

13
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Table 3
Segmentation analysis for FCM with GWO.

optimization technique. image accomplishes 0.98 in accuracy, 0.96 in RI, 0.04 in GCE, and 0.40
The framework performance is broken down into the parameters, for in VI. Similarly, the examined parameters are registered for the other
example, accuracy, Random Index (RI), Global Consistency Error (GCE), brain images. Contrasted with the FCM segmentation investigation, FCM
and Variation of Information (VI). The table demonstrates the detailed with GWO accomplish better accuracy in the portioned tumor brain
investigation of 10 distinctive brain images. The principal segmented image.

14
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Table 4
Segmentation analysis for FCM with SSO with GA.

15
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 12. Comparative analysis.

Fig. 13. Comparative analysis for RI, GCE, and VI.

Fig. 14. Convergence graph analysis.

16
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Fig. 15. Sensitivity analysis.

Fig. 16. Specificity analysis.

Fig. 17. Accuracy analysis.

17
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

Table 4 demonstrates the hybrid approach of segmentation analysis delineates the accuracy of the segmented brain tumor part of the image
for FCM with the optimization methods SSO and GA. The table dem­ with different optimization procedures like GWO, GA, SSO, and SSO
onstrates the aftereffects of dissected parameters, for example, accuracy, with GA. The GWO esteem is contrasted with the hybrid approach
Random Index (RI), Global Consistency Error (GCE), and Variation of esteem, with a distinction is about almost 0.3% for all the six images.
Information (VI) are examined by using FCM segmentation with Social Considering the six images, the most extreme accuracy is achieved in the
Spider optimization and GA technique. The most noteworthy accuracy hybrid approach for the fifth image contrasted with different
0.99 is accomplished in the five brain images. Regarding all the factual methodologies.
parameters, the better outcome is delivered in the FCM –SSO and GA
contrasted with the FCM, FCM with SSO, and FCM with GA. 7. Conclusion
Promote, the GCE factor demonstrates the minimum error in the
recommended strategy. The random index of image 1 in the proposed The segmentation of brain tumor image from the input CT image in
technique is 0.96 and when contrasted with the FCM the distinction is the view of a few feature extractions, classification, and segmentation
0.60 and a comparable contrast is created in every one of the parame­ progression were examined alongside the optimization techniques.
ters. In like manner, the variety of information for image 1 in the pro­ Here, two unique sorts of classifiers, for example, ANFIS and SVM taken
posed hybrid approach is 0.39 and when contrasted with the FCM the VI after by FCM segmentation and two optimization algorithms like GWO
esteem is much high which prompts bringing down the accuracy of the and SSO with GA were examined to identify the tumor part of the image.
segmented image.
Funding details
6.1. Comparison analysis for different segmentation techniques
There are no funding details available.
Fig. 12 demonstrates the comparative examination of various pa­
rameters, for example, sensitivity, specificity, and accuracy for various Informed Consent
optimization procedures, for instance, FCM, FCM-GWO, FCM-SSO, FCM-
GA and FCM-SSO with GA. The imaging sensitivity, specificity, and There is no Informed Consent.
accuracy are low in the FCM investigation when contrasted with the
FCM with optimization strategies. FCM with individual optimization Author’s contribution
procedure gives better outcomes contrasted with FCM yet the coveted
ideal outcome is accomplished in the FCM with hybrid optimization The author ‘s declare no contribution.
only.
Fig. 13 clearly speaks to the near examination of parametric de­
velopments like random index, GCE and VI with different optimization Declaration of competing interest
systems, for example, FCM, FCM-GWO, FCM-SSO, FCM-GA, and FCM-
SSO with GA. In the FCM approach, the arbitrary file esteem is 0.46, GCE The authors declare that they have no known competing financial
is. interests or personal relationships that could have appeared to influence
0.01 and VI are 3.39. The VI measurements are too high in the FCM the work reported in this paper.
when contrasted with alternate methodologies which prompt the least
accuracy in the segmented image. The coveted ideal outcomes are References
accomplished in the FCM with a hybridization procedure where the VI is
[1] Hikmat Khan, Pir Masoom Shah, Munam Ali Shah, Saif ul Islam, Joel, J.P.
0.96 metrics, GCE is 0.03 and RI are 0.98. C. Rodrigues, Cascading handcrafted features and Convolutional Neural Network
for IoT-enabled brain tumor segmentation, Comput. Commun. 153 (2020)
6.2. Convergence graph analysis 196–207.
[2] Farman Ali, SM Riazul Islam, Daehan Kwak, Pervez Khan, Niamat Ullah, Sang-
jo Yoo, Kyung Sup Kwak, Type-2 fuzzy ontology–aided recommendation systems
Fig. 14 shows the convergence graph analysis for the optimization for IoT–based healthcare, Comput. Commun. 119 (2018) 138–155.
strategies, for example, GA, SSO, GWO and SSO with GA, and the chart is [3] Leandro Y. Mano, S. Bruno, Luis Faiçal, H.V. Nakamura, Pedro H. Gomes,
Giampaolo L. Libralon, I. Rodolfo, Meneguete, P.Rocha Geraldo, Filho, et al.,
drawn between fitness function (accuracy in the principal Figure and
Exploiting IoT technologies for enhancing Health Smart Homes through patient
sensitivity in the second Figure) and several iterations. identification and emotion recognition, Comput. Commun. 89 (2016) 178–190.
The fitness function ranges from 40 to 100 and the iterations from [4] Ikram Ud Din, Mohsen Guizani, Joel, J.P.C. Rodrigues, Suhaidi Hassan, V. Valery,
0 to 100. The GWO strategy gives the most minimal accuracy for each Korotaev, Machine learning in the Internet of Things: designed techniques for
smart cities, Future Generat. Comput. Syst. 100 (2019) 826–843.
number of iterations. The hybrid optimization SSO with GA gives the [5] Muhammad Usman Younus, Saif ul Islam, Ihsan Ali, Suleman Khan, Muhammad
most elevated accuracy when contrasted with the individual optimiza­ Khurram Khan, A survey on software defined networking enabled smart buildings:
tion systems. As far as the sensitivity likewise, ideal outcomes are ach­ architecture, challenges and use cases, J. Netw. Comput. Appl. 137 (2019) 62–77.
[6] Fizzah Bhatti, Munam Ali Shah, Carsten Maple, Saif Ul Islam, A novel internet of
ieved in the hybrid approach and the base sensitivity is accomplished in things-enabled accident detection and reporting system for smart city
the GA i.e. for 100th iteration the sensitivity rate is 78. environments, Sensors 19 (9) (2019) 2071.
Fig. 15 depicts the sensitivity examination of six diverse arrange­ [7] Pedro H. Vilela, Joel JPC rodrigues, petar solic, kashif saleem, and vasco furtado.
"Performance evaluation of a fog-assisted IoT solution for e-health applications,
ments of images with various optimization techniques, for example. Future Generat. Comput. Syst. 97 (2019) 379–386.
GWO, GA, SSO, and SSO with GA. The sensitivity esteem ranges from [8] Ankur Limaye, Tosiron Adegbija, HERMIT: a benchmark suite for the internet of
0 to 100. The sensitivity estimation of GWO is 93.8%, GA is 84%, SSO is medical things, IEEE Internet Things J. 5 (5) (2018) 4212–4222.
[9] P.G. Rajan, C. Sundar, Brain tumor detection and segmentation by intensity
80% and SSO with GA is 96% for the 1 image. So also, the sensitivity is adjustment, J. Med. Syst. 43 (8) (2019) 282.
examined for different images. The chart demonstrates the hybrid [10] V. Bhanumathi, R. Sangeetha, CNN based training and classification of MRI brain
approach accomplishes the most extreme sensitivity for all images. images, in: 2019 5th International Conference on Advanced Computing &
Communication Systems (ICACCS), IEEE, 2019, pp. 129–133.
Fig. 16 demonstrates the specificity examination of various brain
[11] Gregor Urban, M. Bendszus, F. Hamprecht, J. Kleesiek, Multi-modal brain tumor
images with four distinctive optimization methods, for example, GWO, segmentation using deep convolutional neural networks" MICCAI BraTS (brain
GA, SSO, and SSO with GA. The diagram is drawn between specificity tumor segmentation) challenge, Proc. win. contribut. (2014) 31–35.
which ranges from 0 to 100 and 6 sets of images. The most astounding [12] Darvin Yi, Mu Zhou, Chen Zhao, Olivier Gevaert, 3-D convolutional neural
networks for glioblastoma segmentation, arXiv preprint arXiv:1611.04534 (2016).
specificity esteem is gotten in image 3 contrasted with different images [13] Baris Kayalibay, Grady Jensen, Patrick van der Smagt, CNN-based segmentation of
and particularly in the SSO with GA optimization strategy. Fig. 17 medical imaging data, arXiv preprint arXiv:1701.03056 (2017).

18
R. Vankdothu and M.A. Hameed Measurement: Sensors 24 (2022) 100440

[14] Darshini Priya, Velusamy, Porkumaran Karandharaj, Medical Image Processing [19] Abdesslem Layeb, Nesrine Lahouesna, Bouchra Kireche, A multi-objective binary
Schemes for Cancer Detection: A Survey Medical Image Processing Schemes for cuckoo search for Bi-criteria knapsack problem‟, I, J. Inf. Eng. Electron. Bus. 4 (2)
Cancer Detection: A Survey”, 2014, pp. 1–40. (2013) 8–15.
[15] Márton Tóth 1, László Ruskó, Balázs Csébfalvi, 2016 “Automatic Recognition of [20] A. Ramaswamy Reddy, E.V. Prasad, Reddy, LSS„Abnormality detection of brain
Anatomical Regions in Three-Dimensional Medical Images”, vol. 35, 2016, MRI images using a new spatial FCM algorithm, Int. J. Eng. Sci. Adv. Technol. 2 (1)
pp. 1240–1251, 5. (2012) 1–7.
[16] Mohammad Havaei, Axel Davy, David Warde-Farley, Biard Antoine, [21] Abdenour Mekhmoukh, Karim Mokrani, Mohamed Cheriet, A modified Kernelized
Aaron Courville, Yoshua Bengio, Chris Pal, Pierre-MarcJodoin, Hugo Larochelle, Fuzzy C-Means algorithm for noisy images segmentation: application to MRI
Brain tumor segmentation with deep neural networks, Med. Image Anal. 35 (2017) images, IJCSI Int. J. Comput. Sci. Iss. 9 (1) (2020) 172–176.
18–31. [22] B. Srikanth, S. Venkata, Suryanarayana “Multi-Class classification of brain tumor
[17] Isaac Jithin Saji, Ramesh Kulkarni, Scaling up of low resolution images using super images using data augmentation with deep neural network”, Mater. Today Proc.
resolution techniques & performing intensity correction for medical imaging” (2021).
december, J. Biomed. Eng. Med. Imag. 2 (6) (2015) 1–40. [23] R. Hashemzehi, S.J.S. Mahdavi, M. Kheirabadi, S.R. Kamel, Detection of brain
[18] R. Khayati, M. Vafadust, F. Towhidkhah, S.M. Nabavi, 2008, „Fully automatic tumors from MRI images base on deep learning using hybrid models CNN and
segmentation of multiple sclerosis lesions in brain MR FLAIR images using adaptive NADE, Biocybern. Biomed. Eng. 40 (3) (2020) 1225–1232.
mixtures method and Markov random field model, Comput. Biol. Med. 38 (2008) [24] Sinar Halimeh, M.D. Teshnehlab, Diagnosing and Classification Tumor and M.S.
379–390. Simultaneous Magnetic Resonance Images Using Convolution Neural Network,
CFIS, 2019.

19