Global Emerging Innovation Summit (GEIS-2021), 2021, 125-137 125

CHAPTER 13
A Comparative Study on Various Machine
Learning Techniques for Brain Tumor Detection
Using MRI
Samia Mushtaq1,*, Apash Roy1 and Tawseef Ahmed Teli2
1
School of Computer Applications, Lovely Professional University, Punjab, India
2
Department of Computer Applications, Amar Singh College, J&K, India

Abstract: The brain is the central controlling system in the human body. If the
structure of the brain changes due to the enlargement of the brain cells, it is diagnosed
as a brain tumor, which can lead to fatality. The techniques such as medical imaging
provide evidence of whether a patient has a brain tumor or not. This paper discusses
various machine learning techniques for brain tumor detection using MRI and provides
a performance analysis of such methods based on the state-of-the-art. A comparison of
ANN and CNN-based models have also been given after implementing the techniques
in Tensorflow to understand the potential benefits of using deep learning-based
techniques.

Keywords: ANN, Brain Tumor, CNN, Deep Learning, Machine learning, MRI.

INTRODUCTION
Images obtained from MRI-Magnetic resonance imaging are utilized to obtain the
complete information of internal brain tissues. During the process of brain tumor
[1] investigation, detection of the tumor core location is the key task to get the
size and shape of the brain tumor. Segmentation techniques are vital in the
detection process of tumors in the brain. To obtain complete knowledge of brain
tissues like gray matter, cerebral spinal fluid, white matter, etc., segmentation
methods are applied. Segmentation can be done in two ways either automatically
or manually. Manual segmentation takes more time and requires expert
knowledge and experience but brings down computational efficiency while
Automatic segmentation works with histograms based on pixel intensities. The
assessment of lesions , using MRI image segmentation, in the brain of a patient
*
Corresponding author Samia Mushtaq: School of Computer Applications, Lovely Professional University, Punjab,
India; E-mail: [email protected]

Dharam Buddhi, Rajesh Singh and Anita Gehlot (Eds.)

© 2021 The Author(s). Published by Bentham Science Publishers
126 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

helps in providing significant data that in turn could help provide insights into the
evaluation and detection of various brain pathologies which could help in the
proper planning and monitoring of treatment. The segmentation incorporated with
human intervention can prove very beneficial in terms of accuracy but it is a very
cumbersome and expensive task that is also a time-consuming task. So, an
efficient and automated segmentation-based detection technique is the need of the
hour that may provide outcomes that could be significantly used for the
quantitative evaluation of tumors.
In neurological sciences, MRI segmentation of the brain is the key basic step and
has applications in functional imaging, operational planning, and quantitative
analysis [2]. Although the brain’s internal structure is accurately described by
MRI-imaging, the segmentation of medical images is a tedious task as a result of
various issues like noise, weakly defined boundaries, poor or weak spatial
resolution, inhomogeneity, low contrast, partial volume, variations in the shape of
objects, acquisition artifacts present in the fetched data and deficiency of anatomy
models that completely give details about the deformations of the internal
structures [2 - 4]. These issues can be solved by the implementation of some
simple techniques which are presented in [4 - 7]. As a result of issues like tumor
heterogeneity, instability in shape, size, tumor recurrence it is very complicated to
formulate the best efficacious segmentation technique [3].
Hybrid techniques are utilized to obtain gainful and reliable results in the
segmentation process by integrating more than one method, by utilizing their pros
and rejecting their cons. Parveen [8] proposed a hybrid technique in which both
SVM and FCM techniques are merged to obtain the correct classification of the
brain tumor region. For segmentation of affected brain core region, FCM is
utilized and in the process of classification, features are extracted using Gray level
run length matrix which is then utilized by SVM for the process of classification.
Region-based ACMs integrated with Machine learning are applied for fruitful
segmentation results by utilizing their pros and rejecting their cons. Intensity
homogeneities are dealt with by machine learning while issues of misclassified
and weakly defined boundaries are handled by region-based ACMs.
Many researchers have used supervised methods for brain tumor segmentation
technique that includes; Support Vector Machines (SVMs), Random Forests and
K-Means, etc. All these methods are known to use hand-crafted feature extraction
techniques which are constrained computationally as compare to any deep
learning-based methods. Many researchers have used different methods of
segmentation for Brain tumor detection. In the following sections, an overview of
different techniques of segmentation with current literature has been discussed.
A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 127

LITERATURE REVIEW
In this paper [9], a tumor classification technique based on wavelet features and
DNN is used. The authors have used OFPA which stands for oppositional flower
pollination algorithm, for the feature selection and Deep Neural Networks have
been used to classify the tumor. Finally, a possibilistic fuzzy c-means clustering
(PFCM) based technique has been used to extract the tumor-affected area. In the
study, the accuracy rate of 92%is achieved with a specificity of 91% and
sensitivity of 86%. Nazir et al. [10] proposed a segmentation method for brain
tumors is used that works in various stages. In the first stage, feature extraction
from input brain images is done using the Discrete Wavelet Transform technique.
Decomposition of an image takes place which results in multiple wavelet sub-
bands, then the bands with the highest energy are selected and divided into
multiple blocks. In the second stage, from every single block high variance
features are chosen using discrete cosine transformation technique and for
classification transferred to neural networks. In the last stage, segmentation is
achieved by k-means clustering. Shivhare et al. [11] provide a complete model
which is based on two constituents.

● For partitioning input images into various clustered images Parameter –free
clustering algorithm is used.
● Hole filling operations proceeded by morphological dilation are used to get to
the core region of the tumor.
Sazzad et al. [12] use gray-scale images obtained from the MRI technique. The
filter operation method is incorporated to eliminate undesirable noise and this
results in better segmentation. OTSU segmentation technique based upon thrush
hold is used to replace color segmentation. Feature information provided by
pathologists is finally used for the identification of the tumor core region.
Khalil et al. [13] proposed the segmentation method applied for brain tumor
identification involves three levels. Pre-processing for the quality improvement of
the image by eliminating undesirable noise like discrepancies and inhomogeneity
from the image. The technique based on two-step clustering is applied to obtain
the exact basic contour points for the level set segmentation method. Dragonfly
algorithm is integrated with k-means in the process of clustering. Level set
segmentation helps in the speedy extraction of tumor edges
In this paper [14], the core area of the tumor is detected using four major modules.
For pre-processing FNLM technique is applied for enhancement. For
segmentation, the OTSU technique is applied. For extraction of various features
like HOG, GEO and LBP are incorporated into one feature vector which is further
128 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

used for prediction. In this study [15], a DL-based technique is used for
segmentation in MR images. The patch-based technique is incorporated in the
convolutional neural network. Output label is predicted while considering both
contextual and local information. A two-phase training mechanism is used to
solve the problem of data imbalance. Preprocessing stage involves image
normalization and usage of the N4ITK technique for the correction of the bias
field. After the processing stage, To enhance segmentation results small false
positives are removed and Dilation proceeded by erosion is done in the same
stage. The authors in [16] use tool-reading. The method includes three modules:
MR images are used as input in preprocessing stage where they are converted into
greyscale images followed by removal of noise using median cleanout, grey
degree coincidence matrix help feature extraction, and lastly, label image is
obtained using watershed segmentation. In this study [17], the researchers use
FCM (fuzzy c-means) with super-resolution is used. Pretrained (CNN)
convolutional neural network-based architecture – SqueezeNet is applied for
extraction of features while (ELM) extreme learning machine is applied for
classification. In this paper [18], the filtering technique-Triangular fuzzy
Median(TFMED) is adopted for enhancing image quality. For the process of
segmentation Fuzzy set method is used which is an unsupervised clustering
technique. Extraction of Gobar features and similar texture features is done and
for the process of classification, they act as an input to ELM-Extreme learning
machine and RELM_LOO methods. In this study [19, 20], the techniques based
on various wavelet transforms and SVM are proposed. In the preprocessing
module removal of noise takes place using wavelet reforms. OTSU technique is
applied for feature extraction in segmentation and CWT and methods are also
applied to avoid dropping of edges. Finally, SVM techniques are applied for the
process of classification and prediction. In this study [21], for the improvement of
brain image NSCT technique is used and then an improvised image is utilized for
the feature extraction process. For classification, ANFIS is used. Morphological
functions are applied for classification. In this research work [22], the authors
have described that the whole architecture works in three main modules. In the
first module, Segmentation is done in two phases, superpixel zoning technique is
applied in the first step followed by discriminative clustering that results in better
clarity and accuracy of MR image of the brain. In the second module, Features are
extracted using the HWT-Haar Wavelet transformation technique. In the third
module, ADBRF (AdaBoost-ADB algorithm merged with RF is applied for
classification. In this paper [23], a methodology is used which has three broad
levels. At level first, preprocessing is done by applying both local smoothening
and non-local means to eliminate noise as much as possible. At the second level
GLRLM-gray level run length matrix, histogram-based method, and GLCM are
applied for extraction of features, ELM (extreme learning machine)-LRF- Local
A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 129

receptive field techniques are used for the classification purpose. At the third
level, a watershed segmentation algorithm is used for segmentation. The
researchers in [24], use improved forms of Fuzzy-C means and Template-base-
-K-Means algorithms. The Template-based-K-Means technique is applied for the
initialization of segmentation to select a perfect template that depends on gray-
level image intensity. The improvised Fuzzy-C means clustering technique is
applied for detection of tumor region, membership function is updated based on
various attributes of tumor-image like energy, contract, entropy, correlation, and
homogeneity. In this study [25], the MMR-Mathematical Morphological
Reconstruction segmentation technique is used. Median Filter is applied for
preprocessing of Images; MMR technique is utilized for proper segmentation.
Textural features and statistical features are extracted and PCA is applied for
selecting the lesser number of features. Finally, SVM is used for classification
using the GRB kernel.
METHODS OF SEGMENTATION

Conventional Methods

Thresholding
For image segmentation, the thresholding method is considered the simplest and
fastest method. In the thresholding technique, a scalar image is converted into a
binary image, and based on the pixel intensity, a threshold value is determined.
Comparison of the threshold value and pixel intensity value takes place. Value 1
is allotted to the pixels whose intensity value is equal or more than the value of
threshold while value 0 is allotted with less value of intensities, thus partitioning
the white pixels-foreground region and the black pixels-background region,
respectively. In the situations, wherein the image region types are greater than two
and equal intensity values are not shared by the objects of the image, then for the
segmentation of an image into multiple regions manifold threshold values are
utilized. For the mitigation of these scenarios, various dynamic and local thresh
holding techniques are suggested in the literature [26].

Region Growing
In the area of image segmentation, Region growing is the prominent and
foundational method utilized for segmenting homogeneous regions which have
equal pixel intensity values. The region growing method can be easily applied to
objects with variable shapes because this technique does not demand prior
information about the shapes. In the implementation of this method, each pixel
needs to be present in any region, connections should be there among the pixels in
130 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

a particular region, certain similarity conditions must be satisfied, bifurcation of

regions must be there and two distinct regions cannot have the identical
characteristics [27]. The seed point is selected either automatically or manually
and every region of concern is initialized with the same. One crucial factor due to
which correctness of segmentation is badly affected is the partial volume effect,
due to this effect intensity differentiation at the boundary of two region tissue
types is diminished, thereupon numerous tissue types are represented by creating
a voxel [28]. This issue can be resolved by introducing MRGM [29] utilizing
gradient information and details for recognizing boundaries. In [30], a
comparative survey of MRGM and region growing is presented and demonstrates
that MRGM provides preferable results of brain tumor segmentation over region
growing method in 3D T1-MRI images.

Supervised Methods
In the supervised learning method, the researchers establish an objective function
that defines the data model and then fine-tunes the attributes of the classifier by
the application of a set of known sample categories. The objective function helps
to predict the value of the output variable from the set of input values. These
methods include the training phase followed by the testing phase. Correctly
labeled data acquired in the phase of training is utilized in the testing phase for the
decision of classes of unlabeled data [31]. A prototype or model is built in the
training phase in which extracted features from data points are mapped to classes
or labels and for the determination of classes in the testing phase same model is
utilized for the unlabeled data set. Human interference is needed in the training
phase results in result variability demonstration. Classification precision is better
in supervised methods as compared to unsupervised methods.

k-Nearest Neighbor
K-NN algorithm collates the fresh unlabeled data occurrences with a group of data
samples that are labeled in a training data set. This algorithm is supervised
learning based on memory. Two levels are followed in the K-NN classification.
The closest neighbors for unlabeled data instances are recognized in the first level
then those closest neighbors are utilized for the determination of label or class for
data instances in the second level. The training data set, Class labels and the
feature vector of the samples are stored in the training step. For the prediction of
labels to the testing data points, gaps or similarities between the training data
instances and fresh unlabeled testing data points are calculated. Using the training
data set, k-Nearest neighbors for the unlabeled data points are identified and by
utilizing their labels unknown data or records are labeled using majority voting.
A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 131

The value of variable k hinges on the total number of cases and the number of
features extracted [32]. The distance metric utilized hinges on the problem
domain [33]. Steenwijk et al. [34] perceive a notable advancement in segmenting
white matter lesions.

Support Vector Machines

SVM transforms data from input space to multidimensional feature space and then
searches for a separating hyper-plane using a kernel function. Various kernel
functions that are commonly used in SVM are linear polynomial, PUK, and radial
basis functions. For the better classification of data in SVM, the image is divided
into two categories by locating a hyperplane [35]. For example, the hyperplane
having the highest distance or gap from the nearby data point from every side is
taken [36].Classification of data points in SVM hinges on the closeness with
segregating hyperplane taking extended time for computations for solving
quadratic and linear problems so created. Wu et al. [37] laid more focus on
dividing the image into multiple superpixels instead of voxels processing. An
effort has been put in for the reduction of errors in classification utilizing SVM as
portrayed in [38]. By utilizing second and first-order statistical features extraction
tumor grades like II, III and IV are classified. Whereas, superpixel-based
perspective or manual approach is included in ROI segmentation. For brain tumor
grading using MRI images, an intensified version is portrayed [39].

Random Forests
It is a classification technique, which is supervised, straightforward, and uses an
ensemble learning strategy while working effectively on big datasets. Without
deleting a variable it operates on enormous input variables and estimates key
features for the process of classification. Random Forests are comparatively
sturdy to noise and outliers.

Artificial Neural Networks

To predict the results of disease diagnosis a machine learning technique known as
Artificial Neural Network can be used. ANN is one of the most powerful tools
used in the analysis of complex clinical datasets. ANNs use known data in the
training process and then identify the complex relation between x and y variables,
acting as input and out variables respectively, followed by the prediction of output
values from a given set of input values. In animals, the nervous system is the
structure of interconnected neurons due to which animals react to the surrounding
132 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

environment. Artificial neural networks react to the environment and attain the
optimal state by emulating the deportment of neurons. To get pertinent output at
the last layer of artificial neurons, input signals are processed and then transmitted
through many in-between interconnected hidden layers. The complications of this
method increase with the increasing network size and thus become more time-
consuming. In addition to this, a large count of images is required for effective
training. Nevertheless, it has a pragmatic advantage of modeling any random non-
trivial distribution. At present various kinds of neural networks are used [40, 41].

Unsupervised Methods
Enormous data is produced by the medical industry, but some data doesn’t
possess corresponding labels. As a result of which the use of supervised learning
techniques is not suitable. In such cases, unsupervised learning can be useful to
discover the relation between hidden structures of unlabeled data. It is suited for
more complex scenarios where the segmentation is done based on the intensity,
gradient, or texture. There is n number of segments created by applying the
algorithms on features of images.

Clustering
The clustering techniques can be used for searching for useful patterns in
unorganized datasets. Clustering group pixels of an image into n groups called
clusters wherein each group is disjoint. There are two variants of clustering
namely hard clustering and soft clustering. Hard clustering like K-clustering
results in the plot of a pixel either in one of the clusters or none. Soft clustering
like Fuzzy C-Means clustering results in a pixel fitting in more than one cluster,
resulting in the overlapping of the clusters.

k-means Clustering
This technique is a vital method in methods that are based on pixels [42]. There is
a k means defined as x1, x1 … Xk for k clusters of an image. After each iteration,
observations are formed and means are calculated again and again until there is no
significant change in the new values.

Fuzzy C-Means Clustering

Fuzzy-based techniques are famous for representing uncertain data [43]. FCM
clustering algorithm was proposed by Bezdek in 1993 [44]. This technique
A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 133

allocates membership values by calculating the Euclidean distance from the center
to the cluster.
COMPARISON OF DIFFERENT TECHNIQUES
Table 1 given below gives the comparison in terms of accuracy for different
techniques using different datasets based on the literature review done.

Table 1. Performance Comparison.

S no Authors Technique Dataset Accuracy

01 [9] Wavelet Texture Features BRATS 2013 92%
02 [11] Parameter-Free Clustering 75.06%
03 [15] MR Images Using DL BRATS 2013 91.0%
04 [18] extreme learning BRATS 2015 96.5% (ELM)
87.5% (RELM-LOO)
05 [20] Wavelet Transforms and Support Vector Machines 92.0%
06 [21] ANFIS Classification 96.4%
07 [25] Mathematical Morphological Reconstruction 92.0%
08 [41] (KMANN) RS 96.6%

EXPERIMENT AND RESULTS

Artificial Neural Network and Convolutional Neural Network were implemented
using Tensorflow to ascertain the accuracy comparison and get an insight into the
possible improvements using Deep Learning models. The dataset used for this
experiment was taken from BRATS which included a total of 2020 images with
1060 tumorous images and 960 non-tumorous images.
After testing the models, the accuracy of ANN was calculated as 68.76% while
the accuracy of CNN was 88.70%. The confusion matrix for ANN is given in
Table 2.

Table 2. Confusion matrix ANN.

- Tumor Tumorless
Tumor TN=173 FP=41
Tumorless FN=88 TP=111

The F1 score for CNN is given in Table 3.

134 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

Table 3. F1 score CNN.

F1 score
Testing Data 0.88
Validation Data 0.91

The accuracy of CNN may not be what one might expect but it was implemented
in the most basic way without considering pre-processing or balancing of the
dataset. It is quite evident from the results that CNN performs way better than
ANN and looks very promising in improving the accuracy profoundly.

CONCLUSION
Machine learning techniques have been used by researchers all around the world
to have an automated system for Brain Tumor Detection. In this research work,
some machine learning techniques were deliberated upon and a comparative
analysis was put forth on the current literature. To understand the benefits of
using Deep Learning approaches like Convolutional Neural Networks, a
comparison of performances between ANN and CNN was given. The
performance difference is a strong endorsement of how well CNN performs even
on an imbalanced dataset without any proper consideration for pre-processing.
Future work should focus on using Deep Learning methods, standalone or hybrid,
to achieve better results in brain tumor detection using MRI.
CONSENT FOR PUBLICATION
Not applicable.
CONFLICT OF INTEREST
The authors declare no conflict of interest, financial or otherwise.
ACKNOWLEDGEMENT
Declare none.
REFERENCES
[1] P.B. Kanade, and P. Gumaste, Brain tumor detection using MRI images. vol. Vol. 3. Brain, 2015.
[2] A. Ben Rabeh, F. Benzarti, and H. Amiri, "Segmentation of brain MRI using active contour model",
Int. J. Imaging Syst. Technol., vol. 27, no. 1, pp. 3-11, 2017.
[http://dx.doi.org/10.1002/ima.22205]
[3] K. Kamnitsas, C. Ledig, V.F.J. Newcombe, J.P. Simpson, A.D. Kane, D.K. Menon, D. Rueckert, and
B. Glocker, "Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion
A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 135

segmentation", Med. Image Anal., vol. 36, pp. 61-78, 2017.

[http://dx.doi.org/10.1016/j.media.2016.10.004] [PMID: 27865153]
[4] K.B. Vaishnave, and K. Amshakala, "An automated MRI brain image segmentation and tumor
detection using SOM-clustering and proximal support vector machine classifier", IEEE Int.Conf.
Engineering and Technology (ICETECH), 2015
[http://dx.doi.org/10.1109/ICETECH.2015.7275030]
[5] N.J. Tustison, B.B. Avants, P.A. Cook, Y. Zheng, A. Egan, P.A. Yushkevich, and J.C. Gee, "N4ITK:
improved N3 bias correction", IEEE Trans. Med. Imaging, vol. 29, no. 6, pp. 1310-1320, 2010.
[http://dx.doi.org/10.1109/TMI.2010.2046908] [PMID: 20378467]
[6] K. Siddiqi, Y.B. Lauzière, A. Tannenbaum, and S.W. Zucker, "Area and length minimizing flows for
shape segmentation", IEEE Trans. Image Process., vol. 7, no. 3, pp. 433-443, 1998.
[http://dx.doi.org/10.1109/83.661193] [PMID: 18276263]
[7] J.G. Sled, A.P. Zijdenbos, and A.C. Evans, "A nonparametric method for automatic correction of
intensity nonuniformity in MRI data", IEEE Trans. Med. Imaging, vol. 17, no. 1, pp. 87-97, 1998.
[http://dx.doi.org/10.1109/42.668698] [PMID: 9617910]
[8] A. Parveen Singh, "Detection of brain tumor in MRI images, using combination of fuzzy C-means and
SVM", 2nd Int. Conf. Signal Processing and Integrated Networks (SPIN), 2015 pp. 98-102
[http://dx.doi.org/10.1109/SPIN.2015.7095308]
[9] S. Preethi, and P. Aishwarya, "Combining Wavelet Texture Features and Deep Neural Network for
Tumor Detection and Segmentation over MRI", J. Intell. Syst., vol. 28, no. 4, pp. 571-588, 2017.
[http://dx.doi.org/10.1515/jisys-2017-0090]
[10] M. Nazir, M.A. Khan, T. Saba, and A. Rehman, "Brain tumor detection from MRI images using multi-
level wavelets", 2019 Int. Conf. Comput. Inf. Sci. ICCIS 2019, 2019 pp. 1-5
[http://dx.doi.org/10.1109/ICCISci.2019.8716413]
[11] S.N. Shivhare, S. Sharma, and N. Singh, An efficient brain tumor detection and segmentation in MRI
using parameter-free clustering. vol. Vol. 748. Springer Singapore, 2019.
[http://dx.doi.org/10.1007/978-981-13-0923-6_42]
[12] T.M. Shahriar Sazzad, K.M. Tanzibul Ahmmed, M.U. Hoque, and M. Rahman, "Development of
Automated Brain Tumor Identification Using MRI Images", 2nd Int. Conf. Electr. Comput. Commun.
Eng. ECCE 2019, 2019 pp. 1-4
[http://dx.doi.org/10.1109/ECACE.2019.8679240]
[13] H.A. Khalil, S. Darwish, Y.M. Ibrahim, and O.F. Hassan, "3D-MRI brain tumor detection model using
modified version of level set segmentation based on dragonfly algorithm", Symmetry (Basel), vol. 12,
no. 8, 2020.
[http://dx.doi.org/10.3390/sym12081256]
[14] M. Sharif, J. Amin, M.W. Nisar, M.A. Anjum, N. Muhammad, and S. Ali Shad, "A unified patch
based method for brain tumor detection using features fusion", Cogn. Syst. Res., vol. 59, pp. 273-286,
2020.
[http://dx.doi.org/10.1016/j.cogsys.2019.10.001]
[15] S. Sajid, S. Hussain, and A. Sarwar, "Brain Tumor Detection and Segmentation in MR Images Using
Deep Learning", Arab. J. Sci. Eng., vol. 44, no. 11, pp. 9249-9261, 2019.
[http://dx.doi.org/10.1007/s13369-019-03967-8]
[16] A.N. Rakhonde, and P. Chippalkatti, Mri Image Based Brain Tumor, no. 3672, pp. 1-10, 2020.
[17] F. Özyurt, E. Sert, and D. Avcı, "An expert system for brain tumor detection: Fuzzy C-means with
super resolution and convolutional neural network with extreme learning machine", Med. Hypotheses,
vol. 134, no. October, p. 109433, 2020.
[http://dx.doi.org/10.1016/j.mehy.2019.109433] [PMID: 31634769]
[18] M. Sharif, J. Amin, M. Raza, M.A. Anjum, H. Afzal, and S.A. Shad, "Brain tumor detection based on
136 Global Emerging Innovation Summit (GEIS-2021) Mushtaq et al.

extreme learning", Neural Comput. Appl., vol. 32, no. 20, pp. 15975-15987, 2020.
[http://dx.doi.org/10.1007/s00521-019-04679-8]
[19] S. Reshmi, R.D. Salagar, and S.S. Veni, "a Survey on Brain Tumor Detection and Segmentation From
Mri Images", Int. J. Comput. Sci. Eng., vol. 6, no. 10, pp. 776-778, 2018.
[http://dx.doi.org/10.26438/ijcse/v6i10.776778]
[20] M. Gurbina, M. Lascu, and D. Lascu, "Tumor detection and classification of MRI brain image using
different wavelet transforms and support vector machines", 2019 42nd Int. Conf. Telecommun. Signal
Process., 2019 pp. 505-508
[http://dx.doi.org/10.1109/TSP.2019.8769040]
[21] A. Selvapandian, and K. Manivannan, "Fusion based Glioma brain tumor detection and segmentation
using ANFIS classification", Comput. Methods Programs Biomed., vol. 166, pp. 33-38, 2018.
[http://dx.doi.org/10.1016/j.cmpb.2018.09.006] [PMID: 30415716]
[22] A. Panda, T.K. Mishra, V.G. Phaniharam, and M.R.I. Automated Brain Tumor Detection Using
Discriminative Clustering Based, Segmentation. vol. Vol. 851. Springer Singapore, 2019.
[23] A. Ari, and D. Hanbay, "Deep learning based brain tumor classification and detection system", Turk. J.
Electr. Eng. Comput. Sci., vol. 26, no. 5, pp. 2275-2286, 2018.
[http://dx.doi.org/10.3906/elk-1801-8]
[24] M.S. Alam, "Automatic human brain tumor detection in mri image using template-based k means and
improved fuzzy c means clustering algorithm", Big Data Cogn. Comput., vol. 3, no. 2, pp. 1-18, 2019.
[http://dx.doi.org/10.3390/bdcc3020027]
[25] B. Devkota, A. Alsadoon, P.W.C. Prasad, A.K. Singh, and A. Elchouemi, Image Segmentation for
Early Stage Brain Tumor Detection using Mathematical Morphological Reconstruction, 2018.
[http://dx.doi.org/10.1016/j.procs.2017.12.017]
[26] D.Y. Kim, and J.W. Park, "Connectivity-based local adaptive thresholding for carotid artery
segmentation using MRA images", Image Vis. Comput., vol. 23, no. 14, pp. 1277-1287, 2005.
[http://dx.doi.org/10.1016/j.imavis.2005.09.005]
[27] R. Adams, and L. Bischof, "Seeded region growing", IEEE Trans. Pattern Anal. Mach. Intell., vol. 16,
no. 6, pp. 641-647, 1994.
[http://dx.doi.org/10.1109/34.295913]
[28] M. Sato, S. Lakare, M. Wan, A. Kaufman, and M. Nakajima, "A gradient magnitude based region
growing algorithm for accurate segmentation", Proc. Int. Conf. Image Processing., vol. 3, 2000 pp.
448-51
[http://dx.doi.org/10.1109/ICIP.2000.899432]
[29] S. Lakare, and A. Kaufman, 3D Segmentation Techniques for Medical Volumes. Center for Visual
Computing, Department of Computer Science, State University of New York, 2000, pp. 59-68.
[30] N.J. Salman, "Modified technique for volumetric brain tumor measurements", J. Biomed. Sci. Eng.,
vol. 02, pp. 16-19, 2009.
[http://dx.doi.org/10.4236/jbise.2009.21003]
[31] N. Gordillo, E. Montseny, and P. Sobrevilla, "State of the art survey on MRI brain tumor
segmentation", Magn. Reson. Imaging, vol. 31, no. 8, pp. 1426-1438, 2013.
[http://dx.doi.org/10.1016/j.mri.2013.05.002] [PMID: 23790354]
[32] N.E.A. Khalid, S. Ibrahim, and P. Haniff, "MRI brain abnormalities segmentation using k-nearest
neighbors (k-NN)", Int. J. Comput. Sci. Eng., vol. 3, no. 2, pp. 980-990, 2011.
[33] S.K. Warfield, M. Kaus, F.A. Jolesz, and R. Kikinis, "Adaptive, template moderated, spatially varying
statistical classification", Med. Image Anal., vol. 4, no. 1, pp. 43-55, 2000.
[http://dx.doi.org/10.1016/S1361-8415(00)00003-7] [PMID: 10972320]
[34] M.D. Steenwijk, P.J. Pouwels, M. Daams, J.W. van Dalen, M.W. Caan, E. Richard, F. Barkhof, and H.
 A Comparative Study on Machine Learning Global Emerging Innovation Summit (GEIS-2021) 137

Vrenken, "Accurate white matter lesion segmentation by k nearest neighbor classification with tissue
type priors (kNN-TTPs)", Neuroimage Clin., vol. 3, pp. 462-469, 2013.
[http://dx.doi.org/10.1016/j.nicl.2013.10.003] [PMID: 24273728]
[35] T.S. Kumar, K. Rashmi, S. Ramadoss, L.K. Sandhya, and T.J. Sangeetha, "Brain tumor detection
using SVM classifier", 3rd Int. Conf. Sensing, Signal Processing and Security (ICSSS), 2017, pp. 318-
23
[36] R. Ayachi, and N. Ben Amor, "Brain tumor segmentation using support vector machines", European
Conf. Symbolic and Quantitative Approaches to Reasoning and Uncertainty, 2009, pp. 736-47
[http://dx.doi.org/10.1007/978-3-642-02906-6_63]
[37] W. Wu, A.Y.C. Chen, L. Zhao, and J.J. Corso, "Brain tumor detection and segmentation in a CRF
(conditional random fields) framework with pixel-pairwise affinity and superpixel-level features", Int.
J. CARS, vol. 9, no. 2, pp. 241-253, 2014.
[http://dx.doi.org/10.1007/s11548-013-0922-7] [PMID: 23860630]
[38] M. Soltaninejad, X. Ye, G. Yang, N. Allinson, and T. Lambrou, Brain tumour grading in different
MRI protocols using SVM on statistical features. Med Image Underst Anal, 2014, pp. 259-264.
[39] M. Soltaninejad, X. Ye, G. Yang, N. Allinson, and T. Lambrou, "An image analysis approach to mri
brain tumour grading", Oncol News, vol. 9, no. 6, pp. 204-207, 2015.
[40] A. Shenbagarajan, V. Ramalingam, C. Balasubramanian, and S. Palanivel, "Tumor diagnosis in MRI
brain image using ACM segmentation and ANN-LM classification techniques", Indian J. Sci.
Technol., vol. 9, no. 1, 2016.
[http://dx.doi.org/10.17485/ijst/2016/v9i1/78766]
[41] M. Sharma, G.N. Purohit, and S. Mukherjee, Information retrieves from brain MRI images for tumor
detection using hybrid technique K-means and artificial neural network (KMANN). Networking
Communication and Data Knowledge Engineering, 2018, pp. 145-157.
[http://dx.doi.org/10.1007/978-981-10-4600-1_14]
[42] J. Vijay, and J. Subhashini, "An efficient brain tumor detection methodology using Kmeans clustering
algorithm", Int. Conf. Communication and Signal Processing, 2013, pp. 653-7
[43] T.A. Teli, and A. Wani, "Fuzzy Logic And Medicine With Focus On Cardiovascular Disease
Diagnosis", 5th IEEE International Conference on Computing for Sustainable Global Development,
2018.
[44] J.C. Bezdek, L.O. Hall, and L.P. Clarke, "Review of MR image segmentation techniques using pattern
recognition", Med. Phys., vol. 20, no. 4, pp. 1033-1048, 1993.
[http://dx.doi.org/10.1118/1.597000] [PMID: 8413011]

© 2021 The Author(s). Published by Bentham Science Publisher. This is an open access chapter published under CC BY 4.0 https://creativecommons.org/licenses/by/4.0/legalcode