See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/370882348

A Comprehensive Review of Brain Tumor Detection and Segmentation

Techniques

Conference Paper in Lecture Notes in Electrical Engineering · May 2023

DOI: 10.1007/978-981-19-9090-8_33

CITATIONS READS
2 272

3 authors, including:

Ankush Azade Sowmya Kamath S

National Institute of Technology Karnataka National Institute of Technology Karnataka
3 PUBLICATIONS 6 CITATIONS 166 PUBLICATIONS 1,349 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ankush Azade on 25 November 2023.

The user has requested enhancement of the downloaded file.

 A Comprehensive Review of Brain Tumor
Detection and Segmentation Techniques

Ankush Azade1 , Praful Kumar2 , and Sowmya Kamath S3

National Institute of Technology Karnataka, Surathkal, India-575025

1
[email protected],2 [email protected],3 [email protected]

Abstract. Brain tumors are particularly dangerous type of tumor and

if this is not treated in time it maybe prove to be deadly and may also
spread across other body parts. Brain tumor is the swelling or growth of
unwanted tissues in the brain that results from the unregulated and dis-
ordered division of cells. The presence of these tissues resulting abnormal
behavior and lot of other complications. The detection of brain tumor is
done by using different techniques out of which through magnetic reso-
nance images (MRI). The scanning process is a time-consuming manual
task that needs the involvement of medical professionals. Automating
the task of detection of the brain tumor while also grading the severity
accurately can help in managing the patients’ disease effectively. As tu-
mor tissue of different patients are different, automating such processes
is often a challenging task. Researchers have incorporated image segmen-
tation for extraction of suspicious regions from MRI, using image pro-
cessing and AI based techniques. Radiomic analysis also plays a big role
in feature extraction processes. In this paper, we present a comprehen-
sive review of existing approaches for brain tumour detection, covering
deep neural models, radiomic analysis and segmentation based methods
for brain tumor classification and segmentation, along with a discussion
on prevalent issues, challenges and future directions of research.

Keywords: Tumour Segmentation, Radiomic Analysis, Medical image

classification, Medical informatics

1 Introduction

The brain is the hub of all nerve activity, making it one of the most important
and delicate organs in the body. The spinal cord and brain constitute the ner-
vous system of a human body, and the body’s vital operations, such as thought,
speech, and movement, are controlled by the Central Nervous System. Each year,
about 350,000 new brain tumour cases are recorded around the world, with a
36% survival rate. A brain tumour is an unregulated cell multiplication that
leads in abnormal growth. Abnormality in the brain and spine growth can result
in negative effects on how a human walks, speaks and performs basic cognitive
activities. Regular headaches, speech difficulties, fatigue, difficulty in balance, in-
explicable nausea, and vomiting are all common signs of a brain tumour. These
2 Ankush Azade and Praful Kumar

symptoms are many times overlooked at early stages and they prove to be very
dangerous later when detected. Typically, these tumours can be classified into
primary and secondary tumors. Primary brain tumors are further divided into
two categories: low grade and high grade. When compared to high-grade tumors,
low-grade tumors grow more slowly. Secondary brain tumors are malignant tu-
mors that originate in another part of the body before spreading to the brain.
The severity of the condition determines the grade of the brain tumour, which
ranges from Grade I to Grade IV.
– Grade 1: The brain cells seem normal and will grow gradually and slowly.
At this stage, the long term survival is very likely.
– Grade 2: The cells will start looking slightly abnormal along with slow growth
of tumor. At this stage, the spread of tumor to the nearby tissues starts and
it may grow to a life threatening grade.
– Grade 3: The growth of tumor in the nearby tissues starts actively and by
the looks the cells will look abnormal.
– Grade 4: At this stage the growth of tumor is very fast and it will spread to
nearby cells very rapidly.
Neurological exams such as CT (computer tomography), MRI and other tests
such as angiogram, spinal tap, and biopsy are used to detect brain tumours.
MRI detection of brain tumours is one of the most effective methods, and cur-
rent diagnosis process mostly involves trained human experts. Often, grading
tumours is a difficult and error-prone process, due to the minute features that
need to be identified for accurate grading. Automated detection and grading of
Brain tumor has thus received significant research attention, as early diagno-
sis and accurate grading can be very beneficial in managing patient treatment
and improve disease prognosis. Detection at an benign stage (lower grade) can
be treated very effectively and can help in successful prevention of the tumour
turning malignant. The application of image processing, machine learning, and
deep learning have helped achieve very good results with high accuracy. Ad-
ditionally, automated segmentation of MRI images of brain tumours can help
enhance diagnostic and treatment choices, particularly in circumstances when
access to qualified specialists and radiologists is restricted.
A comprehensive assessment of current state-of-the-art efforts for brain tu-
mour classification and segmentation is offered in this paper. Deep learning based
works are primarily for classification of tumour in various ways i.e. if the tumour
is present or not and if present, classifying the tumor as per the severity grade of
the tumour. Certain features of the brain tumour like the radius of the tumour,
density of the tumour which can be recognised with radiomic analysis can help
in detection of tumour to a great extent and these properties can be used in
combination with one another in order to obtain better results. Segmentation
techniques are used in order to separate the tumor from other part of the brain in
the MR image so that regions of interest can be identified. The rest of this article
is organized as follows: Section II details the various approaches in deep learning
for brain tumour detection. In Section III, radiomic analysis based techniques
which are mainly used for feature extraction are discussed and segmentation
 Brain Tumor Detection and Segmentation Techniques 3

based approaches are discussed in Section IV. Also, various standard datasets
available for the research community for brain tumour detection research are
discussed in section V. Open challenges and research directions are discussed in
Section VI, followed by concluding remarks and references.

2 Deep Learning Approaches for Brain Tumour Detection

There has been a lot of research in Deep learning based brain tumor detection,
which can be perceived as a classification problem. Several studies have consid-
ered brain tumor detection as binary classification problem where two classes
are tumor and no tumor. A few have addressed this task as a multi-class clas-
sification where different grades of tumor are considered as different classes. We
present a review of existing works that address both binary classification and
multi-class classification in this section.
Sahoo et al. [22] developed a Computer Aided Diagnosis for classifying brain
MR pictures into two categories: normal and abnormal, which represent brain
states and confirm whether or not a tumour is present. To extract significant
information from the photos, the suggested model employs Discrete Wavelet
Transform. They used Symmetric Uncertainty Ranking and Probabilistic Princi-
pal Component Analysis for feature reduction. Finally, Support vector machines
and Filtered Classifiers are utilised for classification. The model was tested using
Harvard Medical School’s DS-75 and DS-160 benchmarks, with accuracy ranging
from 96.4% to 100% for some picture classes and as low as 5.4% for the NITR-
DHH dataset. This suggests that the model did not generalize well enough on
all the datasets and was more biased towards the dataset in which training was
carried out. Choudhary et al. [5] used a 3 layer CNN model including a fully
linked neural network as a backbone to obtain reliable MRI scan results. black-
Their approach is built on a CNN model for binary classification that classifies
the MRI image as Tumor Detected or No Tumor Detected. The CNN is used
for extracting the features and fully connected network is used for classification,
with a mean accuracy of 96%.
Amin et al. [23] proposed that for brain tumour detection, a 7-layer deep
learning model of 3 convolution layers, 3 ReLU layers, and 1 softmax layer was
used to segment and classify MR images. They segmented the input image into
patches of 4*65*65 pixels, determined the centre pixel label for each patch,
and gave it to the deep neural network. The overall results of segmentation
and classification were enhanced using this way of partitioning the image into
patches. They also assessed eight distinct benchmarks, all of which are widely
used in the identification of brain tumours. The results on these bechmarks
namely BRATS2012, BRATS2012 (synthetic images), BRATS2013, BRATS2013
(synthetic images), BRATS2014, BRATS2015, ISLES2015 and ISLES2017 were
in the range 95.5%-100% which shows that model also generalizes very well and
can be further used for real time usage. Fusion of certain hand crafted techniques
were addressed by Saba et al. [24] for the prediction of Glioma categories like
4 Ankush Azade and Praful Kumar

high-grade and low-grade gliomas. High-grade gliomas have a two-year survival

rate when treated early.
The authors used deep learning approaches, particularly Convolutional Neu-
ral Networks, to recognise the patterns. The suggested technique offers a substan-
tial contribution by transforming the input images into one channel. The Grab
cut strategy is used in concert with morphological operations to better segment
and refine the tumour region. Using local binary pattern (LBP) and histogram
of orientated gradients (HOG) features, the classification phase comprises ex-
tracting and integrating deep features in VGG-19 as well as custom features.
An accuracy of around 99% was obtained for the BRATS2015, BRATS2016,
and BRATS2017 benchmarks from BRATS, but there was no testing done with
other datasets, which limits the scalability of this work.
For predicting the progression-free (PFS) and overall survival (OS) of a newly
diagnosed glioblastoma (GBM) patient, Chaddad et al. [20] proposed multi-scale
characteristics based on the Laplacian Gaussian (LoG) filter. T1-weighted imag-
ing (T1WI) and fluid attenuation inversion recovery (FLAIR) images of 40 GBM
patients were used to extract characteristics that were manually categorised as
active tumour, necrosis, or edoema. To extract multi-scale texture features, the
LoG filter was applied to each of these regions of interest, and then univariate ap-
proaches including the Spearman rank correlation coefficient, logrank test, and
Kaplan-Meier estimator were employed to validate the features. Multivariate
analysis, as well as the relationship between them (i.e., Random Forest clas-
sifier). The first three and seventh features were shown to have statistically
significant correlations with PFS and OS, respectively. The three characteristics
derived only from the active tumor area were associated with OS (where p is
less than 0.05), and the hazard ratios (HR) were found to be 2.9, 3, and 3.24,
respectively. Using the random forests classifier to predict the combined charac-
teristics of patients with PFS and OS in GBM patients showed that the AUC
was 85.37% and 85.54%, respectively.
Dastmalchian et al. [25] adopted feature selection phase during which fea-
tures which were found to be highly correlated were picked using Spearman’s
rank coefficient. One of the traits from the correlated pairs was chosen to avoid
redundancy. Furthermore, the authors used the Wilcoxon rank sum test on each
of the tumour groups for the specified attributes in the preceding steps. The text
characteristics were evaluated using ROC and AUC. As a result, 20 out of 39 tex-
ture features were chosen for further investigation. When used to MRF, texture
analysis is a well-established and powerful method for inferring local and regional
heterogeneity in tumours, and it can reveal significant differences between tu-
mour types. This work demonstrated the use of texture analysis on MRF-derived
quantitative maps in malignant brain tumours. Hashemzehi et al. [26] evaluated
their proposed model on the T1-weighted CE-MR images and obtained a preci-
sion of 94.49%. A neural autoregressive distribution distribution (NADE) and a
CNN are combined in this model. They employed a three-step process for their
research, which included density estimation, feature exploitation, and classifica-
tion. In the initial step, the NADE model was trained to establish proper joint
 Brain Tumor Detection and Segmentation Techniques 5

distribution, which facilitated feature extraction in later stages. To limit the ef-
fect of noise and rotation, separate CNNs were utilised to learn features from the
raw image and the generated output. Although the proposed model’s operation
was extremely distinctive, it was not tested on numerous datasets or datasets
with a larger number of photos.

3 Radiomic Analysis for Brain Tumour Detection

Radiomics has been defined as “a quantitative approach to medical imaging,

which aims at enhancing the existing data available to clinicians by means of
advanced mathematical analysis” [27]. It has been reported to have significant
impact on clinical decision making as it utilizes the inherent image information
like pixel intensity, spatial distribution etc, for determining dominant features.
Radiomics also aims to convert medical images into high dimensional data. There
are CAD systems that detect if tumor/cancer is present or not, radiomics are
directed towards extracting large number of features that may be used for dif-
ferent purposes. With the help of radiomics, decision support tools that may
help in accurate prediction of tumor can be developed. Radiomic features can
be combined along with other characteristics of the patients in order to improve
decisions. As discussed the process of feature extraction includes 6 steps: ac-
quiring images, identifying the interest areas in image, segmentation of volumes,
extracting and qualifying descriptive features, using features to for collecting
database, mining this data for development of classifier models [37].
Brunese et al. [21] employed ensemble learning to combine ten distinct ma-
chine learning algorithms, including closest neighbour, Linear SVM, RBF-SVM,
neural network, Gaussian process, Naive bayes, C4.5, Random forest, QDA and
Logistic regression. As radiomic characteristics, first order, shape, Gray level co-
occurence matrix, Gray level run-length matrix, and Gray level size zone matrix
were also used. The first step was to build a classifier using the previously men-
tioned machine learning techniques. The second step in this process was to choose
the best classifier based on the criteria that were considered, such as decision
boundaries, performance obtained by single models, and calibration plots. The
voting ensemble classifier was used to predict the final prediction from among
the many predictions provided by the best classifiers. Following that, they built
a model in two stages: training and testing. In addition, the dataset was divided
into three parts, namely training, testing, and validation, in the ratio 2:1:1. The
accuracies obtained for Grade 1, Grade 2, Grade 3, and Grade 4 were 95.1 %,
95.1 %, 95.2 %, and 95.1 %, respectively, because they predicted four different
Grades of brain tumour.
Supervised learning was used by Mercaldo et al. [28] for the prediction of
the type of lung cancer among the five categories from the magnetic resonance
images. A total of 30 radiomic features were considered which belong to two
basic categories, namely, shape and gray level size zones. Carefully considering
the radiomic features is also very crucial stage as it will effect the final predic-
tion to a very large extent. The neural network considered was a basic single
6 Ankush Azade and Praful Kumar

layer feed forward perceptron with the Heaviside step function as the activation
function. The precision obtained was maximum 93.1% but since the prediction
was to identify the different stages of lung cancer the study is very useful. Gillies
et al. [36] showed that features can be extracted for varied imaging modalities
like position emission tomography, magnetic resonance and computed tomogra-
phy. Radiomics generally have the six steps: collection of image, identification
of volume of interest, volume segmentation, extraction of different features, us-
ing these features for database preparation, development of classifiers based on
these databases. The various usecases of radiomics include enabling diognosis,
tumor prognosis, treatment selection, deciding on location of biopsy or resect.
However, radiomics is a very young discipline and there are certain challenges
faced linke the reproducability of data, data sharing, lot of available data(big
data), lacking of standards, etc. In future, radiomics can prove to be a field of
great use in medical healthcare.

4 Segmentation Approaches for Brain Tumour Detection

blackSegmentation is the process of assigning a specified set of classes to each

pixel in an image, and is an important tool in general medical image analysis
tasks. It is predominantly used for extracting targeted information like regions of
interest (RoI) from medical diagnostic scans. Various segmentation methods like
watershed, threshold, edge-based, region-based and clustering based segmenta-
tion can be used. Recently, deep neural networks also have been used extensively
for RoI segmentation. We discuss some relevant works that adopt these methods
for brain tumor detection below.
Watershed segmentation is based on topological interpretation where result-
ing boundaries form closed and connected regions. Jemimma et al. [13] proposed
an approach using watershed dynamic angle projection and CNNs, for accu-
rate measurement of tumor regions. The textured characteristics of the brain
were extracted using the DAP pattern, and the tumor and non-tumor areas of
the MRI brain picture are classified using the CNN. Threshold Segmentation
is used to break down images into their constituent segments or areas in order
to recognise objects and edges is known as image segmentation. The pixels in
the image are partitioned based on their attributes. The pixels in the image are
either replaced with black or white in threshold segmentation. Natarajan et al.
[14] proposed a method based on sharpening and median filtering for prepro-
cess, and to improve image used histogram equalization and used thresholding
for segmentation of image. Finally, using the image subtraction approach, the
tumor area is derived. Threshold segmentation is not applicable when the object
area ratio changes.
Edge-based segmentation is based on the detection of discontinuities, which
works well in images with high contrast between object and background. Maiti
et al.[15] converted an RGB colour image to an HSV colour image, which is
divided into three regions: hue, saturation, and intensity. After enhancing the
contrast, the watershed algorithm is applied to each section of the image. The
 Brain Tumor Detection and Segmentation Techniques 7

disadvantage of this method is that it is difficult to construct a robust edge

linking. Region-based segmentation methods divide an image into homogeneous
regions. Flow from an inner point to an outer region generates distinct object
boundaries in this approach. Tak et al. [16] proposed object labelling for more
specific information about the tumor site and for detection and extraction they
used MATLAB software. SVM was utilised in an unsupervised way to develop
and retain the pattern for future usage. This method is sequential by nature and
quite expensive in both computation time and memory.
Clustering based Segmentation typically employ unsupervised approaches for
segmenting RoIs. Maksoud et al. [8] proposed a well-formed hybrid segmentation
techniques, based on fuzzy C-Means algorithm. They found that this did not
correctly segment images that are influenced by noise, outliers, or other imaging
distortions. They included a 3D evaluation of brain tumor identification using
3D slices. To recognise edges in CT and MRI images, Sharma et al. [9] used
the Gabor transform in association with soft and hard clustering. Their study
included a total of 4500 MRI scans and 3000 CT pictures.
K-means clustering was used to divide related sub-groups of characteris-
tics. The author used the Fuzzy C Means algorithm to represent the images
as histogram properties. Islam et al.[10] proposed a comparison of k-means, k-
medoids, and hierarchical clustering algorithms. They reported that the k-means
clustering method outperformed other methods and was simple to use. Bhoi and
Gupta[3] for investigating two clustering segmentation techniques, k-means and
fuzzy c-means, in terms of execution duration and computation of the tumour
area. Because it takes less time to execute and extracts the tumour region more
precisely, the k-means segmentation method outperformed fuzzy c-means. One
of the most effective image classification strategies was [29], which was used to
create a fused segmentation. The ensemble methodology, which is based on the
potential field notion of physics, combines data from PFS and other methodolo-
gies. The authors reported an accuracy of 89.8% using the BRATS dataset for
experimental validation of their approach.
Deep learning based Segmentation techniques have been adopted by several
researchers in recent years. Ahuja et al. [6] applied transfer learning for brain
tumor identification and superpixel approach for segmentation. They created a
model based on the BRATS 2019 brain tumor segmentation competition. For
training purposes, they used VGG19 transfer learning and with the help of super-
pixel method tumor was separated into LGG and HGG images. Jia and Chen [7]
used Support Vector Machine to construct a heterogeneous segmentation that is
completely automated. To train and test the accuracy of tumor diagnosis in MRI
images, they used a classification approach known as probabilistic neural network
classification system. This model focused on automatic meningioma segmenta-
tion and employed a multispectral brain dataset. Mittal et al. [30] introduced
the notion of the Stationary Wavelet Transform (SWT) and the novel Grow-
ing Convolution Neural Network (GCNN). The emphasis was on enhancing the
convention system’s correctness. In addition, a comparison study was conducted
using SVM and CNN, which revealed that the suggested model outperformed
8 Ankush Azade and Praful Kumar

both in terms of accuracy. Convolutional Neural Networks are compared to a

combination of texture-based feature extraction methodologies and conventional
techniques by Affoso et al. [31], using different learning models for the classifica-
tion of biological images. This study shows that deep learning-based algorithms
like CNN are the best at image classification.

5 Datasets

Several data sources are currently available, that provide images from brain
tumor of varying tumor ranges. The most commonly used modalities are Mag-
netic Resonance Images (MRI) images. Some of the most common resources
are BraTS, REMBRANDT, images from radiopaedia, etc. RAMBRANDT [32]
and images from Radiopaedia repository contain images that are divided into 4
grades of tumors. The RAMBRANDT dataset contains 110,020 MR Images with
Region of Interest segmentation and medical reports from 130 different patients.
The 4 categories in which the dataset is divided are I, II, III and IV depending
on the grade of the tumor and they contain MR images from 37, 40, 25, 28 pa-
tients respectively. The Radiopedia repository [33] dataset contains images from
121 MRIs which are labelled with four labels divided into I, II, III, IV categories
and they contain 36, 32, 25 and 28 MRIs respectively for the given grades.
BraTS 2019 [34] makes use of data from multiple institutes and the data is
pre-operative MR images and the main focus is on instrinsically heterogeneous
brain tumors, specifically gliomas. Furthermore, BraTS’19 focuses on predict-
ing patient overall survival using integrative analyses of radiomic features and
machine learning algorithms in order to pinpoint the clinical relevance of this
segmentation task. Finally, at BraTS’19, an experimental evaluation of tumour
segmentation uncertainty will be carried out. The dataset is divided into four
subfiles, with dataset volumes classified as T1, T2, Flair, or T1Ce sequences.
There are 155 slices in each volume. MRI images are used to create the slices.
One volume necessitates 155 slices. Glioma dataset with 210 high-grade volumes.
There are 285 volumes in total, with 75 volumes in another type of glioma. Be-
cause it is a standard dataset for detecting brain tumours, BraTS2019 can be
used. The Kaggle dataset [35] containing MRI Images with 155 tumor and 98
no tumor images (253 total images).

6 Open Challenges and Research Directions

Based on the detailed review of various approaches for brain tumor detection,
several interesting insights were revealed. It was observed that deep learning
techniques have been extensively used for accurate classification of brain tu-
mor into multiple classes, for enabling automated severity grading tasks. This
is critical in clinical workflows for supporting decision making tasks that can
help improve the diagnostic process and reduce time taken for diagnosis. Most
studies have focused on binary classification of the brain tumor that reduce the
task to a prediction with respect to if brain tumor is present or not. However,
 Brain Tumor Detection and Segmentation Techniques 9

Fig. 1. Healthy brain MRI Images

Fig. 2. MRI Images containing Tumor

Fig. 3. Sample images from BraTS dataset

10 Ankush Azade and Praful Kumar

Fig. 4. Images of 4 grades of Brain Tumor

the main concern is the type of brain tumor as there is a very high chance of
grade 1 and grade 2 tumors not being detected in time. Also, for many of the
datasets the number of samples in different classes is very unbalanced. Although,
the classes can be balanced using augmentation techniques, the sample images
obtained after augmentation are just copies of the other images by introducing
some small variations. However, if there can be datasets with almost balanced
classes then it would make the results better and more practical to use in real
world applications.
Segmentation is a technique in which a image is divided into number of par-
titions called as segments. The image segments simplify image processing and
analysis by reducing the complexity of the image. Most existing studies have fo-
cused on standard parametric methods like threshold, watershed, region-based
and clustering based segmentation methods. Threshold segmentation is an easy
to use method which takes pixels’ mean value as threshold value, the disadvan-
tage being that it is not suitable when the object area ratio changes. Watershed
is based on topological interpretation which results in closed and connected re-
gions and the problem is it gives excessive segmentation while edge based uses
discontinuity detection and works well in images with good contrast between
object and background. In these standard techniques, k-means clustering is easy
to implement, takes less processing time and also provides accurate predictions.
While applying these segmentation techniques on the MRI images, skull strip-
ping has not been used. This is a significant gap as image quality has a significant
impact on the segmentation method’s performance. The result of image segmen-
tation is determined by a lot of aspects, including pixel color, intensity, image
similarity and issue domain. Hence, automated methods that can address these
issues are the need of the day.

7 Concluding Remarks
There are different methods for identifying brain tumor by looking at Magnetic
Resonance images. Deep learning methods and segmentation along with radiomic
analysis play a huge role in detection of brain tumor. Deep learning methods
 Brain Tumor Detection and Segmentation Techniques 11

majorly focus on learning of a model by learning from different images given as

a train data. Segmentation focuses on division of regions in an image so that it
can identify the tumor by those regions. Radiomic analysis is the extraction of
features like shape and size of tumor from a given image so that these features
can be used for the further training and testing of data. These methods can be
collectively used to obtain good results. Many of the publications combined deep
learning with radiomic analysis, allowing characteristics derived from images to
be input into the model for training and testing. These give better results as
compared to the studies that have individually used one of these methods.

References

1. S. Irsheidat and R. Duwairi.:Brain tumor detection using artificial convolutional

neural networks. 2020 11th International Conference on Information and Commu-
nication Systems, IEEE, April 2020.
2. P. Rajan and C. Sundar.:Brain tumor detection and segmentation by intensity ad-
justment, Journal of Medical Systems, vol. 43, July 2019
3. S. Krishnakumar and K. Manivannan.:Efective segmentation and classifcation of
brain tumor using rough k means algorithm and multi kernel svm in mr images,
Journal of Ambient Intelligence and Humanized Computing, Springer, pp. 6751—-
6760, August 2020
4. M. T.Elmegy and K. M.-E. Magad.:A multiple classifiers system for automaticmulti-
modal brain tumor segmentation, (Cairo, Egypt), 2019 15th International Computer
Engineering Conference (IEEE), Dec 2020
5. C. L. Choudhary, C. Mahanty, R. Kumar, and B. K. Mishra.:Brain tumor detec-
tion and classification using convolutional neural network and deep neural network,
(Gunupur, India), International Conference on Computer Science, Engineering and
Applications, IEEE, 2020.
6. S. Ahuja, B. Panigrahi, and T. Gandhi.:Transfer learning based brain tumor detec-
tion and segmentation using superpixel technique, (Lucknow, India), 2020 Interna-
tional Conference on Contemporary Computing and Applications, (IEEE), 2020.
7. Z. Jia and D. Chen.:Brain tumor detection and classification of mri images using
deep neural network, IEEE Access ( Early Access ), pp. 1–1, Aug. 2020. Available:
8. E. Maksoud, M. Elmogy, and R. Al-Awadi.:Brain tumor segmentation based on a
hybrid clustering technique, Egyptian Informatics Journal Vol. 16, No. 1, vol. 16,
pp. 71––81, Mar. 2015.
9. D. A. Sungheetha.:Gtikf-gabor-transform incorporated k-means and fuzzy c means
clustering for edge detection in ct and mri, Journal of Soft Computing Paradigm
(JSCP), vol. 02, pp. 111–119, 2020.
10. S. Islam1 and D. M. Ahmed.:Implementation of image segmentation for natural
images using clustering methods, International Journal of Emerging Technology and
Advanced Engineering, vol. 3, March 2013.
11. R. Agrawal, M. Sharma, and B. K. Singh.:Segmentation of brain tumor based on
clustering technique: Performance analysis, Journal of Intelligent Systems, vol. 28,
2017
12. Sreeji, V. R, A. Beevi, and N. N.:Survey on different methods of image segmenta-
tion, International Journal of Scientific Engineering Research, vol. 4, April 2013.
12 Ankush Azade and Praful Kumar

13. T. A. Jemimma and Y. J. Vetharaj.:Watershed Algorithm based DAPP features

for Brain Tumor Segmentation and Classification, 2018 International Conference
on Smart Systems and Inventive Technology , 2018, pp. 155-158
14. P. Natarajan, N. Krishnan, N. S. Kenkre, S. Nancy and B. P. Singh.:Tumor de-
tection using threshold operation in MRI brain images, 2012 IEEE International
Conference on Computational Intelligence and Computing Research
15. I. Maiti and M. Chakraborty.:A new method for brain tumor segmentation based
on watershed and edge detection algorithms in HSV colour model, 2012 National
Conference on computing and communication systems, 2012, pp. 1-5
16. R. Tak, N. Kumar, Satyaki, S. Verma and S. Dixit.:Segmentation of medical image
using region based statistical model, 2017 International Conference on Intelligent
Computing and Control Systems, 2017, pp. 1247-1253
17. D. karungan and D. N. Sujatha.:Survey on various image segmentation technique,
vol. 4, 2017.
18. N. Zaitoun and M. Aqel.:Survey on image segmentation techniques, International
Conference on Communication, Management and Information Technology , vol. 65,
pp. 797–806, 2015
19. N. Chakrabarty.:Brain mri images for brain tumor detection, https://www.kaggle.
com/navoneel/brain-mri-images-for-brain-tumor-detection
20. Chaddad A, Sabri S, Niazi T, Abdulkarim B.:Prediction of survival with multi-
scale radiomic analysis in glioblastoma patients. Med Biol Eng Comput. 2018
Dec;56(12):2287-2300.
21. L. Brunese, F. Mercaldo, A. Reginelli, and A. Santone.:An ensemble learning ap-
proach for brain cancer detection exploiting radiomic features, Computer Methods
and Programs in Biomedicine, vol. 185, p. 105134, 2020.
22. Swagatika, M. Sahoo, and S. Bakshi.:A novel privacy-supporting 2-class classifi-
cation technique for brain mri images, Biocybernetics and Biomedical Engineering,
vol. 40, 05 2020.
23. J. Amin, M. Sharif, M. Yasmin, and S. L. Fernandes.:Big data analysis for brain
tumor detection: Deep convolutional neural networks, Future Generation Computer
Systems, vol. 87, pp. 290–297, 2018.
24. T. Saba, A. Sameh Mohamed, M. El-Affendi, J. Amin, and M. Sharif.:Brain tumor
detection using fusion of hand crafted and deep learning features, Cognitive Systems
Research, vol. 59, pp. 221–230, 2020.
25. S. Dastmalchian, O. Kilinc, L. Onyewadume, C. Tippareddy, D. McGivney, D. Ma,
M. Griswold, J. Sunshine, V. Gulani, J. S. Barnholtz-Sloan, et al.:Radiomic analysis
of magnetic resonance fingerprinting in adult brain tumors, European Journal of
Nuclear Medicine and Molecular Imaging, vol. 48, no. 3, pp. 683–693, 2021.
26. R. Hashemzehi, S. J. S. Mahdavi, M. Kheirabadi, and S. R. Kamel.:Detection of
brain tumors from mri images base on deep learning using hybrid model cnn and
nade, Biocybernetics and Biomedical Engineering, vol. 40, no. 3, pp. 1225–1232,
2020.
27. van Timmeren, Janita E., Davide Cester, Stephanie Tanadini-Lang, Hatem Alka-
dhi, and Bettina Baessler.:Radiomics in medical imaging—“how-to” guide and crit-
ical reflection. Insights into imaging 11, no. 1 (2020): 1-16.
28. F. Mercaldo, A. Reginelli, and A. Santone.:Neural networks for lung cancer detec-
tion through radiomic features, in 2019 International Joint Conference on Neural
Networks, pp. 1–10, 2019.
29. I. Cabria and I. Gondra.:Mri segmentation fusion for brain tumor detection, Infor-
mation Fusion, vol. 36, pp. 1–9, 2017.
 Brain Tumor Detection and Segmentation Techniques 13

30. M. Mittal, L. M. Goyal, S. Kaur, I. Kaur, A. Verma, and D. Jude Hemanth.:Deep

learning based enhanced tumor segmentation approach for mr brain images, Applied
Soft Computing, vol. 78, pp. 346–354, 2019.
31. C. Affonso, A. Rossi, F. Vieira.:Deep learning for biological image classification,
Expert Systems with Applications, vol. 85, pp. 114–122, 2017.
32. Scarpace, L., Flanders, A. E., Jain, R., Mikkelsen, T., Andrews, D. W. (2019).
Data From Rembrant [Data set]. The Cancer Imaging Archive.
33. Radiopaedia, https://radiopaedia.org/
34. BraTS 2019, https://www.med.upenn.edu/cbica/brats2019/data.html
35. Kaggle Dataset, https://www.kaggle.com/navoneel/
brain-mri-images-for-brain-tumor-detection
36. Gillies RJ, Kinahan PE, Hricak H.: Radiomics: Images Are More than Pictures,
They Are Data. Radiology. 2016 Feb;278(2):563-77.
37. Abdalla Ibrahim, Martin Vallières, Henry Woodruff, Sergey Primakov, Mohsen
Beheshti, Simon Keek and M. Mottaghy. Radiomics Analysis for Clinical Decision
Support in Nuclear Medicine.

View publication stats