A Comprehensive Survey On Brain Tumor Diagnosis Using Deep Learning and Emerging Hybrid Techniques With Multi Modal MR Image

Archives of Computational Methods in Engineering (2022) 29:4871–4896
https://doi.org/10.1007/s11831-022-09758-z
REVIEW ARTICLE
A Comprehensive Survey on Brain Tumor Diagnosis Using Deep

Learning and Emerging Hybrid Techniques with Multi‑modal MR
Image
Saqib Ali1 · Jianqiang Li1 · Yan Pei2 · Rooha Khurram3 · Khalil ur Rehman1 · Tariq Mahmood1
Received: 2 July 2021 / Accepted: 18 April 2022 / Published online: 9 May 2022
© The Author(s) under exclusive licence to International Center for Numerical Methods in Engineering (CIMNE) 2022
Abstract
The brain tumor is considered the deadly disease of the century. At present, neuroscience and artificial intelligence conspire
in the timely delineation, detection, and classification of brain tumors. The process of manually classifying and segmenting
many volumes of MRI scans is a challenging and laborious task. Therefore, there is an essential requirement to build
computer-aided diagnosis systems to diagnose brain tumors timely. Herein review focuses on the advances of the last decade
in brain tumor segmentation, feature extraction, and classification through powerful and versatile brain imaging modality
Magnetic Resonance Imaging (MRI). However, particular emphasis on deep learning and hybrid techniques. We have
summarized the work of researchers published in the last decade (2010–2019) termed as the 10s and the present decade
(only including the year 2020) termed as the 20s. The decades in review reveal the bore witness to the critical revolutionary
paradigm shift in artificial intelligence viz. conventional/machine learning methods, emerged deep learning, and emerging
hybrid techniques. This review also covers some persistent concerns on using the type of classifier and striking trends in
commonly employed MRI modalities for brain tumor diagnosis. Moreover, this study ensures the limitation, solutions, and
future trends or opens up the researchers’ advanced challenges to develop an efficient system exhibiting clinically acceptable
accuracy that assists the radiologists for the brain tumor prognosis.
Keywords Brain tumor diagnosis · Tumor segmentation · Deep learning · Hybrid techniques · Machine learning
1 Introduction behavior, from lowest to extreme aggressive [2]. Low-

grade gliomas (LGG) (grades I and II) and high-grade
Brain tumor commonness is a significant contributing aspect glioma (HGG) (grades III and IV) are two major categories
to the universal death rate. According to the GLOBOCAN of brain tumors. The HGG grows rapidly, with a maximal
2020 report, the number of new brain cancer cases was life expectation is two years. In contrast, LGG grows slowly
308,102, and 2.5% of people died from brain cancer [1]. and sometimes allows the subject to have many years of life
Tumors originating in the brain can be categorized into four anticipation. Indeed, brain tumors have many characteristics,
main types: gliomas, meningiomas, pituitary adenomas, including variable locations, varying shapes, and sizes, and
and nerve sheath tumors. The World Health Organization poor contrast, leading to overlapping with the intensity
(WHO) categorizes brain tumors through cell origin and values of healthy brain tissues [3]. These characteristics
affect the complexity of tumor growth and predict the
extent of resection at the time of surgical planning, which
* Yan Pei has implications for patient treatment [4]. Therefore,
peiyan@u-aizu.ac.jp
distinguishing healthy tissues from the tumor and exact
1
Faculty of Information Technology, Beijing University classification is not an easy task. Reliable segmentation and
of Technology, Beijing 100124, China brain tumor classification are important to determine the
2
Computer Science Division, University of Aizu, tumor size, exact position, and type.
Aizuwakamatsu 965‑8580, Japan Timely detection of tumors is essential to treat brain
3
Department of Chemistry and Chemical Engineering, Beijing tumors effectively. Medical imaging modalities such as
University of Technology, Beijing 100124, China computed tomography (CT), biopsy, cerebral angiography,
13
Vol.:(0123456789)
4872 S. Ali et al.
myelography, positron emission tomography (PET), and underlying tumor molecular alterations using hand-
MRI contribute a vital role towards brain tumor detection crafted features [11]. These features were extracted from
due to their non-invasive nature [5]. Amongst them, MRI biopsies of 13 subjects using textural metrics. However,
and CT are the two most commonly exercised modalities. deep learning (DL) techniques do not need pre-selection
MRI provides an in-depth scan that can easily spot brain of features because they automatically learn the most
tumors and other infections. appropriate features for identification and prediction.
Moreover, MRI is the most popular scan system in Deep learning automatically mines important features,
detecting several diseases and their treatment planning in evaluates patterns, and categorizes the information by
clinical trials, especially brain tumors [6]. The neurological extracting multi-level features. Lower-level features
MR images for brain tumor diagnosis are captured from include corners, edges, and basic shapes, while higher-
three different views, viz. axial, coronal, and sagittal [7], level features include image texture, more processed
as illustrated in the Fig. 1a. Three primary MR modalities shapes, and particular image patterns [12]. Moreover,
include: T1-weighted (T1-W), T2-weighted (T2-W), and deep learning techniques are used to extract features
FLAIR are utilized for brain tumor analysis [8] as illustrated from additional information and integrate them into the
in the Fig. 1b. Initially, the brain tumor diagnosis relies recommendation process [13]. However, it is unable to
on the radiologist experts after the precise analysis and maintain the spatial consistency and visual delineation
comprehensive monitoring of the image. However, owing of the subject. Therefore, the research paradigm for brain
to the limited availability of domain knowledge expertise, tumor detection, segmentation, and classification has now
this process is time-consuming. been shifted towards hybrid-based techniques. A hybrid
CAD systems truly help radiologists to improve the approach is a method of combining the strengths of several
diagnosis of brain tumors in no time, thereby decreasing classifier systems into a single system to enhance the
the mortality rate due to brain cancer. The fundamental overall accuracy.
rationale of the CAD is to automate the process of The paradigm shift from conventional machine learning
detecting brain tumor images with superior authenticity to deep learning and hybrid approach in the brain tumor
and reliability. Many articles have been published on analysis domain inspired us to do an extensive review over
brain tumor detection, classification, and segmentation the last (10s) and the present decade (20s). The primary
to date. The majority of previous research focused on objective of this work is summarized as follows.
the conventional/machine learning-based approaches.
Machine learning (ML) techniques are uniquely suitable – This review attempts to sum up the previously reported
to address big data challenges such as brain tumor work on brain tumor segmentation, feature extraction,
segmentation. However, it has been used to train machines and classification utilizing brain MRI scans.
for image recognition, which generally requires human – The comprehensive study has been exploited to show the
intervention and intelligence [9]. Typical ML methods development of soft computing, viz. artificial intelligence
apply human-designed based feature extraction techniques (AI) in the entire field of brain tumor analysis, both from
to differentiate tumor properties and features in imaging an application-driven and methodology perspective.
data [10]. For instance, Hu, Leland S., et al. proposed a – The review presented here aims to assist the researcher in
novel study based on a decision tree classifier to predict designing state-of-the-art CAD methods which can help
radiologists for the early diagnosis of brain tumors.
– To present the current trends in the domain of deep
learning and a hybrid-based approach for tumor
prognosis.
– Consequently, the review presents the key findings in the
dedicated discussion part that successfully elaborate the
shift’s pendulum.
– To highlight the prospects and open research challenges
for the successful and fully automatic identification of
brain tumors.
– Moreover, the statistical analysis was carried out by
considering various factors and presented in graphs.
– Lastly, performances of CAD systems of brain tumors
through multi-modal MR scans for tumor segmentation,
Fig. 1 Neurological MRI scans a Three different views (I) Axial, (II)
Coronal, and (III) Sagittal. b Basic MRI modalities (I) T1-W MRI feature extraction, and tumor classification have been
scan, (II) T2-W MRI scan, and (III) FLAIR studied and compared for the 2010–2020 years.
13
A Comprehensive Survey on Brain Tumor Diagnosis Using Deep Learning and Emerging Hybrid… 4873
In this manuscript, we have used freely available search

databases including Google Scholar, Scopus, IEEE explorer,
Science Direct, and PubMed to find the most relevant papers
by applying different queries. We have limited our search to
manuscripts published between the years 2010–2020. We
have used the following queries in various combinations:
“brain cancer diagnosis”, “Brats dataset segmentation”,
“brain tumor segmentation and classification”, “brain Fig. 2 Representation of different tumor segmentation sub-regions
tumor detection using machine learning and deep learning (taken from the BRATS 2013 database)
classifiers”, “brain tumor MRI and deep learning”, “brain
tumor using Harvard dataset”, “brain tumor detection and
BrainWeb dataset”, “brain tumor detection and segmentation For brain tumor segmentation, several conventional
using TCIA dataset”, “artificial intelligence and brain segmentation techniques have been utilized so far, including
tumor”, etc. More than 400 related papers are thoroughly contour and shape-based methods, thresholding-based
reviewed, among them, 190 were most relevant to brain techniques, edge, and region-based algorithms, statistical-
tumor detection, segmentation, and classification, which based approaches, multi-resolution analysis, etc. In this
we have chosen for this manuscript. review, all other methods are categorized as conventional/
After this introduction section, the whole review is ML-based methods (traditional approaches), DL-based
organized as follows. We present the development for brain methods (emerged method), and hybrid-based approaches
tumor segmentation techniques through MRI over the (emerging techniques), as illustrated in Fig. 3. The pros
years 2010–2020 in Sect. 2. Then this review summarizes and cons of the most commonly utilized segmentation
the development for brain tumor feature extraction and and classification techniques are briefly summarized and
classification techniques through MRI over the years compared in Table 1.
2010–2020 in Sect. 3. The statistical analysis of the
decades comprehensively examines the pros and cons of 2.1 ML‑Based Segmentation
published literature for the design of a reliable, automated,
cost-effective, robust, secondary diagnostic tool, i.e., a Many researchers applied ML-based techniques for the
CAD system is done in Sect. 4. The current trend on deep segmentation of brain tumors. Amin et al. have designed
learning-based brain tumor diagnosis is presented in Sect. 5. an automated segmentation network for brain tumor MR
The current trend on hybrid-based brain tumor diagnosis is images. A support vector machine (SVM) classifier is
illustrated in Sect. 6. Then a comprehensive discussion part, employed using different kernels to categorize the cancerous
where the limitations, the research findings, and research or non-cancerous brain images. The performance of the
challenges are briefly elaborated in Sect. 7. The future designed model has been evaluated on standard datasets
research directions for the selection of appropriate technique, named Harvard and Rider. The experimental outcome
image-modality, and dataset for brain tumor segmentation demonstrates that the model performed the segmentation
and classification are briefly explained in Sect. 8. In the end, task very efficiently [14]. Mehmood et al. utilized a self-
the conclusion of this review is made in Sect. 9. organizing map (SOM) clustering algorithm for brain lesions
segmentation [15]. The accuracy of the model was predicted
at 0.76%. In another report, Demirhan and Guler proposed
2 The Development for Tumor the SOM and learning vector quantization (LVQ) to segment
Segmentation Techniques (2010–2020) WM and GM [16]. Zexuan et al. implemented generalized
rough fuzzy C-means clustering for tumor segmentation
The process of cleaving an image into the region of interest [17]. In comparing Fuzzy C means (FCM) methods, a
(ROIs) for the easy depiction and characterization of the state-of-the-art technique was introduced that categorized
data is termed segmentation. The critical objective of WM, cerebrospinal fluid (CSF) spaces, and GM using
segmentation is to locate the tumor regions for the more Adaptive Fuzzy K-mean (AFKM) clustering. Researchers
straightforward prognosis and classification of brain state that with the implementation of the AFKM algorithm,
tumors by changing the representation of the MR images. superior results differ in contrast to FCM qualitatively and
It separates the tumor regions, for instance, necrotic and quantitatively [18].
edema, from the non-tumor regions, mainly WM (white The majority of previous research focused on the
matter) and GM (gray matter) [3] as presented in Fig. 2. machine learning-based approach. Machine learning
Owing to the complex anatomy and high inconsistency, techniques are uniquely suitable to address big data
segmentation or labeling of brain MR scans is challenging. challenges such as brain tumor segmentation. However,
13
4874
Table 1 The benefits and drawbacks of the most popular segmentation and classification techniques
Technique Benefits Drawbacks
13
Deformable model Capable of lengthening or shrinking over time inside an image and verify the When used for noisy data with unclear boundary, might be produce results that have
specific features varying topology
Level set It uses a command-based approach for the segmentation of an image because it can Need substantial thought to build appropriate velocities for progressing the level set
control any of the concavities, convolution, splitting, or cavities function
A beneficial method for a varied class of complex snags
Region growing It can accurately divide the regions of an image that have similar properties It needs a manually selected seed point and eliminates entire pixels connected to the
Capable to perform well to remove noise predefined primary seed
Sensitive in noise removing
Threshold based Threshold segmentation methods are much beneficial for the task of image Because of significant intensity variation, these algorithms are not suitable for all type
linearization of MRI brain images
DWT Preservation of edge sharpness Shift sensitive
Not depend on segmentation of the image Poor directionality
It provides localized frequency information If the phase information is unknown: results with real and unreal projections can be
generated
ANN Suitable for complicated, multivariate nonlinear domains The time-consuming learning process needs a patient-specific learning
No need for statistical modeling Require high training to operate
Perform better on noisy fields
SVM High generalization performance Huge training time
SVM can decrease the misclassifications problem in the training set Patient-specific learning
High generalizability while using a small training dataset Difficult to know the best kernel value for a specific task
KNN Accurately provides distance, a weighted average of pixels Choice of (k) disturbs the performance
KNN is an easy and robust algorithm KNN is memory intensive and has a slow estimation
Useful for datasets having large training images Performance of this algorithm affected in the presence of noisy data
FCM It is a popular technique for the segmentation of a medical image High training time
Gives the best result for the overlapped data set FCM does not make accurate segmentation results due to the arbitrary nature
K-mean It runs fast enough in real-time image segmentation scenarios Difficult to predict K-value; dependent on initial values, which is a time-consuming
Relatively simple to implement process
Easily adapts to new examples K-means has an obstacle where data has varying sizes and density
SOM A self-organizing network has a more effortless and faster training algorithm It has a misclassification problem due to its static nature
A more powerful tool for the visualization of high-dimensional images SOM is computationally expensive
To find important clusters, it uses a statistically salient features map that is useful for
new patterns classification
CNN Suitable for the feature engineering and pattern recognition tasks Relations between parameters selections and output segmentation results needs more
The segmentation mechanism is suitable for medical imaging mathematical explanation
Complex architecture and computationally expensive
GA It finds the solution of the research problem that is a biologically inspired generic It depends upon the fitness function
process implemented in the computing Computationally expensive
Supports multi-objective optimization
Hybrid A hybrid approach is a method of combining the strengths of several classifier –
systems into a single system to enhance the overall accuracy
S. Ali et al.
some ML-based methods utilize manually segmented 2.3 Hybrid‑Based Segmentation

training images. Nevertheless, manual segmentation of
the images is expensive, extensive/tedious, and needs The recent success of hybrid technology in the medical
a team of expert radiologists. Therefore, ML generally domain reflects the interest of researchers in computer
requires human intervention and intelligence [9] as vision. Hybrid systems combine two or more methods to
typical ML methods apply human-designed-based feature overcome the various issues involving high computational
extraction techniques to differentiate tumor properties and time, low accuracy, and effectiveness.
features in imaging data [10]. Mittal M. et al. suggested a combined framework using
SWT-CNN for brain tumor segmentation to enhance the
CNN-based model accuracy performance [28]. Stationary
2.2 DL‑Based Segmentation Wavelet Transform (SWT) technique was applied for feature
extraction rather than Fourier transform that provides
DL techniques do not need an initial feature selection step improved results for discontinuous data followed by the
because it automatically learns the appropriate features random forest (RF) method for the classification task. The
for identification. DL is a subgroup of ML that can suggested technique contributes 2% improvement compared
automatically mine important features, evaluate patterns, with traditional CNN. Nilesh Bhaskarrao et al. proposed
and categorize the information by extracting multi-level Berkeley wavelet transform (BWT) with an SVM framework
features [12]. Various DL models and methods are at hand for tumor segmentation [29]. BWT was employed for the
for tumor segmentation via MRI scans. feature extraction task followed by an SVM to perform the
Havaei et al. utilized Convolutional Neural Network classification task. The author reveals the following results:
(CNN) for performing brain tumor segmentation tasks accuracy 96.51%, specificity 94.2%, and sensitivity 97.72%.
[19]. The experimental results show a 0.88% dice In another research article, the segmentation method based
score and also reduce the segmentation time. Pereira on the fusion of RF and SVM (RF-SVM) was implemented
et al. established a CNN-based automated brain tumor for tumor lesions. It is the two-stage cascaded framework,
delineation system with a 0.88% dice score [20]. where RF learns from tumor labels, and the resultant output
Researchers demonstrated a 3D-CNN model for the is fed to the SVM to classify the labels [30]. Zhao et al. also
segmentation of brain lesions with a DSC of 0.89% in utilized CNN and conditional random fields (CRFs) hybrid
[21]. An automated brain tumor deep neural network technique for efficient brain tumor segmentation. A dice
(DNN) based model was proposed for MRI scans [22]. score of 0.87% was achieved [31]. The review of progress
The 0.72% dice score was observed. A fully convolutional for brain tumor segmentation in the years 2010–2020
residual neural network (FCR-NN) is implemented is described in Table 2. The overview of freely available
for the tumor segmentation, with a 0.87% dice score databases for brain tumor segmentation is shown in Table 3.
[23]. Similarly, DNN based automated segmentation Through this survey, a comparative study of more than
method, with a dice score of 0.87% was employed by fifty segmentation approaches between the years 2010–2020
[24]. In addition, the author utilized DNN, i.e., Fully has urged us to conclude the following findings:
Convolutional Network (FCN), for pixel-wise image
representation for tumor semantic segmentation. The (1) It is evident from Table 2 that various methodologies
MRI scans utilized in the study include T1, T1c, T2, and algorithms have been developed for brain tumor
and Flair. In this way, the tumor regions are segmented segmentation in the past few years. Some fusion/hybrid
more accurately [25]. Despite the benchmark results algorithms are utilized, whereas some are the modified
achieved by deep learning algorithms in the brain version of its basic.
tumor segmentation domain, only a deep learning-based (2) The shift towards the utilization of hybrid techniques
method still has limitations for accurate automated brain is noticeable. However, some researchers are still
tumor segmentation. For instance, the limited capacity struggling with simple ML and DL algorithms to
to delineate visual objects and impotent to consider the achieve touchstone performance. Ito et al. worked on the
spatial consistency and appearance of segmentation segmentation of brain tumors using a semi-supervised
results [26, 27]. deep learning technique from the MR images [32].
The need for an hour is to design architecture for This technique has attained improved results. Tianbao
brain tumor segmentation that can effectively segment Ren et al. developed an automated Kernel-based FCM
the brain tumor regions, require less memory, undergo with a weighted fuzzy kernel clustering model that
fast computation, and improve boundary delineation. enhances brain image segmentation performance [33].
Therefore, the trend of research has been shifted towards Results illustrate that the proposed combined algorithm
utilizing efficient hybrid techniques. achieves an improved misclassification rate which was
13
4876 S. Ali et al.
Table 2 A summary of segmentation methods for brain MR images over the years 2010–2020
Author Year Segmentation method Task performed Performance (%) Paradigm shift Dataset
[40] 2010 FCM based method Brain segmentation – ML Private

[41] 2010 2D-Brain extraction algorithms Brain segmentation – ML Private
(BEA) and 3D-BEA algorithms
[42] 2010 Expectation maximization Brain segmentation Acc = 95.13 ML Private
algorithm
[16] 2011 Integration of self-organizing Brain segmentation WM = 0.70 GM = 0.78 Hybrid IBSR
map (SOM) and learning vector
quantization (LVQ)
[43] 2011 Prossibilistic C-mean (PCM) Brain segmentation Correct = 79 Not Correct = 16 Hybrid Private
clustering+type-II fuzzy models
[44] 2011 SVM based model Brain segmentation Total error = 5.6 ML Private
[45] 2011 Implemented expectation Brain segmentation Acc = 98 Hybrid Private
maximization and Gaussian
mixture (EM-GMM) model
[46] 2011 Active contour methods + SVM Brain segmentation – ML Private
based classification
[47] 2011 FCM clustering method Brain segmentation – ML Standard
[48] 2012 Feedback pulse-coupled neural Brain segmentation SN = 100 SP = 92.8 Acc = 99 ML Harvard
network
[49] 2012 Genetic algorithm (GA) + SVM Brain segmentation SN = 92.3 SP = 99.6 Acc = 99.3 Hybrid Private
based classification
[17] 2012 Generalized rough FCM clustering Segmentation Avg=10.33 ± 2.96 ML BrainWeb
[15] 2013 SOM clustering algorithm using Lesions segmentation SM = 59.0 CC = 53.0 AUC = 76.0 ML Public
prioritization techniques
[50] 2013 Growing hierarchical SOM + Tumor segmentation SN = 81.7 SP = 99.8 ML IBSR
multi-objective-based feature
selection
[51] 2014 Proposed k-nearest neighbors Tumor segmentation ET = 0.53 TC = 0.80 WT = 0.87 ML BRATS 2013
(KNN) and CRF based network
[52] 2014 Region growing technique Tumor segmentation Acc = 80 Dice = 92 ML Public
integrated with cellular automata
edge detection network
[53] 2014 Cellular neural network Brain segmentation Dice Acc = 93 DL Private
[54] 2014 Local independent projection-based Tumor segmentation ET = 0.58 TC = 0.68 WT = 0.84 DL BRATS 2013
classification (LIPC)
[55] 2015 Hybrid model Gaussian Mixture Tumor segmentation ET = 70 TC = 82 WT = 87 Hybrid BRATS 2015
and convolutional restricted
Boltzmann machines (cRBMs)
[56] 2015 2D-CNNs for 3D voxel Tumor segmentation Acc = 0.88 DL BRATS 2013
classification
[57] 2015 Concatenated RF and R project for Tumor segmentation ET = 0.74 TC = 0.78 WT = 0.87 ML BRATS 2013
statistical computing
[30] 2016 RF-SVM cascaded algorithm Tumor segmentation Score = 72 Hybrid BRATS 2012
[23] 2016 Convolutional Residual Neural Tumor segmentation ET = 0.72 TC = 0.81 WT = 0.87 DL BRATS 2016
Network
[58] 2016 Stacked auto-encoder + Stacked Tumor segmentation Acc = 98.04 Hybrid Private
denoising auto encoder
[20] 2016 CNN Tumor segmentation ET = 0.75 TC = 0.65 WT = 0.78 DL BRATS 2013
[19] 2017 DNN Tumor segmentation ET = 0.73 TC = 0.78 WT = 0.85 DL BRATS 2013
[59] 2017 2D fully convolutional neural Tumor Segmentation ET = 0.75 TC = 0.73 WT = 0.88 DL BRATS 2017
networks (FCNNs)
[36] 2017 Deep convolutional neural network Tumor segmentation ET = 0.55 TC = 0.69 WT = 0.81 DL BRATS 2017
(DCNN)
[60] 2017 Neural network + Holistically- Tumor segmentation ET = 0.69 TC = 0.60 WT = 0.86 Hybrid BRATS 2017
nested edge detection (HED)
13
Table 2 (continued)
Author Year Segmentation method Task performed Performance (%) Paradigm shift Dataset
[61] 2017 U-Net based deep fully Tumor segmentation ET = 0.69 TC = 0.60 WT = 0.86 DL BRATS 2015
convolutional networks
[37] 2018 DCNN Tumor segmentation Dice = 0.76 DL TCIA
[62] 2018 Modified U-net convolutional Tumor segmentation DSC = 0.87 DL BRATS 2017
network
[38] 2018 Implemented end-to-end Tumor segmentation ET = 0.81 TC = 0.76 WT = 0.89 DL BRATS 2017
incremental DNN method
[63] 2018 Deep learning based DeepMedic Tumor segmentation ET = 0.78 TC = 0.62 WT = 0.86 DL Public
model
[64] 2018 Integration of FCNNs and CRFs Tumor segmentation ET = 0.62 TC = 0.67 WT = 0.84 DL BRATS 2015
[65] 2019 A mix-pooling CNN model with Tumor segmentation ET = 0.71 TC = 0.75 WT = 0.80 Hybrid BRATS 2013
FCRF
[34] 2019 CNN algorithm Tumor segmentation Acc = 89 DL Private
[66] 2019 Deep learning system Tumor segmentation Acc = 0.91 DL Public
[67] 2019 CNN based U-Net Tumor segmentation Acc = 89 DL BRATS 2017
[39] 2019 Kernel-based CNN with M-SVM Tumor Segmentation Acc = 84 Hybrid Private
deep learning algorithm
[35] 2019 Improved CNN with non- Tumor segmentation Acc = 90.98 Hybrid BRATS 2015
quantifiable local texture feature
[68] 2019 Proposed DE embedded OTSU Tumor segmentation Acc = 94.73 Hybrid IBSR
method and NN
[69] 2020 Ensemble 3D U-Net Tumor segmentation ET = 0.87 TC = 0.91 WT = 0.92 Hybrid BRATS 2018
[70] 2020 RescueNet using residual and Tumor segmentation ET = 0.87 TC = 0.94 WT = 0.94 Hybrid BRATS 2015
mirroring principles
[71] 2020 Tumor GAN Tumor segmentation WT = 0.85 TC = 0.79 DL BRATS 2017
[72] 2020 Path aggregation U-Net (PAU-Net) Tumor segmentation ET = 0.67 TC = 0.72 WT = 0.88 DL BRATS 2018
model
[73] 2020 A novel technique to integrate Tumor segmentation ET = 78.2 TC = 82.3 WT = 90.8 Hybrid BRATS 2018
location information fusion
[74] 2020 One-pass Multi-task Network Tumor segmentation ET = 0.80 TC = 0.79 WT = 0.91 DL BRATS 2018
(OM-Net)
[75] 2020 Introduce a two-step dragonfly Tumor segmentation Acc = 98.15 Recall = 95.4 DL BRATS 2017
algorithm (DA) clustering Precision = 93.5
technique
[76] 2020 Combined Deep supervised 3D Tumor segmentation ET = 0.74 TC = 0.80 WT = 0.89 Hybrid BRATS 2017
Squeeze-and-Excitation V-Net
(DSSE-V-Net) method
[77] 2020 CNN based U-Net Tumor segmentation Dice = 0.84 Acc = 0.92 Hybrid TCIA
less than 2.36%. In the deep learning area, Sundararajan (4) Mainly prior knowledge combined with artificial
et al. use a CNN algorithm for tumor segmentation intelligence led to the framework’s design with
with an accuracy of 89% [34]. Wu Deng et al. utilizes enhanced brain tumor segmentation results.
a basic CNN model with minor modifications [35]. The (5) It has been observed that the most commonly employed
accuracy of the model was enhanced to 90.98%. ML methods are SVM, FCM, and C-means, while the
(3) Over the last few years, deep learning algorithms commonly employed DL method is CNN and DCNN.
are the top performers, especially DCNN [19, 36– Hybrid techniques include the combination of two or
38]. However, the main limitation of DCNN is a more ML or DL techniques. For instance, Thillaikkarasi
dependency on massive training data with expert and Saravanan utilized kernel-based CNN with M-SVM
radiologists annotations from different institutions. It deep learning algorithm for tumor segmentation with
is a pretty tricky task. a dice score of 0.85% [39]. The guiding principle of
13
4878 S. Ali et al.
Table 3 Overview of publicly available databases and their modalities, number of patients and dataset sources
Database Available modalities Images/patients Dataset sources
BRATS2012 T1,T1-weighted(T1Gd),T2-weighted(T2),and T2 FLAIR 45 patients https://www.smir.ch/BRATS/Start2012

BrainWeb T1,T2-,proton-density(PD-weighted) 20 patients http://www.bic.mni.mcgill.ca/brainweb/
Harvard T1-w, T2-w, CE T1-w and FLAIR 13000 brain MRIs http://www.med.harvard.edu/aanlib/
IBSR T1-weighted 39 patients https://www.nitrc.org/frs/?group_id=48
Figshare T1-weighted contrast-enhanced 233 patients https://figshare.com/articles/dataset/
brain_tumor_dataset/1512427
TCIA T1,T2-weighted(T2) 19 patients https://www.cancerimagingarchive.net/
BRATS2017 T1,T1-weighted(T1Gd),T2-weighted(T2),and T2 FLAIR 285 patients https://www.med.upenn.edu/sbia/brats
2017/registration.html
hybrid techniques in achieving a robust, accurate, and analysis (ICA), PCA, and linear discriminant analysis (LDA)
low-cost solution for tumor segmentation. are commonly applied [80]. The amalgamation between
the feature extraction and feature reduction led to the
development of a CAD system that will classify the images
with clinically acceptable accuracy utilizing few features
extracted via low computation resources. Such a developed
3 The Development for Brain Tumor Feature CAD method can be effectively utilized as a secondary
Extraction and Classification Techniques diagnostic tool for brain tumor classification.
(2010–2020) Moreover, to reduce the intensity variation of the MRI
scans, various filters, feature extraction, and selection
The process of allocating the input features to various or their fusion are performed. For instance, the Gabor
categories/classes is termed classification. Before brain wavelet features approach is performed to acquire texture
tumor classification and detection by CAD system, a key information of the MRI scan. Kernel Principal Component
stage is feature analysis and feature selection. The curse of Analysis (KPCA) lowers the redundancy by selecting only
dimensionality is surmounted by reducing the redundancy a small subset of the features, and Gaussian Radial Basis
of feature space through discriminating, appropriate, and Function provides eminent information from any set of
compelling feature sets. The feature extraction step requires features [81]. However, in the pre-trained CNNs method,
many MRI slices (such as axial, coronal, and sagittal planes) the fine-tuning-based feature extraction is employed [82].
with ground truth.
3.2 Tumor Classification
3.1 Feature Extraction
The procedure of categorizing tumor grade or tumor as
It is the process of converting an MRI scan into a set of benign or malignant is called tumor classification. Owing
its features for classification purposes. Extracting the to the distinct shape, location, size, and contrast of tumorous
set of distinctive features is a challenging task. Various cells, brain tumor classification is challenging. Acquiring
feature extraction techniques are used for this purpose, superior classification mainly depends on the extraction of
including principal component analysis (PCA), spectral an optimum set of features for classification and the choice of
mixture analysis (SMA), texture features, Gabor features, a suitable classifier. The factors like classification accuracy,
nonparametric weighted and decision boundary feature computational resources, and algorithm performance should
extraction, feature based on wavelet transform, discriminant be considered to choose the optimum classifier.
analysis (DA), and so on [78], as shown in Fig. 3. Recently, Input patterns are classified into analogous classes via
the most efficient CAD system performs DWT (discrete two types of classification techniques: (1) unsupervised
wavelet transform) [79] to acquire the wavelet coefficients classification, which includes hierarchical clustering, FCM,
at various levels. K-mean clustering, SOM, etc. (2) supervised classification,
Feature reduction is an additional step to lessen the data which includes decision tree, SVM, LDA, KNN, Bayesian
dimension. For this purpose, independent components classifier, etc. [83], as illustrated in Fig. 3. Unsupervised
13
classification is the recognition of natural classes or groups accuracy of 92.3%, which was more enhanced to 97.8% by
in multi-spectral data. This classification required no prior applying multiclass SVM. Mohsen, H. et al. implemented
knowledge; it recognizes classes as distinct units and has FCM followed by DWT, combined with the DNN for the
fewer chances for operator error. tumor classification, utilizing 66 T2-W MRI scans [108].
However, in supervised classification, the samples of The model depicted a 96.97% classifying rate. Moreover,
known identity are used to classify the samples of unknown Chaplot S. et al. proposed the novel approach of combining
identity. Supervised classification requires prior knowledge, wavelets with SOM and SVM for tumor classification
labels are provided for the input dataset, and significant by the use of 52 T2-W MRI scans [127]. The proposed
errors might be detected. The overview of detailed literature method showed more than 94% accuracy for SOM and 98%
related to various feature extraction and classification classification accuracy for SVM.
methodologies for MRI images published during 2010–2020 Through the survey of Table 4 and literature study, it was
is presented in Table 4. observed that:
3.3 ML, DL, and Hybrid‑Based Feature Extraction (1) Discrete wavelet transform, PCA, and texture analysis
and Classification (TA) is the commonly employed feature extractor
methods.
The brain tumor classification techniques followed the same (2) CNNs attain high classification and prediction
paradigm shift as described earlier in the segmentation performance when the algorithm is already pre-
techniques. As the 10s started, a considerable number of trained as a feature extractor. In brain tumor patients,
researchers focused on conventional ML-based classifiers the overall survival time prediction played a significant
for classification. role in the deep feature extractor methods. Numerous
For instance, Alfonse et al. use the SVM for automated methods that embody feature choice, feature pooling,
tumor classification using MR scans [123]. Firstly brain and data augmentation networks are included in CNN
images are segmented employing adaptive thresholding. + activation feature methodology [124].
Secondly, features are extracted using Fast Fourier (3) Hybrid systems (combined with a pre-feature extractor
Transform (FFT), then minimal redundancy maximal and different deep learning and machine learning
relevance methods are used for feature selection. This approaches) are commonly employed for efficient
technique achieved 98.9% classification accuracy. In SVM, tumor classification.
the classification of different points based on proximity (4) Most of the CAD systems for the classification
accompanied by splitting hyperplane required more of brain tumors are imperfect in terms of higher
execution time to calculate linear or quadratic complications. complexity, high dimensions of feature vectors, and
As the decade progressed, deep learning methodologies high generalization capability. Even though significant
are employed for classification purposes. In an article, efforts have been made in the last decade, much work
features are extracted using segmentation algorithms, is still needed to establish a CAD system with a high
i.e., Dense CNN, while the features are classified using success rate.
recurrent neural networks (RNN) [112]. Convolutional and (5) We believe the hybrid intelligent systems designed by
fully connected networks are the most commonly employed integrating machine learning approaches with other
DL-based classification models of brain tumors [124]. methodologies offer a highly proficient, accurate
Recently (in the 20s), the emerging hybrid techniques classification system. It appears to give higher
are commonly employed classification methods. Moreover, classification accuracy in the range of 95%-100%.
hybrid intelligent systems are also implemented for the
design of classifiers utilizing soft computing approaches.
Soft computing intelligent paradigms include neural 4 Statistical Analysis of Conducted Research
networks, bio-inspired algorithms such as genetic algorithm
(GA), used to build robust classification systems. Sujan M The prime objective of this comprehensive section is to
et al. proposed a combined technique with k-mean and acquire the answer to various queries:
FCM [125]. This technique implemented a median filter
for MR brain images denoising and brain surface extractor 4.1 Commonly Employed MRI Modalities
for features extraction; then clustering is done through an
integrated hybrid method. Deepak and Ameer developed a Previous studies already reveal that MRI is the most
CNN-based GoogleNet transfer learning classification model commonly used modality for performing brain tumor
to classify brain tumors including glioma, meningioma, and classification and segmentation tasks [128]. However, this
pituitary [126]. The proposed algorithm attained a better research attempts to highlight the frequently used MRI
13
4880 S. Ali et al.
Table 4 State-of-the-art feature extraction along with classification methods for brain MRIs
References Year Feature extraction Classification Performance (%) Paradigm shift Dataset
[84] 2010 DWT + PCA ANN and KNN SN = 98 SP = 90 Acc = 97 Hybrid Harvard
[85] 2010 Texture based feature removal SVM and ANN Acc = 99 ML Harvard
[86] 2010 SGLDM +WT GA + SVM SN = 92 SP = 95 Acc = 100 Hybrid Harvard
[87] 2010 DWT + PCA ACPSO+NN Acc = 98.7 ML Harvard
[88] 2010 Gray level features Rule-based Level set SVM SN = 93.5, 88.7, 81.5, Acc ML Private
= 84.3
[89] 2011 DWT + PCA BPNN SN = 100 SP = 98 Acc = ML Private
99.8
[90] 2012 PCC, ICA and PCA SVM SN = 89 SP = 84 Acc = 85 ML Private
[91] 2012 PCA + LDA ANN + KNN Acc = 100 Hybrid Harvard
[92] 2012 DWT + PCA KNN and ANN SN = 96 SP = 97 Acc = 98 ML Private
[93] 2013 2D-DWT PNN SN = 83.3 SP = 100 Acc = DL Private
95.7
[94] 2013 2D DWT + GARCHA KNN+SVM SN = 98.2 SP = 98.2 Acc Hybrid Harvard
= 97.6
[95] 2013 SC-ICA and ICA SVM SN = 76.4 SP = 99.9 Acc ML Harvard
= 98
[96] 2013 PCA BPNN Acc = 96.3 DL Private
[97] 2013 Gabor wavelets CCANN SN = 92.5 SP = 89.5 Acc DL Private
= 91.8
[98] 2014 – Neuro-fuzzy SN = 88.9 SP = 88.9 Acc ML Private
= 95.6
[99] 2014 PCA SVM SN = 100 SP = 50 Acc = 84 ML BrainWeb
[100] 2014 FDCT and GLCM PNN-RBF Acc = 99.7 DL Private
[101] 2014 Polynomial domain Normalized cross correlation Acc = 99.8 ML Private
(NCC)
[102] 2015 CNN activations trained by CNN Acc = 97.5 DL Harvard
ImageNet
[103] 2015 2D-DWT SVM SN = 98.1 SP = 92 Acc = ML –
97.7
[104] 2015 Grayscale, symmetry and SVM-KNN SN = 100 SP = 93.7 Acc Hybrid Harvard
texture = 98
[105] 2015 DWT-SGLDM GA-SVM Acc = 95.6 Hybrid BrainWeb
[106] 2016 DWT Genetic algorithms Acc = 95.6 – Private
[107] 2016 Watershed transform CNN Acc = 98.5 DL Private
[108] 2017 DWT and PCA DNN and FCM Acc = 0.98 Hybrid Harvard
[109] 2017 CNN CNN + FCNN Acc = 91.4 Hybrid Private
[110] 2017 PCA SVM, RF, KNN, LOG, MLP Acc = 83 ML TCIA
[111] 2018 CNN ConvNets Acc = 97 DL Private
[112] 2018 DenseNet based auto-encoder DenseNet Acc = 92.1 DL Private
[113] 2018 Gabor-wavelet features ELM-LRF SN = 96.8 SP = 97.1 Acc ML Harvard
= 97.1
[114] 2018 CNN Capsule Networks Acc = 86.5 DL Private
[115] 2019 CNN based feature extraction SVM, GA, CNN Acc = 84.5 83.6 91 DL UCI
[116] 2019 DWT and PCA SVM SN = 0.79 SP = 0.91 Acc ML BRATS 2015
= 0.86
[117] 2019 Pre-train CNN model Alex and Google networks SN = 0.99 SP = 1.00 Acc Hybrid BRATS 2017
= 0.99
[118] 2019 Circular context-sensitive Random forest SN = 0.84 SP = 0.71 Acc ML BRATS 2015
(CCS) = 0.87
[119] 2020 Upsamples feature maps Multi-Scale 3D U-Nets SN = 0.86 SP = 0.86 Acc DL BRATS 2015
= 0.85
[120] 2020 CNN based features CNN based on GoogLeNet SN = 94 SP = 97 Acc = 0.98 DL Private
13
Table 4 (continued)
References Year Feature extraction Classification Performance (%) Paradigm shift Dataset
[121] 2020 Texture and HOG features ANN SN = 89 SP = 94 Acc = 92.1 Hybrid Private
[122] 2020 Gray Level Co-occurrence FBSO SN = 95.7 SP = 88.9 Acc Hybrid BRATS 2018
Matrix (GLCM) = 93.8
Fig. 3 Illustration of decade analysis brain tumor diagnosis and general CAD system for the brain tumor segmentation and classification
modality. The fact that among the different MRI modalities, 4.2 Year‑wise Increment in Publication for Brain
the most commonly used single modality is T1-W which Tumor Analysis
is 18% of total reviewed studies. While the image dataset
is constituting T1, T2, T1-CE, and FLAIR are the current A range of well-renowned databases, including IEEE
winners by acquiring 32% of the total publications illustrated Xplorer, Science Direct, PubMed, Scopus, were searched
in Fig. 4. utilizing the brain tumor/CNS cancer? keyword combined
13
4882 S. Ali et al.
4.3 Commonly Employed Datasets
Acquiring the brain tumor dataset is the primary task.

Certain datasets are easily and freely assessed for
experimentation like BrainWeb, BRATS [129], Harvard,
The Cancer Imaging Archive (TCIA), Oasis, etc. Most of
the publications were exploited BRATS data sets which
constitute 30% of the current reviewed articles. However,
the private dataset ranked second with 25% of the
reviewed article, as shown in Fig. 6. This is because before
the launch of the representative BRATS dataset research
community was utilizing the private dataset acquired from
the local laboratories and hospitals. However, the research
community is still utilizing the private dataset to compare
Fig. 4 Commonly used MRI scans modalities (in %) for brain tumor
analysis reviewed in the current study their proposed models.
with MRI for the segmentation, classification, and 4.4 Classifier‑Based Publication Statistics
detection tasks through machine learning, deep learning,
and hybrid approaches. Afterward, the most relevant 248 Deep learning technology resulted in great realistic
articles were scrutinized between the years 2010–2020. performances in brain tumor image analysis [70]. CNN
Despite the benchmark results obtained so far. The era can be known as an archetypical classifier owing to
of brain tumor analysis has propelled intense research immense usage in the prognosis of various diseases
efforts in the last and present decade to acquire the robust such as brain tumor classification, segmentation, and
computer vision technique. The yearly development of soft detection [20]. Moreover, the number of CNN-based
computing viz. artificial intelligence from application- architectures (CNN combined with other architectures) has
driven and methodology perspectives has been assessed developed between the years 2010–2020 [14, 130–132].
from the rapidly increasing year-wise publications. The The retrospective analysis reveals that the CNN-based
year-wise distribution is illustrated in Fig. 5. Undoubtedly, architectures have enhanced the performance accuracy for
the ?0s would be the decade of computer vision arena for brain tumor detection. In this review, 64 CNN and CNN-
brain tumor segmentation, classification, and detection based architectures were utilized for brain tumor analysis;
tasks. it forms 26% of the designated study. However, the SVM
(ML algorithm) ranked second with 39 publications, which
constitute 16% of current research. Fig. 7 presented the
core statistics of classifier based publications.
Fig. 5 Year-wise, rapidly increase in publication for brain tumor

analysis Fig. 6 Most popularly employed databases
13
and third, as shown in Fig. 8. This is because radiologists

mostly find it difficult to segment the tumor from an MR
image to classify the tumor type. It is also a time-consuming
task. Therefore, more research is carried out on brain
tumor segmentation to assist radiologists and clinicians in
diagnosing brain tumors and their sub-regions. Automated
segmentation also helps in distinguishing tumor regions
from the non-tumor region in no time.
4.6 Summarization of Previously Reported Works
This review attempts to summarize the various reviews

published between 2010 and 2020 on brain tumor
segmentation, classification, and detection to the best of
our knowledge. Most of the existing literature covers the
Fig. 7 Classifier based publication statistics
conventional ML-based methods for brain tumor analysis,
as depicted in Table 5. In comparison, this study will cover
a few ML-based methods with special emphasis on deep
learning and hybrid-based methods. It depicts that the
present review attempts to address all the limiting issues
and lacks in the existing surveys.
5 Current Trends on DL‑Based Brain Tumor

Diagnosis
In contrast to the machine learning algorithms, the deep

learning methods showed more usage for the segmentation
and classification of brain tumor MR images. Herein,
we made a comprehensive study to show the significant
advancement in the deep learning approach over the year
2010–2020.
Fig. 8 Most studied CAD task
A deep learning network has multiple hidden layers of
the network representing input data with various layers of
4.5 Most Studied CAD Tasks extraction, supporting the reduction of many problems in
conventional machine learning methods. Furthermore, deep
From the two-decade analysis, it comes to know that brain learning methods have features such as self-learning and
tumor segmentation is the most studied CAD task so far. generalization ability, enabling good quantitative analysis of
However, classification and detection are ranked second medical imaging features. Due to these characteristics, deep
Table 5 Comprehensive analysis and comparison of our study with existing surveys for brain tumor
Review articles Literature Theme of study Learning method No. of Analyzed Dataset Future
converge years current converge prospects
range covered trends
[133] 2003–2012 Segmentation and classification Conventional 9 × × ×

[134] 2009–2013 Tumor classification Conventional 4 × × ×
[135] 2004–2013 Segmentation and classification Conventional 9 × × ×
[80] 2003–2013 Segmentation and classification Conventional 10 × ✓ ✓
[136] 2008–2016 Segmentation and classification Conventional 8 × ✓ ✓
[137] 2016–2019 Tumor classification Deep learning 4 ✓ ✓ ✓
This review 2010–2020 Segmentation, feature Deep learning and hybrid 11 ✓ ✓ ✓
extraction, and classification
13
4884 S. Ali et al.
learning-based techniques achieve accurate detection results The diversity of tumors in the human brain demands high
of neurological disorders and are greatly acknowledged in accuracy, which surges complexity; in this situation, input
the medical image processing domain [138]. Many CAD MR image and its characteristics play an important role. In
systems incorporated deep learning-based segmentation another article, the author proposed a novel segmentation
and classification approaches in medical image processing, method based on a multi-modal super voxel with the RF
including chest, breast, pulmonary nodules, and brain tumors classifier [150]. To evaluate the performance of the method,
[58, 139–141]. BRATS and clinical datasets include; MRI scans and
Different deep learning-based networks such as DCNNs diffusion tensor imaging (DTI) datasets are used in such
[142], CNN’s, and auto-encoders [143] are designed for a way to classify each super voxel as normal or tumorous
effective and accurate segmentation, feature detection, (core or edema) scan. Sensitivity and dice score is used for
and classification of brain tumors via MRI scans. Many performance measures 86% and 0.84% accuracy is reported
researchers with great motivation are doing more research for clinical dataset while better accuracy 96% and 0.89%
in different institutes and developing new algorithms to are achieved for multi-modal images taken BRATS. Iqbal,
improve performance. Sajid, et al. introduced three different approaches for the
A novel brain tumor segmentation technique named segmentation of brain tumors using BRATS 2015 MR scans.
WMMFCM was presented to reduce the challenges of Interpolated Network (IntNet), show supremacy over Skip-
FCM by using three different stages, including wavelet Net, and SE-Net, IntNet reached top values 90%, 88%, 73%
multi-resolution (WM), morphological pyramid (M), and for three considered factors dice coefficient, sensitivity, and
FCM clustering technique. BrainWeb (152 MR scans) and specificity respectively [151].
BRATS (81 glioma images) datasets are employed to verify Moreover, various techniques are developed to enhance
the performance of the proposed architecture. Implemented and outdo the CNN capabilities in terms of accuracy,
algorithms achieved 97.05% accuracy and 95.85% accuracy computational time particularly (when handling massive size
for BrainWeb and BRATS, respectively [144]. Researchers datasets), and hardware specifications [108]. Researchers
proposed a tumor segmentation technique using semi- designed the FCM technique to segment 66 T2-W MRI
automatic software [145] to register multi-modal 159 T1-W scans into four groups: sarcoma, metastatic bronchogenic
and T2-W low-grade gliomas MR scans [146]. The CNN carcinoma normal, and glioblastoma brain tumor. This
model was applied for the classification task and obtained study used a combination of DWT with the DNN algorithm
a cross-domain performance of 87.7%, 93.3%, and 97.7% and achieved 96.97% classification accuracy. In the present
for accuracy, sensitivity, and specificity. In another report, scenario, improved CNN is employed to resolve the issue
the author uses the concept of CNN with small kernels to of the manual diagnosing process. Another presented an
overcome the problem of overfitting and provided small enhanced CNN (ECNN) model integration of the BAT
weights in the CNN architecture [20]. Firstly starts with algorithm to segment brain tumors automatically [152]. They
infrequent intensity and patch normalization; the research implemented an approach that is also useful in controlling
showed efficiency and effectiveness with the combinations over-fitting using function loss of BAT and small kernels
of data augmentation. Secondly, training of patches is features of ECNN. The accuracy of ECNN is found 3% more
conducted through artificially rotating the images. Finally, classical CNN. The author of [38] proposed an end-to-end
a defined threshold is employed to enforce volumetric incremental EnsembleNet algorithm for glioblastomas
limitations, which means removing small clusters that may segmentation and obtained a dice score of 0.88% on BRATS
be predicted as trivial tumors. Experimental results achieve 2017.
84% accuracy for the baseline network while 88% accuracy As the 20s decade progressed, the number of publications
rate was achieved by U-net. has been increased where the combination of two or more
Over the past few years, the DCNN model has also deep learning or machine learning techniques is employed
shown significant advancements. The authors designed an to overcome the limitations of both individual techniques. In
automatic DCNN model aiming to overcome the issue of this way, a new robust hybrid system can be designed with
over-fitting by combining DCNN with max-out and drop-out improved performance metrics.
layers [147]. BRATS 2013 dataset MR scans collection is
used to evaluate the performance of the proposed algorithm.
Results demonstrated that the dice similarity coefficient was 6 Current Trend on Hybrid‑Based Brain
achieved 80%, 67%, 85% for WT, TC, and ET, respectively. Tumor Diagnosis
Few studies are published to overcome the issues of
segmentation using an improved version of DCNN [148, Hybrid algorithms are a combination of two or more
149]. Another important problem investigated by a large algorithms used to achieve superior performance compared
group of authors is the existence of multiple tumors [108]. with single methods. Such hybrid algorithms aim to
13
overcome the shortcomings of one method by a second Development in hybrid systems is increased further
alternative method. SVM is a commonly acknowledged through the amalgamation of more than two approaches.
approach in different applications and integrated with Fuzzy k-means (FKM) poorly supervised problems
conventional and also with current trending approaches. associated with the huge amount of data. To increase the
A novel hybrid wavelet separately and SOM system for abilities of data supervision, FKM is combined by SOM to
the tumor segmentation of 52 axial T2-W MR slices was develop a tumor detection method [156]. SOM supports early
designed by [127]. SVM technique is applied to classify clustering and decreases the dimensionality of input images.
healthy and unhealthy images affected by Alzheimer’s Harvard brain repository is used to evaluate the performance
disease. Significant classification performances, 94% and of the proposed model on MRI scans. This integration
98%, were reported for SOM and SVM, respectively. Results of algorithms achieves 96.18% accuracy and 87.18%
determine the effectiveness of the implemented system. sensitivity. Vishnuvarthanan et al. designed a combined
A combination of SVM with GA is presented to segment optimization technique with FKM to efficiently process MR
normal and tumor regions. The spatial gray level dependence image sequences using the bacteria foraging optimization
method (SGLDM) is applied by [88] for texture features (BFO) method integrated through a modified FKM approach
extraction. Harvard medical dataset consisting of 83 brain [157]. The suggested model produces promising results
images including (29 normal, 22 malignant, and 32 benign with 97.14% sensitivity and 93.94% specificity. Another
tumor slices) was used for experiments. Results show the group designed a hybrid clustering system by merging
varying performance ranging 94.44% to 98.14% accuracy three different algorithms (k-means, FCM, and SOM) to
and 91.9% to 97.3% for sensitivity. automatically segment brain tissue [158]. Firstly, pixel
FCM clustering is a well-known technique in the hybrid intensity values are fixed to improve image resolutions,
class. Rajendran, A. and R. Dhanasekaran proposed an secondly applied a super-pixel algorithm to link pixels with
effective region-based fuzzy clustering segmentation related intensity into objects. Finally, extracted features
approach termed enhanced possibilistic FCM [153]. The and their labels produced by proposed clustering methods
proposed method is implemented to control the initialization are used to train a neural network (NN) for classification.
and weak boundaries challenges in region-based methods. Results achieved 98.10%, 98.97%, and 79.66% for accuracy
The integration is applied on 15 CE-T1W and FLAIR and specificity. These results outperform other clustering
scans to classify tumors. Results reported average accuracy methodologies. Namburu, A. et al. proposed a novel soft
indices, 95.3%, and 82.1% for similarity and Jaccard, fuzzy rough c-means (SFRCM) technique to extract soft
respectively. Another works to detect and classify brain tissues like WM, GM, and CSF [159]. To evaluate the
tumors by merging FCM clustering with an SVM classifier efficiency of the SFRCM method 20, 10, and 20 images of
[28]. The model’s performance is compared on different BrainWeb, BRATS, and IBSR databases respectively were
sets of images of 120 patients using ANN and SVM. SVM utilized. MR scans are employed to categorize high-grade
classifier performs well on small datasets, while for larger glioma and achieved 94.04% accuracy.
datasets, ANN performs better. Abdel-Maksoud et al. This comprehensive study reveals that the hybrid
explore a novel combination called K-means Integrated techniques outperform the deep learning approach for brain
with Fuzzy C-means (KIFCM) clustering for brain tumor tumor segmentation and classification in the 20s decade.
segmentation using three datasets, including 81 images The whole summary of a few hybrid-based approaches
of BRATS, 152 images of BrainWeb, and 22 images from along their association with MR scans modalities, the task
digital imaging and communications in medicine (DICOM) performed, data set, and size influencing the performance
[154]. K-means decreases computational time, whereas evaluation is given in Table 6. The abbreviation list of this
FCM increases the accuracy of brain tumor recognition. paper is presented in Table 7.
KIFCM and FCM algorithms reported the same accuracy,
but KIFCM uses a small execution time and achieved 90.5%,
100%, and 100% accuracy for all datasets. An effective and 7 Discussion
novel detection model based on SOM incorporation with
LVQ is presented in [155]. For the experiments, 20 patients In retrospect, this review reveals that brain tumor analysis
of glial tumor MRI scans including; T1-W, T2-W, and attains state-of-the-art results in the domain of neuroimaging
FLAIR are used. This research also applied a skull stripping analysis. This review reports the substantial diversity of
algorithm on the IBSR database, which outperforms other various algorithms over the last decade and the past year
algorithms. However, experimental results on BRATS 2012 for brain tumor analysis in terms of segmentation, feature
obtained dice similarity of 91%, 87%, 96%, 61%, and 77% extraction, and classification. In consequence of immense
WM, GM, CSF, tumor, and edema. scrutiny following outcomes are made:
13
4886 S. Ali et al.
Table 6 Summary of hybrid techniques utilized for brain tumor MRIs

Ref Method MRI modality Results (%) Size/database Task
[14] Radial based function (RBF) – Acc = 97.1 SN = 91.9 SP 311 images Harvard Classification
and SVM classifier = 98
[87] DWT + PCA and adaptive T2-W Acc = 98.7 160 images Harvard Detection
chaotic particle swarm
optimization (ACPSO)
[126] GoogleNet + k-nearest T1-CE Men = 0.9 Glioma = 0.97 3064 images Figshare Classification
neighbors and SVM Pituitary = 0.98
[153] Enhanced Possibilistic Fuzzy CE-T1W, FLAIR JC = 82.1 SI = 95.3 15 images Private Segmentation
C-Means (EPFCM) with
parametric deformable
model
[131] SWT with Growing – Acc = 98.6 Recall = 0.98 5 Images BRAINIX Segmentation
Convolution Neural
Network
[130] GLCM and DWT + K-means T1, T1-c, T2, and FLAIR Acc = 99.8 SN = 89.7 SP 274 Patients BRATS 2015 Detection
= 99.9
[160] Multi-cascaded convolutional FLAIR, T1c and T2 WT = 88.2 SN = 90.7 SP 285 Patients BRATS 2018 Segmentation
neural networks and = 99.1
conditional random field
[154] K-means integrated with T1-W, T2-W, CE-T1W and Acc = 90.5 255 Images Public Classification
FCM FLAIR
[161] Tsallis entropy + Bat FLAIR Acc = 0.94 SN = 0.94 SP 110 Slices BRAINIX Detection
Algorithm = 0.96
[155] SOM + LVQ T1-W, T2-W and FLAIR WM = 91 GM = 96 CSF 20 Subjects Private Classification
= 61
[29] BWT + SVM T1-W, T2-W and FLAIR Acc = 96.5 SP = 94.2 SN 135 images BrainWeb Detection
= 97.7
[156] SOM + FKM T1-W, T2-W, CE T1W and Acc = 96.1 SN = 87.1 4 Patient Harvard Segmentation
FLAIR
[158] Super-pixel + (FCM-SOM) T1-W Acc = 98.1 SN = 79.6 SP = 28 patient Private Detection
98.9 DSC = 79.3
[162] Cognition based modified T1-W, T2-W, CE-T1W and Acc = 98 384 images Private Segmentation
level set and optimized ann FLAIR
classifier
[81] Adaptive firefly T2-W Acc = 99.8 SN = 97.2 SP 81 images BRATS 2015 Classification
backpropagation neural = 99.8
network
[76] Deep supervised 3D Squeeze T1, T1c, T2, and FLAIR ET = 0.74 WT = 0.89 TC 285 samples BRATS 2017 Segmentation
and V-Net (DSSE-V-Net) = 0.80
[151] IntNet, SkipNet, SENet T1-W, T2-W, CE-T1W and DSC = 90 SN = 88 484 images Private Segmentation
FLAIR
[163] K-means clustering and SVM T1-W, T2-W, CE-T1W and Acc = 94.06 274 Patients BRATS 2015 Classification
FLAIR
[164] Deep CNN T1-CE Acc = 98.43 SN = 99.03 SP Figshare Dataset Classification
= 98.17
[165] U-Net Deep learning model T1-W, T2-W, CE-T1W and DSC = 0.98 285 Patients BRATS 2018 Segmentation
FLAIR
[166] Multiscale CNN T1-W, T2-W, CE-T1W and Acc = 0.97 3064 images Figshare Classification
FLAIR
[167] VGG Stacked Classifier – Precision = 99.2 Recall = 253 images Kaggle Detection
Network 99.1 F1 score = 99.2
– In the year 2010–2015, conventional/machine learning extraction, and classification. From 2015 to 2019, the
algorithm was the trend for brain tumor diagnosis, traditional ML techniques have been replaced by deep
including all its aspects viz. image segmentation, feature learning-based algorithms for medical image analysis,
13
Table 7 List of abbreviations 75–95%. For example, Macyszyn et al. illustrated the
Description Abbreviations
classification of 105 high-grade gliomas (HGG) patients
into long and short-term categories utilizing the SVM
Magnetic resonance imaging MRI modal [168]. The accuracy of the architecture lies
Computer-aided diagnosis CAD within the range of 82–85%. In another report, Emblem
World Health Organization WHO et al. demonstrated an SVM classifier using histogram
Computed tomography CT data of 235 patients to predict glioma patient overall
Positron emission tomography PET survival [169]. The accuracy, sensitivity, and specificity
Region of interests ROIs were 0.79%, 78%, and 81% at six months and 0.85%,
Support vector machine SVM 85%, and 86% respectively at three years for overall
Random forest RF survival prediction. Sarkiss et al. undergo a systematic
Weighted fuzzy kernel clustering WFKC literature analysis (2000–2018) to provide evidence
Kernel-based fuzzy C mean KFCM for the utilization of machine learning techniques for
K-means integrated with fuzzy C-means KIFCM glioma detection [170]. The outcome reveals a sensitivity
Fuzzy C mean FCM between 78 and 93% and specificity between 76 and 95%.
Fuzzy k-means FKM However, the 20s are the emerging hybrid technique
Adaptive fuzzy K-mean AFKM era. Hybrid techniques included the amalgamation of
Machine learning ML one or more deep learning or hybrid techniques and
Deep neural network DNN were integrated into current neuroimaging analysis
Conditional random fields CRFs pipelines. These techniques have impactful results
Fully convolutional residual neural network FCR-NN regarding efficiency and accuracy of classification (in
Fully convolutional network FCN the range 92-100%. For instance, Nie et al.’s findings
Stationary wavelet transform SWT suggest that hybridization of traditional ML-based
Berkeley wavelet transform BWT approach named SVM with deep learning framework
Brain extraction algorithms BEA produce better results than bare models in terms of
Self-organizing map SOM accurate prediction of overall survival [171]. 3D CNN
Learning vector quantization LVQ (deep learning architecture), when combined with SVM
Prossibilistic C-mean PCM (machine learning architecture), attains an accuracy of
Genetic algorithm GA 96% for the prediction of OS in 69 HGG patients. The
Expectation maximization and Gaussian mixture EM-GMM comprehensive analysis was carried out to investigate the
Convolutional restricted Boltzmann machines cRBMs paradigm shift conventional/Machine learning → Deep →
Deep convolutional neural network DCNN Hybrid approach in the domain of brain tumor analysis.
Fully convolutional neural networks FCNN It is justified in the form of a graph presented in Fig. 9.
Principal component analysis PCA – The perfect design/architecture for the ML, DL, and
Spectral mixture analysis SMA hybrid-based techniques is not the sole determinant for
Discrete wavelet transform DWT achieving significant results. However, after the literature
Accuracy Acc survey carried out in this review, one can distill the high
Sensitivity SN accuracy architecture method for individual tasks with
Specificity SP its application area in the specific type of brain tumor
Area under curve AUC detection. The progress and development of high-
Similarity measure SM performance brain tumor CAD systems are presented
Correlation coefficient CC in Fig. 10. We found hybrid-based architectures and
Whole tumor WT deep learning approaches compete for performing brain
Tumor core TC tumor segmentation and classification task through
Enhancing tumor ET this analysis. However, the researchers that come by
White matter WM significant performance on ML or DL-based systems
Gray matter GM than the hybrid system might be due to trade outside
Cerebrospinal fluid CSF the network, such as normalization in pre-processing
techniques or data augmentation. For instance, Zhang
et al. investigated the gliomas grading in 120 patients
especially brain tumor investigation. With the critical [172]. Researchers could classify LGG and HGG with
analysis of Tables 1 and 2, it was observed that the ML 94-96% accuracy by utilizing combined SVM and
and DL-based systems have accuracy in the range of SMOTE (synthetic minority over-sampling technique).
13
4888 S. Ali et al.
hybrid system. The researchers who acquire robust

performances for hybrid approaches could do so
because they implemented the best augmentation and
pre-processing techniques. This is an easy way to boost
up the generalizability of the network without altering
the architecture. The key contributors to the significant
performance of any network are data augmentation
techniques, pre-processing techniques, hyper-parameter
optimization (i.e., learning rate and drop out), etc.
Moreover, changes in the network and receptive field’s
input size could help domain experts achieve good
performance results. However, unfortunately, so far 10s
there is a lack of exact techniques or the best suitable
hyperparameters for practical implementation. In the
brain image analysis domain, Bayesian methods to
Fig. 9 Graph illustrating the paradigm shift for brain tumor optimize hypermeters have not been implemented until
recognition tasks
the 20s.
– Brain tumor segmentation through radiotherapy treatment
Moreover, by observing different BRATS challenges, planning depends on manual segmentation of tumors
even using similar architecture for the same type of by expertise, making the process slow, arduous, and
network, extensively varying results were obtained. sensitive due to differences of opinion among physicians.
However, accuracy could also be increased by adding For the automatic and accurate segmentation of gliomas,
more layers to the framework [173, 174]. numerous tools and algorithms have been proposed in
– Designing hybrid architecture for specific task properties the 10s [175, 176], and the process continues in the 10s.
attains significant results than utilizing straightforward In this direction, to bring out efficient approaches and
machine learning or deep learning architecture. routes to solve the challenging problem, BRATS (multi-
Since selecting and integrating one or more systems modal brain tumor segmentation challenge) is organized
for attaining desired results could be possible in the annually [177, 178]. During half last decade (2015–
Fig. 10 Development toward brain tumor in last decade
13
2019), most of the exploited approaches of the BRATS expert annotations from multiple institutions is difficult.
rely on deep learning architectures, e.g., 3D-CNNs The BRATS challenge was organized using pre-operative
[179]. However, the top-performing approaches utilize various institutional data of MRI scans for brain tumor
ensembles of deep learning architectures [180, 181] sub-regions detection to provide the research community
or they even hybridized the various deep learning with a rich amount of images and a platform for comparing
architectures with algorithms like CRFs (conditional and evaluating various brain tumor algorithms. The dataset
random fields) [182] known to be as emerging hybrid is increasing every year. Despite much momentum gained
techniques. Moreover, in the BRATS 2017 and 2018 by ML and DL methods in terms of accuracy, the emerging
challenges, the top-performing methods include cascaded hybrid techniques have replaced them and have integrated
networks, multi-view and multiscale approaches [183] them successfully into neuroimaging pipelines.
generic U-Net architecture with data augmentation and One of the significant factors that hamper the
post-processing for brain tumor segmentation [184]. effectiveness of deep learning techniques is the requirement
Thus, emerging hybrid techniques are considered a of a bulky dataset to train the framework, thus requiring
robust and practical way to improve rugged segmentation more computation resources. However, acquiring such a
results. huge amount of data in the medical domain is challenging.
Therefore, to conquer this loophole, various architectures
have been designed so far to overcome the above problem.
For instance, the utilization of generative adversarial
network (GAN) [187]. This network requires scarce data
8 Future Research Directions for training. Even this fact still cant be ignored that more
data will give better performance. The dataset is increasing
For the last decade, the direction of research on brain every year. More publically available databases with experts
tumor diagnosis from MRI has been turned into a hybrid labeling could be created like BRATS to bridge this issue.
intelligent system derived from the combination of different Furthermore, the data augmentation method can be used
algorithms and networks as shown in Fig. 10 [19, 20, 38, to enhance the training dataset. Second, most DL-based
41, 45, 49, 50, 52, 57, 70, 160]. This is the easiest way frameworks are impotent due to their restricted capacity to
to employ the strengths and weaknesses, leading to more delineate visual objects to consider the spatial consistency
robust and exceptional CAD system performance. Despite and appearance of segmentation results [26, 27]. Therefore,
extensive research, strength, and huge popularity in terms of a robust hybrid/fusion (a new learning-based method)
accuracy, conventional machine learning, and deep learning segmentation method is suggested. Thanks to the hybridized
methods, especially CNN’s, encounter various challenges. approach, a more accurate and efficient methodology based
For example, they need a large amount of training data on integrating machine-learned features and hand-crafted
which could either be difficult to acquire for each domain. features can resolve this issue for the efficient automated
Moreover, it can be tough to have the desired accuracy segmentation of brain tumors. Although some research
for a target problem [185]. Furthermore, the increment in has already been conducted to resolve the issue [188, 189].
the number of layers in the deep learning model cannot However, research in this realm is still ongoing for further
guarantee the increment in classification accuracy. Similarly, advancement.
owing to running GPU and RAM (hardware devices), The current research community is deprived of perfect
the DL models are computationally expensive. Lack design for brain tumor MRI segmentation and classification
of computational power results in more time to train the in terms of superior accuracy, low computational time,
network, which depends on the size of the training dataset. minimum cost, acquiring a massive amount of data for
Thus, employing DL models in real-time scenarios, for training, and requiring a group of experts for evaluation so
instance, the clinical practice remains a mystery [186]. on. Certain areas like excellent accuracy still need extensive
The employment of deep learning in neuroimaging research. In a report, accuracies are compared for different
analysis endure black-box problems in artificial ML and DL methods, including SVM, KNN, LDA, and
intelligence (AI). The researchers are well known LR. Whereby, the algorithms are tested on 163 samples of
for inputs and outputs but not known for internal BRATS 2017. The study reveals that the best recognition
representations. Therefore, DL methods are highly performance is attained using a hybrid system fusing LDA
affected by the inherent problems of medical images, with the CNN classifier [190]. All the other snags of ML
i.e., noise and illumination. However, the solution to and DL methods can be overcome by the fusion of two or
this problem is introducing pre-processing steps before more techniques that effectively vanquish the ambiguity
sending input to the model to improve the performance. in the field of brain tumor segmentation and classification
Moreover, acquiring a massive amount of data with utilizing brain MRI scans.
13
4890 S. Ali et al.
9 Conclusion 4. Ganau L, Paris M, Ligarotti GK, Ganau M (2015) Management

of gliomas: overview of the latest technological advancements
and related behavioral drawbacks. Behav Neurol 1–8:2015
This study systematically covered various brain tumor 5. Jayadevappa D, SrinivasKumar S, Murty DS (2011) Medical
segmentation, feature extraction, and classification image segmentation algorithms using deformable models: a
techniques over 2010–2020. Two-decade analysis reveals review. IETE Tech Rev 28(3):248–255
6. Yazdani S, Yusof R, Karimian A, Pashna M, Hematian A (2015)
certain facts related to the development in the usage of Image segmentation methods and applications in mri brain
artificial intelligence-based approaches for performing images. IETE Tech Rev 32(6):413–427
tumor segmentation and classification tasks. It will assist 7. Shijin Kumar PS, Dharun VS (2016) A study of mri segmentation
radiologists and clinicians in early treatment planning methods in automatic brain tumor detection. Int J Eng Technol
8(2):609–614
and diagnosis of brain tumors. Finally, considering the 8. Preston DC (2006) Magnetic resonance imaging (mri) of the
statistical analysis of two decades, it has been observed that brain and spine: Basics. https://case.edu/med/neurology/NR/
researchers should practice more deep learning, ensemble, MRI%20Basics.htm
or hybrid-based techniques to design robust CAD systems. 9. Goodfellow I, Bengio Y, Courville A, Bengio Y (2016) Deep
learning, vol 1. MIT Press, Cambridge
In the future, it can be explored that the combination 10. Zacharaki EI, Wang S, Chawla S, Yoo DS, Wolf R, Melhem ER,
of few-shot learning techniques with CNN could be more Davatzikos C (2009) Classification of brain tumor type and grade
effective towards fulfilling the requirements of segmentation using mri texture and shape in a machine learning scheme. Magn
and classification of brain tumors. Since few-shot learning Reson Med 62(6):1609–1618
11. Hu LS, Ning S, Eschbacher JM, Baxter LC, Gaw N, Ranjbar
is an advanced technique where less number of images is S, Plasencia J, Dueck AC, Peng S, Smith KA et al (2017)
required for the training the network. Since acquiring a Radiogenomics to characterize regional genetic heterogeneity
huge amount of data with expert annotations from multiple in glioblastoma. Neuro Oncol 19(1):128–137
institutions is difficult in each domain. Moreover, research 12. Zeiler MD, Fergus R (2014) Visualizing and understanding
convolutional networks. In Proceedings of the European
can be conducted using other medical modalities like conference on computer vision, Zürich, Switzerland, pp 818–
computed tomography (CT) for brain tumor detection and 833. Springer
classification. 13. Yin C, Shi L, Sun R, Wang J (2020) Improved collaborative
filtering recommendation algorithm based on differential privacy
protection. J Supercomput 76(7):5161–5174
14. Amin J, Sharif M, Yasmin M, Fernandes SL (2020) A distinctive
Author Contributions Each author took part in the present work
approach in brain tumor detection and classification using mri.
conception and/or design. Tasks of data collection, material
Pattern Recogn Lett 139:118–127
preparation, data analysis, and writing of the original draft were
15. Mehmood I, Ejaz N, Sajjad M, Baik SW (2013) Prioritization of
executed by Saqib Ali. Jianqiang Li and Yan Pei supervise this study.
brain mri volumes using medical image perception model and
Rooha Khurram and Khalil ur Rehman helped in reviewing, and editing
tumor region segmentation. Comput Biol Med 43(10):1471–1483
the manuscript. All authors read and approved the final manuscript.
16. Demirhan A, Güler I (2011) Combining stationary wavelet
transform and self-organizing maps for brain mr image
Funding This study is supported by the National Key R&D Program segmentation. Eng Appl Artif Intell 24(2):358–367
of China with Project No. 2020YFB2104402. 17. Ji Z, Sun Q, Xia Y, Chen Q, Xia D, Feng D (2012) Generalized
rough fuzzy c-means algorithm for brain mr image segmentation.
Declarations Comput Methods Programs Biomed 108(2):644–655
18. Sulaiman SN, Non NA, Isa IS, Hamzah N (2014) Segmentation
of brain mri image based on clustering algorithm. In: Proceedings
Conflict of interest The authors declare that there is no conflict of
of the 2014 IEEE Symposium on Industrial Electronics &
interest regarding the publication of this paper.
Applications (ISIEA), Kota Kinabalu, Malaysia, pp 60–65. IEEE
19. Havaei M, Davy A, Warde-Farley D, Biard A, Courville A,
Bengio Y, Pal C, Jodoin P-M, Larochelle H (2017) Brain tumor
segmentation with deep neural networks. Med Image Anal
35:18–31
References 20. Pereira S, Pinto A, Alves V, Silva CA (2016) Brain tumor
segmentation using convolutional neural networks in mri images.
1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal IEEE Trans Med Imaging 35(5):1240–1251
A (2020) Erratum: Global cancer statistics 2018: Globocan 21. Kamnitsas K, Ledig C, Newcombe VFJ, Simpson JP, Kane
estimates of incidence and mortality worldwide for 36 cancers AD, Menon DK, Rueckert D, Glocker B (2017) Efficient multi-
in 185 countries. CA Cancer J Clin 70(4):313 scale 3d cnn with fully connected crf for accurate brain lesion
2. Louis DN, Perry A, Reifenberger G, Von Deimling A, Figarella- segmentation. Med Image Anal 36:61–78
Branger D, Cavenee WK, Ohgaki H, Wiestler OD, Kleihues 22. De Smedt F, Hulens D, Goedemé T. On-board real-time tracking
P, Ellison DW (2016) The 2016 world health organization of pedestrians on a uav. In: Proceedings of the IEEE conference
classification of tumors of the central nervous system: a on computer vision and pattern recognition workshops, Boston,
summary. Acta Neuropathol 131(6):803–820 pp 1–8. IEEE, 2015
3. Gordillo N, Montseny E, Sobrevilla P (2013) State of the art 23. Chang PD (2016) Fully convolutional deep residual neural
survey on mri brain tumor segmentation. Magn Reson Imaging networks for brain tumor segmentation. In: Proceedings of
31(8):1426–1438 the International workshop on Brainlesion: Glioma, multiple
13
sclerosis, stroke and traumatic brain injuries, Athens, Greece, 40. Shanthi KJ, Sasikumar MN, Kesavadas C (2010) Neuro-fuzzy
pp 108–118. Springer approach toward segmentation of brain mri based on intensity
24. Randhawa RS, Modi A, Jain P, Warier P (2016) Improving and spatial distribution. J Med imaging Radiat Sci. 41(2):66–71
boundary classification for brain tumor segmentation and 41. Somasundaram K, Kalaiselvi T (2010) Fully automatic brain
longitudinal disease progression. In: Proceedings of the extraction algorithm for axial t2-weighted magnetic resonance
International Workshop on Brainlesion: Glioma, Multiple images. Comput Biol Med 40(10):811–822
Sclerosis, Stroke and Traumatic Brain Injuries, Athens, Greece, 42. Donoso R, Veloz A, Allende H (2010) Modified expectation
pp 65–74. Springer maximization algorithm for mri segmentation. In Proceedings of
25. Rao V, Sarabi MS, Jaiswal A (2015) Brain tumor segmentation the Iberoamerican Congress on Pattern Recognition, Sao Paulo,
with deep learning. MICCAI Multimodal Brain Tumor Brazil, pp 63–70. Springer
Segmentation Challenge (BraTS) 59:1–4 43. FazelZarandi MH, Zarinbal M, Izadi M (2011) Systematic image
26. Zheng S, Jayasumana S, Romera-Paredes B, Vineet V, Su Z, processing for diagnosing brain tumors: a type-ii fuzzy expert
Du D, Huang C, Torr PHS (2015) Conditional random fields system approach. Appl Soft Comput 11(1):285–294
as recurrent neural networks. In Proceedings of the IEEE 44. Nan Zhang S, Ruan SL, Liao Q, Zhu Y (2011) Kernel feature
international conference on computer vision, Santiago, Chile, selection to fuse multi-spectral mri images for brain tumor
pp 1529–1537. IEEE segmentation. Comput Vis Image Underst 115(2):256–269
27. Zhao F, Xie X (2013) An overview of interactive medical image 45. Ramasamy R, Anandhakumar P (2011) Brain tissue classification
segmentation. Ann BMVA 2013(7):1–22 of mr images using fast Fourier transform based expectation-
28. Singh A et al. (2015) Detection of brain tumor in mri images, maximization gaussian mixture model. In: Proceedings of
using combination of fuzzy c-means and svm. In Proceedings of the International Conference on Advances in Computing and
the 2015 2nd International Conference on Signal Processing and Information Technology, Berlin, Heidelberg, pp 387–398.
Integrated Networks (SPIN), pp 98–102. IEEE, 2015 Springer
29. Bahadure NB, Ray AK, Thethi HP (2017) Image analysis for 46. Tanoori B, Azimifar Z, Shakibafar A, Katebi S (2011) Brain
mri based brain tumor detection and feature extraction using volumetry: an active contour model-based segmentation followed
biologically inspired bwt and svm. Int J Biomed Imaging by svm-based classification. Comput Biol Med 41(8):619–632
1–13:2017 47. Noreen N, Hayat K, Madani SA (2011) Mri segmentation through
30. Amiri S, Rekik I, Mahjoub MA (2016) Deep random forest- wavelets and fuzzy c-means. World Appl Sci J 13:34–39
based learning transfer to svm for brain tumor segmentation. 48. Mohsen H, Ahmed El-Dahshan E-S, Salem A-BM (2012). A
In Proceedings of the 2016 2nd International Conference on machine learning technique for mri brain images. In Proceedings
Advanced Technologies for Signal and Image Processing of the 2012 8th International Conference on Informatics and
(ATSIP), Monastir, Tunisia, pp 297–302. IEEE Systems (INFOS), Cairo University, pp 1–161. IEEE
31. Zhao X, Wu Y, Song G, Li Z, Fan Y, Zhang Y (2016) Brain tumor 49. Gasmi K, Kharrat A, Messaoud MB, Abid M (2012) Automated
segmentation using a fully convolutional neural network with segmentation of brain tumor using optimal texture features
conditional random fields. In Proceedings of the International and support vector machine classifier. In Proceedings of the
Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke International Conference Image Analysis and Recognition,
and Traumatic Brain Injuries, Athens, Greece, pp 75–87. Aveiro, Portugal, pp 230–239. Springer
Springer, 2016 50. Ortiz A, Górriz JM, Ramírez J, Salas-Gonzalez D, Llamas-Elvira
32. Ito R, Nakae K, Hata J, Okano H, Ishii S (2019) Semi-supervised JM (2013) Two fully-unsupervised methods for mr brain image
deep learning of brain tissue segmentation. Neural Netw segmentation using som-based strategies. Appl Soft Comput
116:25–34 13(5):2668–2682
33. Ren T, Wang H, Feng H, Chensheng X, Liu G, Ding P (2019) 51. Havaei M, Jodoin P-M, Larochelle H (2014) Efficient interactive
Study on the improved fuzzy clustering algorithm and its brain tumor segmentation as within-brain knn classification.
application in brain image segmentation. Appl Soft Comput In: Proceedings of the 2014 22nd International Conference on
81:1–9 Pattern Recognition, pp 556–561. IEEE
34. Sundararajan RSS, Venkatesh S, Jeya Pandian M (2019) 52. Charutha S, Jayashree MJ (2014) An efficient brain tumor
Convolutional neural network based medical image classifier. detection by integrating modified texture based region growing
Int J Recent Technol Eng 8(3):4494–4499 and cellular automata edge detection. In: Proceedings of the
35. Deng W, Shi Q, Luo K, Yang Y, Ning N (2019) Brain tumor 2014 International Conference on Control, Instrumentation,
segmentation based on improved convolutional neural network Communication and Computational Technologies (ICCICCT),
in combination with non-quantifiable local texture feature. J Med pp 1193–1199. IEEE
Syst 43(6):1–9 53. Duraisamy M, Mary F, Jane M (2014) Cellular neural network
36. Hu Y, Xia Y (2017) 3d deep neural network-based brain based medical image segmentation using artificial bee colony
tumor segmentation using multimodality magnetic resonance algorithm. In: Proceedings of the 2014 International conference
sequences. In: Proceedings of the International MICCAI on green computing communication and electrical engineering
Brainlesion Workshop, Granada, Spain, pp 423–434. Springer (ICGCCEE, Coimbatore, India), pp 1–6. IEEE
37. AlBadawy EA, Saha A, Mazurowski MA (2018) Deep learning 54. Huang M, Yang W, Yao W, Jiang J, Chen W, Feng Q (2014)
for segmentation of brain tumors: impact of cross-institutional Brain tumor segmentation based on local independent projection-
training and testing. Med Phys 45(3):1150–1158 based classification. IEEE Trans Biomed Eng 61(10):2633–2645
38. Saouli R, Akil M, Kachouri R et al (2018) Fully automatic brain 55. Agn M, Puonti O, Munck af Rosenschöld P, Law I, Van Leemput
tumor segmentation using end-to-end incremental deep neural k (2015) Brain tumor segmentation using a generative model
networks in mri images. Comput Methods Programs Biomed with an rbm prior on tumor shape. In: Proceedings of the
166:39–49 BrainLes 2015, Munich, Germany, pp 168–180. Springer
39. Thillaikkarasi R, Saravanan S (2019) An enhancement of deep 56. Zhao L, Jia K (2015) Deep feature learning with discrimination
learning algorithm for brain tumor segmentation using kernel mechanism for brain tumor segmentation and diagnosis. In:
based cnn with m-svm. J Med Syst 43(4):1–7 Proceedings of the 2015 international conference on intelligent
13
4892 S. Ali et al.
information hiding and multimedia signal processing (IIH-MSP), 72. Lin F, Qiang W, Liu J, Wang D, Kong X (2020) Path
Adelaide, SA, Australia, pp 306–309. IEEE aggregation u-net model for brain tumor segmentation.
57. Tustison NJ, Shrinidhi KL, Wintermark M, Durst CR, Kandel Multimedia Tools Appl 80(15):1–14
BM, Gee JC, Grossman MC, Avants BB (2015) Optimal 73. Kao P-Y, Shailja F, Jiang J, Zhang A, Khan A, Chen JW,
symmetric multimodal templates and concatenated random Manjunath BS (2020) Improving patch-based convolutional
forests for supervised brain tumor segmentation (simplified) with neural networks for mri brain tumor segmentation by
antsr. Neuroinformatics 13(2):209–225 leveraging location information. Front Neurosci 13:1–14
58. Xiao Z, Huang R, Ding Y, Lan T, Dong RF, Qin Z, Zhang X, 74. Zhou C, Ding C, Wang X, Zhentai L, Tao D (2020) One-pass
Wang W (2016) A deep learning-based segmentation method for multi-task networks with cross-task guided attention for brain
brain tumor in mr images. In: Proceedings of the 2016 IEEE 6th tumor segmentation. IEEE Trans Image Process 29:4516–4529
International Conference on Computational Advances in Bio and 75. Khalil HA, Darwish S, Ibrahim YM, Hassan OF (2020) 3d-mri
Medical Sciences (ICCABS), Atlanta, GA, USA, pp 1–6. IEEE, brain tumor detection model using modified version of level
2016 set segmentation based on dragonfly algorithm. Symmetry
59. Kim G (2017) Brain tumor segmentation using deep fully 12(8):1–22
convolutional neural networks. In: Proceedings of the 76. Liu P, Dou Q, Wang Q, Heng P-A (2020) An encoder-
International MICCAI Brainlesion Workshop, Granada, Spain, decoder neural network with 3d squeeze-and-excitation and
pp 344–357. Springer deep supervision for brain tumor segmentation. IEEE Access
60. Pourreza R, Zhuge Y, Ning H, Miller R (2017) Brain tumor 8:34029–34037
segmentation in mri scans using deeply-supervised neural 77. Naser MA, Deen MJ (2020) Brain tumor segmentation and
networks. In: Proceedings of the International MICCAI grading of lower-grade glioma using deep learning in mri
Brainlesion Workshop, Granada, Spain, pp 320–331. Springer images. Comput Biol Med 121:1–8
61. Dong H, Yang G, Liu F, Mo Y, Guo Y (2017) Automatic brain 78. Khalid S, Khalil T, Nasreen S (2014) A survey of feature
tumor detection and segmentation using u-net based fully selection and feature extraction techniques in machine
convolutional networks. In: Proceedings of the annual conference learning. In: Proceedings of the 2014 science and information
on medical image understanding and analysis, Edinburgh, United conference, Park Inn by Radisson Hotel, London Heathrow, pp
Kingdom, pp 506–517, Springer 372–378. IEEE
62. Kamrul Hasan SM, Linte CA (2018) A modified u-net 79. Ranjan Nayak D, Dash R, Majhi B (2016) Brain mr image
convolutional network featuring a nearest-neighbor re-sampling- classification using two-dimensional discrete wavelet
based elastic-transformation for brain tissue characterization and transform and adaboost with random forests. Neurocomputing
segmentation. In: Proceedings of the 2018 IEEE Western New 177:188–197
York Image and Signal Processing Workshop (WNYISPW), 80. Mohan G, Monica Subashini M, Survey on brain tumor grade
Rochester, NY, USA, pp 1–5. IEEE classification (2018) Mri based medical image analysis.
63. Perkuhn M, Stavrinou P, Thiele F, Shakirin G, Mohan M, Biomed Signal Process Control 39:139–161
Garmpis D, Kabbasch C, Borggrefe J (2018) Clinical evaluation 81. Deepa AR, Emmanuel Sam WR (2019) An efficient
of a multiparametric deep learning model for glioblastoma detection of brain tumor using fused feature adaptive firefly
segmentation using heterogeneous magnetic resonance imaging backpropagation neural network. Multimedia Tools Appl
data from clinical routine. Invest Radiol 53(11):1–8 78(9):11799–11814
64. Zhao X, Yihong W, Song G, Li Z, Zhang Y, Fan Y (2018) A 82. Diagnosis C-A, Ahmed KB, Hall LO, Goldgof DB, Liu R,
deep learning model integrating fcnns and crfs for brain tumor Gatenby RA (2017) Fine-tuning convolutional deep features
segmentation. Med Image Anal 43:98–111 for mri based brain tumor classification. In: Proceedings of the
65. Chang J, Zhang L, Naijie G, Zhang X, Ye M, Yin R, Meng Q Medical Imaging 2017. Orlando, Florida, USA, vol 10134, pp
(2019) A mix-pooling cnn architecture with fcrf for brain tumor 1–7
segmentation. J Vis Commun Image Represent 58:316–322 83. Yassin NIR, Omran S, El Houby EMF, Allam H (2018) Machine
66. Chang K, Beers AL, Bai HX, Brown JM, Ina Ly K, Li X, Senders learning techniques for breast cancer computer aided diagnosis
JT, Kavouridis VK, Boaro A, Su C et al (2019) Automatic using different image modalities: a systematic review. Comput
assessment of glioma burden: a deep learning algorithm for fully Methods Programs Biomed 156:25–45
automated volumetric and bidimensional measurement. Neuro 84. El-Dahshan E-SA, Hosny T, Salem A-BM (2010) Hybrid
Oncol 21(11):1412–1422 intelligent techniques for mri brain images classification. Digital
67. Kumar S, Negi A, Singh JN (2019) Semantic segmentation Signal Process 20(2):433–441
using deep learning for brain tumor mri via fully convolution 85. Qurat-Ul-Ain GL, Kazmi SB, Jaffar MA, Mirza AM (2010)
neural networks. In: Proceedings of the Information and Classification and segmentation of brain tumor using texture
Communication Technology for Intelligent Systems, pp 11–19. analysis. In: Proceedings of the Recent advances in artificial
Springer intelligence, knowledge engineering and data bases, pp 147–155
68. Sharma A, Kumar S, Narayan Singh S (2019) Brain tumor 86. Kharrat A, Gasmi K, Messaoud MB, Benamrane N, Abid M
segmentation using de embedded otsu method and neural (2010) A hybrid approach for automatic classification of brain
network. Multidimension Syst Signal Process 30(3):1263–1291 mri using genetic algorithm and support vector machine.
69. Feng X, Tustison NJ, Patel SH, Meyer CH (2020) Brain tumor Leonardo J Sci 17(1):71–82
segmentation using an ensemble of 3d u-nets and overall survival 87. Zhang Y-D, Wang S, Lenan W (2010) A novel method for
prediction using radiomic features. Front Comput Neurosci magnetic resonance brain image classification based on adaptive
14:1–12 chaotic pso. Progress Electromagn Res 109:325–343
70. Nema S, Dudhane A, Murala S, Naidu S (2020) Rescuenet: 88. Yamamoto D, Arimura H, Kakeda S, Magome T, Yamashita Y,
an unpaired gan for brain tumor segmentation. Biomed Signal Toyofuku F, Ohki M, Higashida Y, Korogi Y (2010) Computer-
Process Control 55:1–9 aided detection of multiple sclerosis lesions in brain magnetic
71. Li Q, Yu Z, Wang Y, Zheng H (2018) Tumorgan: a multi-modal resonance images: false positive reduction scheme consisted
data augmentation framework for brain tumor segmentation. of rule-based, level set method, and support vector machine.
Sensors 20(15):1–16 Comput Med Imaging Graph 34(5):404–413
13
89. Jafari M, Kasaei S (2011) Automatic brain tissue detection in mri Proceedings of the 2015 International Conference on Industrial
images using seeded region growing segmentation and neural Instrumentation and Control (ICIC), Pune, India, pp 60–65. IEEE
network classification. Aust J Basic Appl Sci 5(8):1066–1079 105. Kharrat A, Ben Halima M, Ben Ayed M (2015) Mri brain
90. Zöllner FG, Emblem KE, Schad LR (2012) Svm-based glioma tumor classification using support vector machines and meta-
grading: optimization by feature reduction analysis. Z Med Phys heuristic method. In: Proceedings of the 2015 15th International
22(3):205–214 Conference on Intelligent Systems Design and Applications
91. Jafarpour S, Sedghi Z, Amirani MC (2012) A robust brain mri (ISDA), Marrakech, Morocco, pp 446–451. IEEE
classification with glcm features. Int J Comput Appl 37(12):1–5 106. Rajesh Chandra G, Ramchand K, Rao H (2016) Tumor
92. Arimura H, Tokunaga C, Yamashita Y, Kuwazuru J (2012) detection in brain using genetic algorithm. Procedia Comput Sci
Magnetic resonance image analysis for brain cad systems with 79:449–457
machine learning. In: Proceedings of the Machine learning in 107. Ishikawa Y, WashiyaK, Aoki K, Nagahashi H (2016) Brain tumor
computer-aided diagnosis: medical imaging intelligence and classification of microscopy images using deep residual learning.
analysis, pp 258–296. IGI Global In: Proceedings of the SPIE BioPhotonics Australasia, Adelaide,
93. Saritha M, Paul Joseph K, Mathew AT (2013) Classification of Australia, vol 10013, pp 1–10. SPIE
mri brain images using combined wavelet entropy based spider 108. Mohsen H, El-Dahshan E-SA, El-Horbaty E-SM, Salem A-BM
web plots and probabilistic neural network. Pattern Recogn Lett (2018) Classification using deep learning neural networks for
34(16):2151–2156 brain tumors. Future Comput Inf J 3(1):68–71
94. Kalbkhani H, Shayesteh MG, Zali-Vargahan B (2013) Robust 109. Paul JS, Plassard AJ, Landman BA, Fabbri D (2017) Deep
algorithm for brain magnetic resonance image (mri) classification learning for brain tumor classification. In: Proceedings of the
based on garch variances series. Biomed Signal Process Control Medical Imaging 2017: Biomedical Applications in Molecular,
8(6):909–919 Structural, and Functional Imaging, Orlando, Florida, USA, vol
95. Sindhumol S, Kumar A, Balakrishnan K (2013) Spectral 10137, pp 1–16. SPIE
clustering independent component analysis for tissue 110. Alberts E, Tetteh G, Trebeschi S, Bieth M, Valentinitsch
classification from brain mri. Biomed Signal Process Control A, Wiestler B, Zimmer C, Menze BH (2017). Multi-modal
8(6):667–674 image classification using low-dimensional texture features for
96. Ibrahim WH, AbdelRhman A, Osman A, Ibrahim Mohamed Y genomic brain tumor recognition. In: Proceedings of the Graphs
(2013) Mri brain image classification using neural networks. In: in Biomedical Image Analysis, Computational Anatomy and
Proceedings of the 2013 international conference on computing, Imaging Genetics, Québec City, QC, Canada, pp 201–209.
electrical and electronic engineering (ICCEEE), Khartoum, Springer
Sudan, pp 253–258. IEEE, 2013 111. Banerjee S, Mitra S, Masulli F, Rovetta S (2018) Brain tumor
97. Amsaveni V, Albert Singh N, Dheeba J (2013) Computer aided detection and classification from multi-sequence mri: Study
detection of tumor in mri brain images using cascaded correlation using convnets. In: Proceedings of the International MICCAI
neural network. In: Proceedings of the IET Chennai Fourth brainlesion workshop, Granada, Spain, pp. 170–179. Springer
International Conference on Sustainable Energy and Intelligent 112. Zhou Y, LiZ, Zhu H, Chen C, Gao M, Xu K, Xu J (2018) Holistic
Systems (SEISCON 2013), Chennai, India, pp 527–532. IET brain tumor screening and classification based on densenet and
98. Bhanumurthy MY, Anne K (2014) An automated detection and recurrent neural network. In: Proceedings of the International
segmentation of tumor in brain mri using artificial intelligence. MICCAI Brainlesion Workshop, Granada, Spain, pp 208–217,
In: Proceedings of the 2014 IEEE International Conference Springer
on Computational Intelligence and Computing Research, 113. Ari A, Hanbay D (2018) Deep learning based brain tumor
Coimbatore, India, pp 1–6. IEEE classification and detection system. Turk J Electr Eng Comput
99. Babu Nandpuru H, Salankar SS, Bora VR. Mri brain cancer Sci 26(5):2275–2286
classification using support vector machine. In: Proceedings of 114. Afshar P, Mohammadi A, Plataniotis KN (2018) Brain tumor
the 2014 IEEE Students’ Conference on Electrical, Electronics type classification via capsule networks. In: Proceedings of the
and Computer Science, Bhopal, India, pp 1–6. IEEE 2018 25th IEEE International Conference on Image Processing
100. Preethi G, and Sornagopal V (2014) Mri image classification (ICIP), Athens, Greece, pp 3129–3133. IEEE
using glcm texture features. In: proceedings of the 2014 115. Hemanth G, Janardhan M, Sujihelen L (2019) Design and
international conference on green computing communication implementing brain tumor detection using machine learning
and electrical engineering (ICGCCEE), Coimbatore, India, pp approach. In: Proceedings of the 2019 3rd International
1–6. IEEE Conference on Trends in Electronics and Informatics (ICOEI),
101. Nasir M, Khanum A, Baig A (2014) Classification of brain Tirunelveli, India, pp 1289–1294. IEEE
tumor types in mri scans using normalized cross-correlation 116. Tahir B, Iqbal S, Usman Ghani Khan M, Saba T, Mehmood Z,
in polynomial domain. In: Proceedings of the 2014 12th Anjum A, Mahmood T (2019) Feature enhancement framework
International Conference on Frontiers of Information for brain tumor segmentation and classification. Microsc Res
Technology, Islamabad Pakistan, pp 280–285. IEEE Techn 82(6):803–811
102. Xu Y, Jia Z, Ai Y, Zhang F, Lai M, Eric I, Chang C (2015) 117. Amin J, Sharif M, Yasmin M, Saba T, Anjum MA, Fernandes
Deep convolutional activation features for large scale brain SL (2019) A new approach for brain tumor segmentation and
tumor histopathology image classification and segmentation. classification based on score level fusion using transfer learning.
In: Proceedings of the 2015 IEEE international conference J Med Syst 43(11):1–16
on acoustics, speech and signal processing (ICASSP), South 118. Zhao J, Meng Z, Wei L, Sun C, Zou Q, Ran S (2019) Supervised
Brisbane, QLD, Australia, pp 947–951. IEEE brain tumor segmentation based on gradient and context-
103. Yang G, Zhang Y, Yang J, Ji G, Dong Z, Wang S, Feng C, Wang sensitive features. Front Neurosci 13:1–11
Q (2016) Automated classification of brain images using wavelet- 119. Peng S, Chen W, Sun J, Liu B (2020) Multi-scale 3d u-nets: An
energy and biogeography-based optimization. Multimedia Tools approach to automatic segmentation of brain tumor. Int J Imaging
Appl 75(23):15601–15617 Syst Technol 30(1):5–17
104. Machhale K, Babu Nandpuru H, Kapur V, Kosta L (2015) Mri 120. Grøvik E, Yi D, Iv M, Tong E, Rubin D, Zaharchuk G (2020)
brain cancer classification using hybrid classifier (svm-knn). In: Deep learning enables automatic detection and segmentation of
13
4894 S. Ali et al.
brain metastases on multisequence mri. J Magn Reson Imaging 139. Cheng J-Z, Ni D, Chou Y-H, Qin J, Tiu C-M, Chang Y-C,
51(1):175–182 Huang C-S, Shen D, Chen C-M (2016) Computer-aided
121. Arunkumar N, Abed Mohammed M, Mostafa SA, Ibrahim DA, diagnosis with deep learning architecture: applications to
Rodrigues JJPC, de Albuquerque VHC (2020) Fully automatic breast lesions in us images and pulmonary nodules in ct scans.
model-based segmentation and classification approach for mri Sci Rep 6(1):1–13
brain tumor using artificial neural networks. Concurr Comput 140. Kooi T, Litjens G, Van Ginneken B, Gubern-Mérida A,
Pract Exp 32(1):1–9 Sánchez CI, Mann R, den Heeten A, Karssemeijer N (2017)
122. Narmatha C, Eljack SM, Tuka AARM, Manimurugan S, Mustafa Large scale deep learning for computer aided detection of
M (2020) A hybrid fuzzy brain-storm optimization algorithm for mammographic lesions. Med Image Anal 35:303–312
the classification of brain tumor mri images. J Ambient Intell 141. Lakhani P, Sundaram B (2017) Deep learning at chest
Human Comput 4:1–9 radiography: automated classification of pulmonary
123. Alfonse M, Salem A-BM (2016) An automatic classification of tuberculosis by using convolutional neural networks.
brain tumors through mri using support vector machine. Egy Radiology 284(2):574–582
Comp Sci J 40(3):1–11 142. Hussain S, Anwar SM, Majid M (2018) Segmentation of
124. Nadeem MW, Ghamdi MAA, Hussain M, Khan MA, Khan KM, glioma tumors in brain using deep convolutional neural
Almotiri SH, Butt SA (2020) Brain tumor analysis empowered network. Neurocomputing 282:248–261
with deep learning: a review, taxonomy, and future challenges. 143. Myronenko A (2018) 3d mri brain tumor segmentation using
Brain Sci 10(2):1–33 autoencoder regularization. In: Proceedings of the International
125. Sujan M, Alam N, Noman SA, Islam MJ (2016) A segmentation MICCAI Brainlesion Workshop, Granada, Spain, pp 311–320.
based automated system for brain tumor detection. Int J Comput Springer
Appl 153(10):41–49 144. Ali H, Elmogy M, El-Daydamony E, Atwan A (2015)
126. Deepak S, Ameer PM (2019) Brain tumor classification using Multi-resolution mri brain image segmentation based on
deep cnn features via transfer learning. Comput Biol Med morphological pyramid and fuzzy c-mean clustering. Arab J
111:1–7 Sci Eng 40(11):3173–3185
127. Chaplot S, Patnaik LM, Jagannathan NR (2006) Classification 145. Akkus Z, Sedlar J, Coufalova L, Korfiatis P, Kline TL,
of magnetic resonance brain images using wavelets as input to Warner JD, Agrawal J, Erickson BJ (2015) Semi-automated
support vector machine and neural network. Biomed Signal segmentation of pre-operative low grade gliomas in magnetic
Process Control 1(1):86–92 resonance imaging. Cancer Imaging 15(1):1–10
128. Liu J, Li M, Wang J, Fangxiang W, Liu T, Pan Y (2014) A survey 146. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee
of mri-based brain tumor segmentation methods. Tsinghua Sci JC (2011) A reproducible evaluation of ants similarity
Technol 19(6):578–595 metric performance in brain image registration. Neuroimage
129. Burgos N, Bottani S, Faouzi J, Thibeau-Sutre E, Colliot O 54(3):2033–2044
(2021) Deep learning for brain disorders: from data processing 147. Hussain S, Muhammad Anwar S, Majid M (2017). Brain
to disease treatment. Brief Bioinform 22(2):1560–1576 tumor segmentation using cascaded deep convolutional neural
130. Arunachalam M, Royappan Savarimuthu S (2017) An efficient network. In: Proceedings of the 2017 39th Annual International
and automatic glioblastoma brain tumor detection using shift- Conference of the IEEE Engineering in Medicine and Biology
invariant shearlet transform and neural networks. Int J Imaging Society (EMBC), Jeju, Korea (South), pp 1998–2001. IEEE
Syst Technol 27(3):216–226 148. Singh L, Chetty G, Sharma D (2012) A novel machine learning
131. Mittal M, Goyal LM, Kaur S, Kaur I, Verma A, Hemanth DJ approach for detecting the brain abnormalities from mri
(2019) Deep learning based enhanced tumor segmentation structural images. In: Proceedings of the IAPR International
approach for mr brain images. Appl Soft Comput 78:346–354 Conference on Pattern Recognition in Bioinformatics, Tokyo,
132. Ali S, Li J, Pei Y, Aslam MS, Shaukat Z, Azeem M (2020) An Japan, pp 94–105. Springer
effective and improved cnn-elm classifier for handwritten digits 149. Pan Y, Huang W, Lin Z, Zhu W, Zhou J, Wong J, Ding Z
recognition and classification. Symmetry 12(10):1–15 (2015) Brain tumor grading based on neural networks and
133. El-Dahshan E-SA, Mohsen HM, Revett K, Salem A-BM (2014) convolutional neural networks. In: Proceedings of the 2015
Computer-aided diagnosis of human brain tumor through mri: aA 37th Annual International Conference of the IEEE Engineering
survey and a new algorithm. Expert Syst Appl 41(11):5526–5545 in Medicine and Biology Society (EMBC), Milan, Italy, pp
134. Vidyarthi A, Mittal N (2014) Comparative study for brain 699–702. IEEE
tumor classification on mr/ct images. In: Proceedings of the 150. Soltaninejad M, Yang G, Lambrou T, Allinson N, Jones TL,
Third International Conference on Soft Computing for Problem Barrick TR, Howe FA, Ye X (2018) Supervised learning based
Solving, Noida Campus of Indian Institute of Technology multimodal mri brain tumour segmentation using texture
Roorkee, India, pp 889–897. Springer features from supervoxels. Comput Methods Programs Biomed
135. Saad NM, Bakar SARSA, Muda AS, Mokji MM (2015) Review 157:69–84
of brain lesion detection and classification using neuroimaging 151. Iqbal S, Ghani MU, Saba T, Rehman A (2018) Brain tumor
analysis techniques. Jurnal Teknologi 74(6):1–13 segmentation in multi-spectral mri using convolutional neural
136. Tandel GS, Biswas M, Kakde OG, Tiwari A, Suri HS, Turk networks (cnn). Microsc Res Techn 81(4):419–427
M, Laird JR, Asare CK, Ankrah AA, Khanna NN et al (2019) 152. Thaha MM, Kumar KPM, Murugan BS, Dhanasekeran S,
A review on a deep learning perspective in brain cancer Vijayakarthick P, Selvi AS (2019) Brain tumor segmentation
classification. Cancers 11(1):1–32 using convolutional neural networks in mri images. J Med Syst
137. Muhammad K, Khan S, Ser JD, de Albuquerque VHC (2020) 43(9):1–10
Deep learning for multigrade brain tumor classification in smart 153. Rajendran A, Dhanasekaran R (2012) Fuzzy clustering and
healthcare systems: a prospective survey. IEEE Trans Neural deformable model for tumor segmentation on mri brain image:
Netw Learn Syst 32(2):507–522 a combined approach. Procedia Eng 30:327–333
138. Vasilakos AV, Tang Y, Yao Y et al (2016) Neural networks for 154. Abdel-Maksoud E, Elmogy M, Al-Awadi R (2015) Brain tumor
computer-aided diagnosis in medicine: a review. Neurocomputing segmentation based on a hybrid clustering technique. Egypt Inf
216:700–708 J 16(1):71–81
13
155. Demirhan A, Toru M, Güler I (2014) Segmentation of 172. Zhang X, Yan L-F, Yu-Chuan H, Li G, Yang Yang Yu, Han Y-ZS,
tumor and edema along with healthy tissues of brain using Liu Z-C, Tian Q, Han Z-Y et al (2017) Optimizing a machine
wavelets and neural networks. IEEE J Biomed Health Inf learning based glioma grading system using multi-parametric mri
19(4):1451–1458 histogram and texture features. Oncotarget 8(29):47816–47830
156. Vishnuvar thanan G, Rajasekaran MP, Subbaraj P, 173. McKinley R, Jungo A, Wiest R, Reyes M (2017) Pooling-
Vishnuvarthanan A (2016) An unsupervised learning method free fully convolutional networks with dense skip connections
with a clustering approach for tumor identification and tissue for semantic segmentation, with application to brain tumor
segmentation in magnetic resonance brain images. Appl Soft segmentation. In: Proceedings of the International MICCAI
Comput 38:190–212 Brainlesion Workshop, Quebec City, Canada, pp 169–177.
157. Vishnuvarthanan A, Rajasekaran MP, Govindaraj V, Zhang Y, Springer
Thiyagarajan A (2017) An automated hybrid approach using 174. Mlynarski P, Delingette H, Criminisi A, Ayache N (2019) Deep
clustering and nature inspired optimization technique for learning with mixed supervision for brain tumor segmentation.
improved tumor and tissue segmentation in magnetic resonance J Med Imaging 6(3):1–14
brain images. Appl Soft Comput 57:399–426 175. Pertierra LR, Hughes KA, Vega GC, Olalla-Tárraga M (2017)
158. Al-Dmour H, Al-Ani A (2018) A clustering fusion technique for High resolution spatial mapping of human footprint across
mr brain tissue segmentation. Neurocomputing 275:546–559 Antarctica and its implications for the strategic conservation of
159. Namburu A, Samay SK, Edara SR (2017) Soft fuzzy rough avifauna. PLoS ONE 12(1):1–20
set-based mr brain image segmentation. Appl Soft Comput 176. Zhao Z, Yang G, Lin Y, Pang H, Wang M (2018) Automated
54:456–466 glioma detection and segmentation using graphical models. PLoS
160. Kai H, Gan Q, Zhang Y, Deng S, Xiao F, Huang W, Cao C, ONE 13(8):1–22
Gao X (2019) Brain tumor segmentation using multi-cascaded 177. Menze BH, Jakab A, Bauer S, Kalpathy-Cramer J, Farahani K,
convolutional neural networks and conditional random field. Kirby J, Burren Y, Porz N, Slotboom J, Wiest R et al (2014) The
IEEE Access 7:92615–92629 multimodal brain tumor image segmentation benchmark (brats).
161. Rajam RA, Reshmi R, Suresh A, Suresh A, Sindhuja S (2018) IEEE Trans Med Imaging 34(10):1993–2024
Segmentation and analysis of brain tumor using meta-heuristic 178. Bakas S, Akbari H, Sotiras A, Bilello M, Martin Rozycki
algorithm. In: Proceedings of the 2018 International Conference JS, Kirby JBF, Farahani K, Davatzikos C (2017) Advancing
on Recent Trends in Electrical, Control and Communication the cancer genome atlas glioma mri collections with expert
(RTECC), Malaysia, pp 256–260. IEEE segmentation labels and radiomic features. Sci Data 4(1):1–13
162. Amarapur B et al (2020) Computer-aided diagnosis applied 179. Milletari F, Navab N, Ahmadi S-A (2016) V-net: Fully
to mri images of brain tumor using cognition based modified convolutional neural networks for volumetric medical image
level set and optimized ann classifier. Multimedia Tools Appl segmentation. In: Proceedings of the 2016 fourth international
79(5):3571–3599 conference on 3D vision (3DV), Stanford, CA, USA, pp 565–
163. Khan AR, Khan S, Harouni M, Abbasi R, Iqbal S, Mehmood Z 571. IEEE
(2021) Brain tumor segmentation using k-means clustering and 180. Kamnitsas K, Bai W, Ferrante E, McDonagh S, Sinclair M,
deep learning with synthetic data augmentation for classification. Pawlowski N, Rajchl, Matthew Lee M, Kainz B, Rueckert D, et al
Microsc Res Techn 4:1–11 (2017) Ensembles of multiple models and architectures for robust
164. Ayadi W, Elhamzi W, Charfi I, Atri M (2021) Deep cnn for brain brain tumour segmentation. In: Proceedings of the International
tumor classification. Neural Process Lett 53(1):671–700 MICCAI brainlesion workshop, Quebec City, Canada, pp 450–
165. Arora A, Jayal A, Gupta M, Mittal P, Chandra Satapathy S (2021) 462. Springer
Brain tumor segmentation of mri images using processed image 181. Zhou C, Chen S, Ding C, Tao D (2018) Learning contextual
driven u-net architecture. Computers 10(11):1–17 and attentive information for brain tumor segmentation. In:
166. Javier Díaz-Pernas F, Martínez-Zarzuela M, Antón-Rodríguez Proceedings of the International MICCAI brainlesion workshop,
M, González-Ortega D (2021) A deep learning approach for Granada, Spain, pp 497–507. Springer
brain tumor classification and segmentation using a multiscale 182. Chandra S, Vakalopoulou M, Fidon L, BattistellaE, Estienne T,
convolutional neural network. In: Proceedings of the Healthcare, Sun R, Robert C, Deutsch E, Paragios N (2018) Context aware
vol 9, pp 1–14. Multidisciplinary Digital Publishing Institute 3d cnns for brain tumor segmentation. In: Proceedings of the
167. Majib MS, Rahman MM, Sazzad TMS, Khan NI, Dey SK (2021) International MICCAI Brainlesion Workshop, Granada, Spain,
Vgg-scnet: a vgg net-based deep learning framework for brain pp 299–310. Springer
tumor detection on mri images. IEEE Access 9:116942–116952 183. Wang G, Li G, Ourselin S, Vercauteren T (2017) Automatic brain
168. Macyszyn L, Hamed Akbari JM, Da Pisapia X, Attiah M, Pigrish tumor segmentation using cascaded anisotropic convolutional
V, Bi Y, Sharmistha Pal RV, Davuluri LR et al (2015) Imaging neural networks. In: Proceedings of the International MICCAI
patterns predict patient survival and molecular subtype in brainlesion workshop, Quebec City, Canada, pp 178–190.
glioblastoma via machine learning techniques. Neuro-oncology Springer
18(3):417–425 184. Isensee F, Kickingereder P, Wick W, Bendszus M, Maier-Hein
169. Emblem KE, Pinho MC, Zöllner FG, Paulina Due-Tonnessen KH (2018) No new-net. In: Proceedings of the International
JK, Hald LR, Schad TRM, Rapalino O, Bjornerud A (2015) A MICCAI Brainlesion Workshop, Granada, Spain, pp 234–244.
generic support vector machine model for preoperative glioma Springer
survival associations. Radiology 275(1):228–234 185. Hoo-Chang Shin HR, Roth MG, Lu L, Ziyue X, Nogues I, Yao
170. Sarkiss CA, Germano IM (2019) Machine learning in neuro- J, Mollura D, Summers RM (2016) Deep convolutional neural
oncology: can data analysis from 5346 patients change decision- networks for computer-aided detection: Cnn architectures,
making paradigms? World nNeurosurg 124:287–294 dataset characteristics and transfer learning. IEEE Trans Med
171. Nie D, Zhang H, Adeli E, Liu L, Shen D (2016) 3d deep learning Imaging 35(5):1285–1298
for multi-modal imaging-guided survival time prediction of brain 186. Nima Tajbakhsh JY, Shin SR, Gurudu RT, Hurst CB, Gotway
tumor patients. In: Proceedings of the International conference MBK, Liang J (2016) Convolutional neural networks for medical
on medical image computing and computer-assisted intervention, image analysis: Full training or fine tuning? IEEE Trans Med
Athens, Greece, pp 212–220. Springer Imaging 35(5):1299–1312
13
4896 S. Ali et al.
187. Mondal AK, Dolz J, Desrosiers C (2018) Few-shot 3d multi- prediction in the brats challenge. arXiv preprint arXiv:1811.
modal medical image segmentation using generative adversarial 02629, pp 1–49
learning. arXiv preprint arXiv:1810.12241, pp 1–10 190. Anwar SM, Majid M, Qayyum A, Awais M, Alnowami M, Khan
188. Xiaojun H, Luo W, Jiliang H, Guo S, Weilin Huang MR, Scott MK (2018) Medical image analysis using convolutional neural
RW, Dahlweid M, Reyes M (2020) Brain segnet: 3d local networks: a review. J Med Syst 42(11):1–13
refinement network for brain lesion segmentation. BMC Med
Imaging 20(1):1–10 Publisher's Note Springer Nature remains neutral with regard to
189. Bakas S, Reyes M, Jakab A, Bauer S, Rempfler M, Crimi jurisdictional claims in published maps and institutional affiliations.
A, Shinohara RT, Berger C, Ha SM, Rozycki M et al (2018)
Identifying the best machine learning algorithms for brain tumor
segmentation, progression assessment, and overall survival
13

A Comprehensive Survey On Brain Tumor Diagnosis Using Deep Learning and Emerging Hybrid Techniques With Multi Modal MR Image

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

A Comprehensive Survey On Brain Tumor Diagnosis Using Deep Learning and Emerging Hybrid Techniques With Multi Modal MR Image

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Comprehensive Survey On Brain Tumor Diagnosis Using Deep Learning and Emerging Hybrid Techniques With Multi Modal MR Image

Uploaded by

Copyright:

Available Formats

Archives of Computational Methods in Engineering (2022) 29:4871–4896

A Comprehensive Survey on Brain Tumor Diagnosis Using Deep

1 Introduction behavior, from lowest to extreme aggressive [2]. Low-

In this manuscript, we have used freely available search

some ML-based methods utilize manually segmented 2.3 Hybrid‑Based Segmentation

[40] 2010 FCM based method Brain segmentation – ML Private

BRATS2012 T1,T1-weighted(T1Gd),T2-weighted(T2),and T2 FLAIR 45 patients https://​www.​smir.​ch/​BRATS/​Start​2012

4.3 Commonly Employed Datasets

Acquiring the brain tumor dataset is the primary task.

Fig. 5 Year-wise, rapidly increase in publication for brain tumor

and third, as shown in Fig. 8. This is because radiologists

4.6 Summarization of Previously Reported Works

This review attempts to summarize the various reviews

5 Current Trends on DL‑Based Brain Tumor

In contrast to the machine learning algorithms, the deep

[133] 2003–2012 Segmentation and classification Conventional 9 × × ×

Table 6 Summary of hybrid techniques utilized for brain tumor MRIs

hybrid system. The researchers who acquire robust

Fig. 10 Development toward brain tumor in last decade

9 Conclusion 4. Ganau L, Paris M, Ligarotti GK, Ganau M (2015) Management

You might also like

some ML-based methods utilize manually segmented 2.3 Hybrid‑Based Segmentation

BRATS2012 T1,T1-weighted(T1Gd),T2-weighted(T2),and T2 FLAIR 45 patients https://www.smir.ch/BRATS/Start2012

4.3 Commonly Employed Datasets

4.6 Summarization of Previously Reported Works

5 Current Trends on DL‑Based Brain Tumor

9 Conclusion 4. Ganau L, Paris M, Ligarotti GK, Ganau M (2015) Management