ArticlePDF AvailableLiterature Review

Abstract and Figures

Background: Among the brain-related diseases, brain tumor segmentation on magnetic resonance imaging (MRI) scans is one of the highly focused research domains in the medical community. Brain tumor segmentation is a very challenging task due to its asymmetric form and uncertain boundaries. This process segregates the tumor region into the active tumor, necrosis and edema from normal brain tissues such as white matter (WM), grey matter (GM), and cerebrospinal fluid (CSF). Introduction: The proposed paper analyzed the advancement of brain tumor segmentation from conventional image processing techniques, to deep learning through machine learning on MRI of human head scans. Method: State-of-the-art methods of these three techniques are investigated, and the merits and demerits are discussed. Results: The prime motivation of the paper is to instigate the young researchers towards the development of efficient brain tumor segmentation techniques using conventional and recent technologies. Conclusion: The proposed analysis concluded that the conventional and machine learning methods were mostly applied for brain tumor detection, whereas deep learning methods were good at tumor substructures segmentation.
Content may be subject to copyright.
Advancements of MRI-based brain Tumor Segmentation from Traditional to
Recent Trends- A Review
Padmapriya Thiyagarajan1, Sriramakrishnan Padmanaban2, Kalaiselvi
Thiruvenkadam1*, Somasundaram Karuppanagounder1
1Department of Computer Science and Applications,
The Gandhigram Rural Institute (Deemed to be University),
Gandhigram 624 302, Tamil Nadu, India.
stpadmapriya@gmail.com Kalaiselvi.gri@gmail.com ka.somasundaram@gmail.com
2Department of Computer Applications,
Kalasalingam Academy of Research and Education (Deemed to be University),
Krishnankoil 626128, Tamil Nadu, India.
sriram0210@gmail.com
Abstract
The proposed paper analyzed the advancement of brain tumor segmentation techniques
from conventional, to deep learning through machine learning on brain tumor segmentation
(BraTS) repository. State-of-the-art methods of these three techniques are investigated, and the
merits and demerits are discussed. Among the brain-related diseases, brain tumor segmentation
on magnetic resonance imaging (MRI) scans is one of the highly focused research domains in the
medical community. Brain tumor segmentation is a very challenging task due to its asymmetric
form and uncertain boundaries. This process segregates the tumor region into the active tumor,
necrosis and edema from normal brain tissues such as white matter (WM), grey matter (GM),
and cerebrospinal fluid (CSF). The proposed analysis concluded that the conventional and
machine learning methods were mostly applied for brain tumor segmentation, whereas deep
learning methods were good at tumor substructures segmentation. The prime motivation of the
paper is to instigation the young researchers towards the development of novel brain tumor
segmentation techniques using recent technologies.
Keywords
Brain tumor, Magnetic resonance imaging, Tumor segmentation, Machine learning, Deep
learning, BraTS dataset, Graphics processing units.
1. Introduction
Object segmentation plays a significant role in image processing that provides several
exciting applications like automobile, argument reality, biometrics, medical applications,
military applications, optical character recognition, remote sensing, and robotics. Among them,
medical image segmentation is more valuable because it increases the survival of human beings
[1]. The brain is a complex organ of the human body which controls all other organs. Brain
related disorders are alzheimer, dementias, epilepsy, parkinson’s, tumor, and etc. Reproduction
of a set of abnormal cells within the brain in an uncontrolled manner is known as a brain tumor.
Brain tumors are classified into two broad categories, namely primary brain tumors
(benign), and metastatic brain tumors (malignant). A primary brain tumor forms within the brain
region, and it will not spread to any other parts of the body. A metastatic brain tumor forms in
any parts of the body and later spreads to the brain. About 40% of all cancer’s spreads to the
brain. Benign tumors are non-cancerous, less aggressive, and it is easy to extract from brain
surgically. Malignant tumors are cancerous and more aggressive. Among the most common
tumors, central nervous system (CNS) tumors constitute about 2% approximately. In India, a
CNS tumor ranges from 5 to 10 persons per one lakh population. In India, the second most
occurring cancer in children is a brain tumor. It accounts for about 26% of childhood cancers [2].
The treatment of brain tumor differs from person to person based on the characteristics of
tumor such as location, shape, size and type of tumor. The most suggestible modality for the
treatment of brain tumor is magnetic resonance imaging (MRI) due to its principles of the non-
invasive and non-ionizing image capturing characteristics [3]. MRI provides accurate
information about the location, extent of the tumor and its relationship to the surrounding
structures without exposing the patient to high ionizing radiation. MRI image sequences are T1-
weighted, T1-weighted with contrast enhancement (T1c), T2-weighted, proton density-weighted
(PD), fluid-attenuated inversion recovery (FLAIR) and are shown in Fig.1 [4].
(a) (b)
(c) (d)
Fig. 1. MRI Image Sequences (a) T1 Weighted (b) T1c Weighted (c) T2 Weighted (d) FLAIR
For brain tumor analysis, MRI image sequences play a significant role. T1 is most
commonly used since it allows healthy tissues annotation. In T1c image, the borders of the tumor
region are brighter, and these images are taken after injecting a contrast agent named gadolinium
to the patient. The necrotic core and the active cell regions can be distinguished very easily in the
T1c image sequence. The edema region is brighter in T2 weighted images. FLAIR helps to
separate edema from CSF because the water molecule signals are suppressed during the FLAIR
imaging process [5].
MRI produces very large volumes of tumor images for a single patient and it is very
crucial for the medical professionals to annotate them manually, and segment these images in a
short time. Manual segmentation of tumor is also hard due to low contrast, partial volume effect,
artifacts and patient moments during the scanning process. To overcome these issues, automatic
segmentation methods had gained its importance. With the technological advancement of the
medical industry requires the unsupervised methods for guiding the brain tumor treatment
planning.
Fig. 2. Brain tumor substructures
Before starting the tumor treatment, brain tumor segmentation gives essential qualitative
and quantitative information. In the qualitative analysis, volumetric tumor data can be visualized
for identifying tumor location, shape and size. Quantitative data is useful to identify the
percentage of substructures in the tumor. The substructures of brain tumor such as necrotic core,
active cells and edema are shown in Fig. 2. Active region in brain tumor is called as active cells,
whereas the dead cells in tumor portion are called necrosis. Region which surrounds the tumor
portion is called edema. An intracranial pressure occurs due to the presence of edema [6]. In the
segmentation process, substructures are segregated from the normal brain tissues such as WM,
GM, and CSF.
Many researchers contribute their works in this field, and that can be categorized into
conventional, machine learning and deep learning methods. The conventional methods include
threshold-based, region-based and edge-based segmentation [7-20]. In the thresholding methods,
segmentation can be computed from the value given by the local or global thresholding. This
technique is more suitable for limited regions present in the image. Region-based segmentation
methods worked on adjacent regions starting with a seed point and suggested for multiple
regions. Edge-based methods are appropriate for prediction of uncertainty in the intensity using
edge detectors.
Machine learning methods are support vector machines, native bayes, random forest, k-
means, fuzzy c means, etc. [21-28]. These algorithms are adjusting themselves to give better
results based on their learning rate in the training phase. The testing stage is used to evaluate the
accuracy of the trained model on unknown data. The data can be obtained from the process of
feature extraction, which reduces the number of resources required to describe the large data.
Feature selection helps to maximize the accuracy of machine learning algorithms using feature
selection models like principal component analysis (PCA), independent component analysis, etc.
Deep learning or deep neural networks (DNNs) is a part of machine learning that includes
various forms of convolutional neural networks for segmentation such as patch-based
segmentation, Unet based segmentation, multimodal segmentation, etc. [29-32]. It learns from
vast amounts of unstructured data. It uses multiple layers between the input and output layers to
extract higher-level features from the raw data. The learning process can be supervised, semi-
supervised or unsupervised. Two major issues of DNN is overfitting and computation time. The
overfitting occurs due to added layers of abstraction, which makes rare dependencies in the
training data. Several parameters help to exclude the overfitting such as pruning, weight decay,
sparsity, and dropout. The computation complexity of DNN depends upon the number of
parameters, number of layers, units per layers, the learning rate, and initial weights. Graphics
processing units (GPUs) have large processing capabilities of many-core architectures to
overcome the computation issue in DNN [33].
1.1.Key contributions
This study provides a detailed overview of existing and recent methods on brain tumor
segmentation of MRI images. Some of the state-of-the-art methods for brain tumor segmentation
on brain tumor segmentation (BraTS) dataset are illustrated in Fig. 3. Key contributions of this
study are as follows:
a) The paper motivates the young researchers to contribute their research knowledge in
this field.
b) It can guide the researchers to choose suitable methods for future development.
c) Various methods can be consolidated on BraTS dataset and compared using Dice
similarity metrics. Consolidation of results from existing and recent methods is useful
to compare their contribution in the field.
d) This study creates a pool of research on brain tumor segmentation.
Fig. 3.
State
1.2 Dataset details
Increasing bandwidth of the
tumor repositories to develop and enhance the methods.
available in the form
of clinical or simulated datasets for brain tumors
2019), BrainWeb repository,
internet brain segmentation repositories (
stroke lesion s
egmentation (ISLES),
provided in Table 1. From
the reviewed papers, about 46 percent used BraTS datasets,
percent used Harvard dataset,
and about
BrainWeb repository and remaining 18 percent used
hospitals.
-of-the-art brain tumor segmentation methods
Increasing bandwidth of the
i
nternet helps to access the huge volume of online brain
tumor repositories to develop and enhance the methods.
Numerous online repositories are
of clinical or simulated datasets for brain tumors
such as
internet brain segmentation repositories (
IBSR),
Harvard,
egmentation (ISLES),
etc.
The datasets and their corresponding
the reviewed papers, about 46 percent used BraTS datasets,
and about
13 percent used IBSR datasets, about 3 percent from
BrainWeb repository and remaining 18 percent used
by other
datasets obtained from various
nternet helps to access the huge volume of online brain
Numerous online repositories are
such as
BraTS (2012-
Harvard,
Ischemic
The datasets and their corresponding
websites are
the reviewed papers, about 46 percent used BraTS datasets,
about 20
13 percent used IBSR datasets, about 3 percent from
datasets obtained from various
Table 1: Open databases for brain tumor analysis
Dataset Name Website
BraTS
(2012-2019)
https://www.smir.ch/BRATS/Start2012
https://www.smir.ch/BRATS/Start2013
https://www.virtualskeleton.ch/BRATS/Start2014
https://www.smir.ch/BRATS/Start2015
https://www.smir.ch/BRATS/Start2016
https://www.med.upenn.edu/sbia/brats2017/data.html
https://www.med.upenn.edu/sbia/brats2018/data.html
https://www.med.upenn.edu/cbica/brats2019/data.html
IBSR https://www.nitrc.org/projects/ibsr/
BrainWeb https://Brainweb.bic.mni.mcgill.ca/brainweb
Harvard http://www.med.harvard.edu/aanlib/
Recently, researchers are contributing more on BraTS variants datasets from BraTS2012
to BraTS2019. These datasets have high-grade glioma (HGG) and low-grade glioma (LGG) for
tumor segmentation with T1 and contrast-enhanced T1 (T1C), T2, FLAIR and gold standard.
BraTS2012 to BraTS2015 datasets consists of 30 patient volumes among which 20 volumes
belong to HGG, and 10 volumes belong to LGG. BraTS2016 dataset consists of 22 HGG
volumes and 15 LGG volumes. BraTS2017 to BraTS2019 datasets are different from earlier
ones. These datasets are collected from The Cancer Genome Atlas (TCGA) and The Cancer
Imaging Archive (TCIA). They were categorized into pre-operative or post-operative by TCGA.
All the pre-operative TCIA scans (135 HGG and 103 LGG) training volumes were included in
BraTS2017 to BraTS2019 datasets. A survival data file associated with the HGG and LGG
volumes also provided in .CSV format. This data is used for predicting the survival rate of
patients.
One of the important issues in medical image analysis for brain tumor segmentation is
validation. Since validation has its direct impact on surgical planning, calculation of overlap
between the ground truth and the segmented image plays a vital role in medical imaging. Dice
similarity coefficient is evaluation metrics in brain tumor segmentation and their value between
zero to one where zero indicates no overlap and one indicates perfect overlap.
The rest of the paper organizes as follows: Section 2 provides a detailed overview of
conventional methods on brain tumor segmentation. Section 3 explains the machine learning
methods, and Section 4 focus on the deep learning methods for brain tumor segmentation.
Comparison and discussion are in section 5, and section 6 depicts the conclusions.
2. Conventional Methods for Brain Tumor Segmentation
The standard image processing methods such as threshold-based methods, region-based
methods and edge-based methods are known to be conventional segmentation methods. These
methods are commonly employed by for two-dimensional segmentation.
2.1. Threshold-based segmentation
Threshold-based segmentation methods give their results based on one or more threshold
values (multilevel threshold) for converting grey scale image into a binary image [34]. There are
two types of thresholding methods namely, and global or local thresholding. Global thresholding
plays a vital role in segmenting objects and the background using a single threshold value for an
image. It is suitable for images containing different objects with homogenous intensity and high
background. The greatest disadvantage of global thresholding is that the threshold selection will
become very difficult if the image contrast is low [35]. Local thresholding makes use of unique
threshold value applied on sub-images obtained from the whole image. Local statistical feature
and prior knowledge are used to estimate the threshold value for local thresholding [36].
Multilevel thresholding segments a grey image into 𝑛 distinct regions for multi-region
segmentation. Some researchers had proposed multilevel thresholding along with particle swarm
optimization technique for segmenting brain tumor from MRI brain scans [37].
2.2. Region-based segmentation
Region-based segmentation methods work with a similarity criterion, according to the
homogenous neighborhood pixels [38]. In the brain tumor segmentation, region growing and
watershed methods are most commonly used. Region growing connects the regions of similar
pixels from an image based on similarity criteria and seed point [39]. Seeds are determined by
either hard or automatic [40]. Chong et al. had mentioned that region-based methods are very
effective and consume a less computation time for MRI brain tumor segmentation [41]. Partial
volume effect is the drawback of region growing method since it limits the accuracies due to
blurs the distinctive intensities at the borders of different tissue classes [42].
Watershed segmentation is based on the idea behind the behaviour of water on the
landscape. Multi-scale watershed transformation is used to segment the brain tumors [43, 44].
Rather than performing automatic watershed methods, some researchers have performed
hierarchical user-assisted watershed segmentation methods on MRI brain images [45]. The
disadvantage of watershed methods is the over-segmentation problem. This problem had been
overcome by some of the researchers using automatic methods [46 - 48]. Ratan et al. had
proposed a multi-parameter based brain tumor segmentation and quantization method using MRI
brain images.
2.3. Edge-based segmentation
Edge-based segmentation is based on the deformable contours and classified into
parametric deformable and geometric deformable. Under the influence of internal and external
forces, these deformable contour areas are iteratively deformed [49 - 53]. Parametric deformable
methods are used to estimate the accurate boundaries of any type of brain tumors. Khotanlu et al.
had developed a parametric deformable model to segment tumor from MRI of human brain scans
[54]. Geometric deformable models overcome the difficulty in handling topological changes for
the splitting and merging of contours. Malladi et al. had developed a geometric deformable
model for brain tumor segmentation [55].
The major disadvantages of edge-based segmentation methods are that they required
more users’ interactions. As a consolidation, conventional methods are tough to achieve good
segmentation results, and faced difficulties are also discussed. Mostly these methods were
combined with machine learning and deep learning methods for refinement or as preprocessing.
3. Machine Learning Methods for Brain Tumor Segmentation
Machine learning algorithms are capable of learning different patterns and relationships
from the data and make accurate predictions [56]. It can be categorized into supervised,
unsupervised and semi-supervised algorithms based on the utilization of labeled training data
[57]. Datahasfeature with labels named as supervised learning. If the data contains unlabelled
features and called as unsupervised learning. The combination of supervised and unsupervised
learning is known to be semi-supervised learning. In literature, fuzzy c means (FCM), k-means,
Markov random fields (MRF), Bayes, artificial neural networks (ANN), support vector machines
(SVM), atlas-based methods are used for brain tumor segmentation. Machine learning methods
are more reliable and precise than established radiological methods in the treatment of brain
tumors. It generally reduces the burden of radiologists in detecting abnormal slices from a huge
volume of data [58].
Fuzzy C means (FCM) clustering separates a group of data into two or more clusters in
pattern recognition [59]. It works by measuring the distance between the center of cluster and the
data point and produces the best results for brain tumor segmentation when compared to k-means
algorithm [60]. Researchers have proposed a knowledge-based method with multispectral
histogram using FCM for brain tumor segmentation [61]. Also, a knowledge-based FCM was
implemented for brain tumor segmentation, followed by three-dimensional connected
components to build the tumor shape [62]. A combination of region growing technique and
knowledge-based fuzzy technique was proposed for brain tumor segmentation [63]. This method
segments multispectral images more effectively when compared to other methods such as k-
means, SVM, etc. FCM algorithm takes time to convergence due to its iterative nature. This
problem was solved using the fast-generalized clustering algorithm and bias-corrected FCM
algorithm [64]. Bias corrected FCM proved to be the best in segmenting brain tumor in very less
computation time [65]. The standard FCM algorithm is modified to obtain accurate and fast
segmentation results in the case of mixed noises such as Gaussian, impulse, etc. [66]. In FCM
based methods, it is very difficult to determine the degree of attraction, which affects the
segmentation results. Degree of attraction is dependent on its location and features of
neighborhood pixels. The genetic algorithms (GA) are the best solutions for reaching an optimal
solution, but it doesn’t find an exact solution. Particle swarm optimization (PSO) enhances the
optimal solution search. Since GA’s and PSO’s were combined to determine the optimum value
of the degree of attraction [67].
Atlas-based algorithms are used for image registration of different images [68]. In
general, atlas-based segmentation methods were used for segmenting brain tissues and locate the
tumor done in three steps. Initially, affline registration is done. Next, the brain atlas provides a
brain tumor template, and finally, the seeded atlas is deformed by some flow principles and
growth of brain tumor [69].
SVM deals with supervised classification problems, and it is considered as a kernel-based
method [70]. It is the popular method that has been used for brain tumor segmentation because of
its ability of classification [71-74]. SVM based methods were proposed based on the ability to
learn the data without prior knowledge which achieved better segmentation results of brain MRI
when compared to FCM [75]. For segmenting brain tumor, a multi-kernel SVM along with
feature selection and fusion process was proposed [76]. When compared to single kernel SVM,
the results of multi-kernel SVM had improved the accuracy and reduced the error rates. State-of-
the-art machines learning methods for brain tumor segmentation are inferred in Table 2.
Table 2: State-of-the-art machines learning methods for brain tumor segmentation on BraTS dataset
Author Method Dataset Dice Score
Shivaraja et al. (2020) [77] Bayesian fuzzy clustering BraTS 2016 0.97
Zaka et al. (2020) [78] Texture Based Localization BraTS 2012 0.88
Khalid et al. (2017) [79] Multimodal segmentation
using machine learning BraTS 2013 0.88
Mohammadneza et al. (2017) [80] Random Forest BraTS 2012 0.88
Amiri et al. (2016) [81] Random Forest BraTS 2013 0.72
Srinivas Rao et al. (2016) [82] FCM BraTS 2013 0.96
Laszlo et al. (2015) [83] FCM BraTS 2012 0.70
Tustison et al. (2015) [84] Random Forest BraTS 2013 0.87
Wu et al. (2014) [85] Conditional Random Forest BraTS 2012 0.86
4. Deep Learning Methods for Brain Tumor Segmentation
The rapid development of technology introduces deep learning, which is good at
processing unstructured data like images and videos. Recent advancements of deep learning are
used to research in the biomedical industry. To be very specific, convolutional neural networks
(CNN) is a part of deep learning plays a very important role in brain tumor segmentation [86]. In
the conventional and machine learning segmentation methods, the hand-crafted image features
are fed into the system, whereas CNN learns the distinctive features directly from the input data.
The research in deep learning is based on the designing of network architecture rather than
extracting features. In general, CNN extracts the patches from images as inputs. It uses trainable
convolutional filters and sub-sampling to extract more complex features from the inputs.
Multimodal MRI brain tumor segmentation is the most recent methodology in deep
learning. It extracts the patches from all the imaging modalities of MRI such as T1, T1c, T2 and
FLAIR. These patches or cubic voxels are fused together and fed as input to the convolutional
neural networks. A three-dimensional CNN architecture for multimodal MRI glioma was
proposed [87]. In contrast to this method, an interpretation method was developed using two
dimensional CNN architectures [88]. This eradicates the problem of handling high dimensional
CNN architectures. In, the hyperbolic tangent function has been used as an activation function
whereas in the rectified linear unit is used as the activation function.
Due to the complex nature of deep learning technique, GPU has additional advantages
over traditional CPUs; these include having more computational units and higher bandwidth for
accessing the memory [89]. Deep learning takes more time on training with features and utilizes
the additional computation power from GPU. Speed of GPU measures in terms of folds (×)
compared with CPU computation. The invention of GPU reduces the computation time of deep
learning from weeks to hours. The computation time of deep learning algorithms is varying
based on the size of layers, parameters, number of features, size of the training data etc.
A novel approach for brain tumor segmentation was proposed with cascaded two
pathway CNN architecture using Nvidia GPU [90]. This two-dimensional method extracts small-
sized patches and large-sized patches at the same time. A similar method which was
implemented using two pathway approaches using a single CNN is proposed [91]. Some authors
had proposed brain tumor segmentation using deeper architectures along with small 3 x 3 filters
[92]. They have inferred that small-sized kernels produce better accuracies when compared to
large-sized kernels. Here, the leaky rectified linear unit is used as a nonlinear activation function.
An ensemble method which combines CNN with clustering techniques was proposed [93]. This
method uses local structure prediction along with CNN for brain tumor segmentation. The voxels
of image patches are extracted from ground truth images, and they were clustered using k-means
algorithm. Later two-dimensional CNN is used for classifying multimodal input image patches
into three clusters. Another method, which extracts input patches from T1, T1c, T2 and FLAIR
modalities separately [94]. These patches are processed by four different CNN’s, and outputs of
these four CNN’s are concatenated and used as feature maps for random forest classifier with
GPU. The performance of some notable CNN based segmentation methods are listed in Table 3.
Table 3: Recent Deep Learning Methods for Brain Tumor Segmentation
Author Method Dataset
Dice Coefficient
GPU Name
Complet
e Core Enha
nced
Bal et al. (2020)
[95]
Deep Convolutional
Neural Network BraTS2015 0.87 0.79 0.77 -
Nema et al. (2020)
[96]
RescueNet based on
generative adversarial
networks
BraTS2017 0.94 0.85 0.93
NVIDIA Tesla
V100 4 × 16 GB
GPU
Zexun et al. (2020)
[97]
Atrous Convolution
Feature Pyramid
BraTS2013,
BraTS2015,
BraTS2018
0.86 0.76 0.74
NVIDIA Titan XP
with a single GPU
core
Chen et al. (2019)
[98]
Dual force
convolutional neural
networks
BraTS2017 0.89 0.73 0.73 NVIDIA GTX
Titan Xp
Mlynarski et al.
(2019) [99]
U-NET with mixed
supervision BraTS2018 0.80 0.63 0.66 Nvidia GeForce
GTX 1080 Ti
Sajid et al. (2019)
[100]
Patch-based hybrid
convolutional neural
networks
BraTS2013 0.86 0.86 0.88 -
Wang et al. (2019)
[101]
Multimodal brain
tumor segmentation
using PP-NET
BraTS2018 0.94 - - NVIDIA Titan X
Hussain et al.
(2018) [102]
Patch based deep
convolutional neural
networks
BraTS2015 0.86 0.87 0.90 GPU Compatible
Zhao et al. (2018)
[103]
Integration of CNN
and conditional
random fields
BraTS2015 0.81 0.65 0.60 NVIDIA Tesla K80
GPUs
Urban et al. (2014)
[87]
Three-dimensional
convolutional neural
network architecture
BraTS2013 0.86 0.77 0.73 -
Zikie et al. (2014)
[88]
Patch-based
convolutional neural
network architecture
BraTS2013 0.83 0.73 0.69 -
Davy et al. (2014)
[91]
Double pathway
convolutional neural
network
BraTS2013 0.85 0.74 0.68 -
Dvorak and Menze
(2015) [93]
Combination of
convolutional neural
networks and k-means
algorithm.
BraTS2013 0.83 0.75 0.77 -
Havaei et al. (2016)
[90]
Simultaneous local
and global processing
using cascaded two
pathway CNN
BraTS2013 0.88 0.79 0.73 GTX Titan Black
Pereira et al. (2016)
[92]
Convolutional neural
networks with small
3x3 filters
BraTS2013 0.88 0.83 0.77 NVIDIA GeForce
GTX 980
5. Comparison and Discussion
In general, analysis of brain tumor is performed manually by medical professionals. This
is a challenging task due to the asymmetric nature of a brain tumor. Automatic segmentation
methods overcome the problems of manual segmentation since they assist the medical
professionals to make a final decision on further treatment. In this paper, several conventional,
machine learning and deep learning automatic segmentation methods and their performances
were presented. Difficulties found in the tumor segmentation; these three methods played a vital
role in the past few decades.
Generally, conventional methods are given crisp type results like edge detection,
thresholding, region merging or split. Therefore, this survey addressed that conventional methods
are useful in the preprocessing like brain skull stripping, enhancement, denoising, approximate
tumor region extraction, etc. [104]. Most of the conventional methods are embedded with the
advanced techniques on the pipeline to produce better segmentation results. It was lightweight
algorithms and required less computation power from the CPU.
In the machine learning methods, it was beneficial for complete or whole tumor
segmentation. Few machine learning methods are used in the preprocessing like tumorous image
selection [6]. Table 2 showed that different methods under machine learning using Dice
similarity index for complete tumor segmentation on BraTS2012 and BraTS2013. From Table 2,
it was observed that fuzzy based segmentation yields better accuracy than other methods. The
Table shows that performance of the method proposed by Shivaraja et al. using Bayesian fuzzy
clustering for whole tumor segmentation than other methods. These methods can effectively
utilize the computation power from CPU, and complex algorithms expect additional computation
power from GPU [105].
In the extension of soft computing, deep learning was effectively making the computer-
assisted systems in tumor diagnostics. Deep learning yielded accurate segmentation based on
their architecture such as optimal features, training data, number of layers and parameters etc. It
took more time on training data, and mandatorily required additional computation power from
GPU. Generally, cost of GPUs makes a deep learning process very expensive. Table 3 shows the
recent deep learning methods on BraTS datasets for segmenting brain tumor substructures. Table
showed that the deep learning methods are very suitable for tumor substructures segmentation
and classification. From the Table 3, Namma et al. developed a RescueNet based on generative
adversarial networks were given high segmenting Dice accuracy for complete, core, enhanced
tumor in 94%, 85% and 93% respectively. For in-depth classification, deep learning architectures
are more suggestible compare than machine learning and conventional methods.
6. Conclusion
This paper presented a overview on conventional to deep learning methods for brain
tumor segmentation. We have discussed several state-of-the-art methods for brain tumor
segmentation in the literature. This study gave an idea for selecting methods and techniques for
their research in tumor. This study concludes that the convention methods were frequently used
in the preprocessing step and support to adapt with other methods. Then, machine learning
methods are more suggestible for complete tumor segmentation and classification. Finally, deep
learning methods are very suitable for tumor substructures segmentation and classification. The
invention of GPU accelerates the speed of deep learning computation. This study will help the
researchers to find the best deep neural network architecture that suits for brain tumor analysis.
References
[1] Prince, J. L., & Links, J. M. (2006). Medical imaging signals and systems. Upper Saddle River, NJ:
Pearson Prentice Hall.
[2] Dhage, P., Phegade, M. R., & Shah, S. K. (2015, January). Watershed segmentation brain tumor detection.
In 2015 International Conference on Pervasive Computing (ICPC) (pp. 1-5). IEEE.
[3] Liang, Z. P., & Lauterbur, P. C. (2000). Principles of magnetic resonance imaging: a signal processing
perspective. SPIE Optical Engineering Press.
[4] Kalaiselvi, T. (2011). Brain portion extraction and brain abnormality detection from magnetic resonance
imaging of human head scans. Pallavi Publication.
[5] Drevelegas, A., & Papanikolaou, N. (2011). Imaging modalities in brain tumors. In Imaging of brain
tumors with histological correlations (pp. 13-33). Springer, Berlin, Heidelberg.
[6] Sriramakrishnan, P., Kalaiselvi, T., & Rajeswaran, R. (2019). Modified local ternary patterns technique for
brain tumour segmentation and volume estimation from mri multi-sequence scans with gpu cuda
machine. Biocybernetics and Biomedical Engineering, 39(2), 470-487.
[7] Donoso, R., Veloz, A., & Allende, H. (2010, November). Modified expectation maximization
algorithm for MRI segmentation. In Iberoamerican Congress on Pattern Recognition (pp. 63-70).
Springer, Berlin, Heidelberg.
[8] Pedoia,V., Balbi, S., and Binaghi, E., (2015). Fully automatic brain tumor segmentation by using
competitive EM and graph cut. Image analysis and processing–ICIAP,568- 578.
[9] Yousefi, S., Azmi, R., and Zahedi, M., (2012).Braintissue segmentation in MR images based on a
hybrid of MRF and social algorithms. Medical Image Analysis,16(4), 840-848.
[10] Merisaari, H., Parkkola, R., Alhoniemi, E., Teräs, M., Lehtonen, L., Haataja, L., ... &Nevalainen, O. S.
(2009). Gaussian mixture model-based segmentation of MR images taken from premature infant
brains. Journal of neuroscience methods, 182(1), 110-122.
[11] Somasundaram, K., & Kalaiselvi, T. (2010). Fully automatic brain extraction algorithm for axial T2-
weighted magnetic resonance images. Computers in biology and medicine, 40(10), 811-822.
[12] Aslam, A., Khan, E., & Beg, M. M. (2015). Improved edge detection algorithm for brain tumor
segmentation. Procedia Computer Science.
[13] Anithadevi, D., Perumal, K., (2018). Roughset and multi - thresholds based seeded region
growing algorithm for image segmentation. Artificial intelligence and evolutionary computations
in engineering systems, Springer,369-379.
[14] Rajendran, R. Dhanasekaran, (2012).Fuzzy clustering and deformable model for tumor segmentation on
MRI brain image: A combined approach. Procedia Eng., 30, 327-333.
[15] Sachdeva, J., Kumar, V., Gupta, I., Khandelwal, N., & Ahuja, C. K. (2012). A novel content-based active
contour model for brain tumor segmentation. Magnetic resonance imaging, 30(5), 694-715.
[16] Tanoori, B., Azimifar, Z., Shakibafar, A., & Katebi, S. (2011). Brain volumetry: an active contour model-
based segmentation followed by SVM-based classification. Computers in biology and medicine, 41(8),
619-632.
[17] Zabir, I., Paul, S., Rayhan, M. A., Sarker, T., Fattah, S. A., & Shahnaz, C. (2015, December). Automatic
brain tumor detection and segmentation from multi-modal MRI images based on region growing and level
set evolution. In 2015 IEEE International WIE Conference on Electrical and Computer Engineering
(WIECON-ECE) (pp. 503-506). IEEE.
[18] Zanaty, E. A., &Ghoniemy, S. (2016). Medical image segmentation techniques: an overview. International
Journal of informatics and medical data processing, 1(1), 16-37.
[19] Pei, L., Reza, S. M., Li, W., Davatzikos, C., &Iftekharuddin, K. M. (2017, March). Improved brain tumor
segmentation by utilizing tumor growth model in longitudinal brain MRI. In Medical Imaging 2017:
Computer-Aided Diagnosis (Vol. 10134, p. 101342L). International Society for Optics and Photonics.
[20] Shubhangi, D.C., Hiremath, P.S., (2009). Support vector machine (SVM)classifier for brain tumor
detection. In Proceedings of the International Conference on Advances in Computing, Communication and
Control,444- 448.
[21] Kapás, Z., Lefkovits, L., & Szilágyi, L. (2016, September). Automatic detection and segmentation of
brain tumor using random forest approach. In International Conference on Modeling Decisions for
Artificial Intelligence (pp. 301-312). Springer, Cham.
[22] Havaei, M., Jodoin, P. M., & Larochelle, H. (2014, August). Efficient interactive brain tumor
segmentation as within-brain kNN classification. In 2014 22nd International Conference on Pattern
Recognition (pp. 556-561). IEEE.
[23] Abdullah, S., Huda, S. N., Bohani, F. A., Nayef, B. H., Sahran, S., Al Akash, O., ... & Ismail, F. (2016).
Round randomized learning vector quantization for brain tumor imaging. Computational and mathematical
methods in medicine, 2016.
[24] Benson, C. C., Deepa, V., Lajish, V. L., & Rajamani, K. (2016, September). Brain tumor segmentation
from MR brain images using improved fuzzy c-means clustering and watershed algorithm. In 2016
International Conference on Advances in Computing, Communications and Informatics (ICACCI) (pp. 187-
192). IEEE.
[25] Singh, B., & Aggarwal, P. (2017, October). Detection of brain tumor using modified mean-shift based
fuzzy c-mean segmentation from MRI Images. In 2017 8th IEEE Annual Information Technology,
Electronics and Mobile Communication Conference (IEMCON) (pp. 536-545). IEEE.
[26] Anand, A. (2017). Brain tumor segmentation using watershed technique and self organizing maps. Indian
Journal of Science and Technology, 10(44).
[27] Ganesh, M., Naresh, M., & Arvind, C. (2017). MRI brain image segmentation using enhanced adaptive
fuzzy K-means algorithm. Intelligent Automation & Soft Computing, 23(2), 325-330.
[28] Pan, Y., Huang, W., Lin, Z., Zhu, W., Zhou, J., Wong, J., & Ding, Z. (2015, August). Brain tumor grading based on neural
networks and convolutional neural networks. In 2015 37th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC) (pp. 699-702). IEEE.
[29] Dong, H., Yang, G., Liu, F., Mo, Y., & Guo, Y. (2017, July). Automatic brain tumor detection and
segmentation using u-net based fully convolutional networks. In annual conference on medical image
understanding and analysis (pp. 506-517). Springer, Cham.
[30] Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., ... & Larochelle, H. (2017).
Brain tumor segmentation with deep neural networks. Medical image analysis, 35, 18-31.
[31] Chen, L., Wu, Y., DSouza, A. M., Abidin, A. Z., Wismüller, A., & Xu, C. (2018, March). MRI tumor
segmentation with densely connected 3D CNN. In Medical Imaging 2018: Image Processing (Vol. 10574,
p. 105741F). International Society for Optics and Photonics.
[32] Kalaiselvi, T., Sriramakrishnan, P., & Somasundaram, K. (2017). Survey of using GPU CUDA
programming model in medical image analysis. Informatics in Medicine Unlocked, 9, 133-144.
[33] Suzuki, H., & Toriwaki, J. I. (1991). Automatic segmentation of head MRI images by knowledge guided
thresholding. Computerized medical imaging and graphics, 15(4), 233-240.
[34] Harris, G. J., Barta, P. E., Peng, L. W., Lee, S., Brettschneider, P. D., Shah, A., ... & Pearlson, G. D.
(1994). MR volume segmentation of gray matter and white matter using manual thresholding: dependence
on image brightness. American journal of neuroradiology, 15(2), 225-230.
[35] Adams, R., & Bischof, L. (1994). Seeded region growing. IEEE Transactions on pattern analysis and
machine intelligence, 16(6), 641-647.
[36] Kaur, T., Saini, B. S., & Gupta, S. (2016). Optimized multi threshold brain tumor image segmentation
using two-dimensional minimum cross entropy based on co-occurrence matrix. In Medical imaging in
clinical applications (pp. 461-486). Springer, Cham.
[37] Wong, K. P. (2005). Medical image segmentation: methods and applications in functional imaging.
In Handbook of biomedical image analysis (pp. 111-182). Springer, Boston, MA
[38] Mittelhaeusser, G., & Kruggel, F. (1995, April). Fast segmentation of brain magnetic resonance
tomograms. In International Conference on Computer Vision, Virtual Reality, and Robotics in
Medicine (pp. 237-241). Springer, Berlin, Heidelberg.
[39] Kaus, M. R., Warfield, S. K., Nabavi, A., Black, P. M., Jolesz, F. A., & Kikinis, R. (2001). Automated
segmentation of MR images of brain tumors. Radiology, 218(2), 586-591.
[40] Chong, V. F., Zhou, J. Y., Khoo, J. B., Huang, J., & Lim, T. K. (2004). Tongue carcinoma: tumor volume
measurement. International Journal of Radiation Oncology* Biology* Physics, 59(1), 59-66.
[41] Sato, M., Lakare, S., Wan, M., Kaufman, A., & Nakajima, M. (2000, September). A gradient magnitude
based region growing algorithm for accurate segmentation. In Proceedings 2000 International Conference
on Image Processing (Cat. No. 00CH37101) (Vol. 3, pp. 448-451). IEEE.
[42] Letteboer, M., Niessen, W., Willems, P., Dam, E. B., & Viergever, M. (2001). Interactive multi-scale
watershed segmentation of tumors in MR brain images. In Proc. of the IMIVA workshop of MICCAI.
[43] Dam, E., Loog, M., & Letteboer, M. (2004, August). Integrating automatic and interactive brain tumor
segmentation. In Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR
2004. (Vol. 3, pp. 790-793). IEEE.
[44] Cates, J. E., Whitaker, R. T., & Jones, G. M. (2005). Case study: an evaluation of user-assisted hierarchical
watershed segmentation. Medical Image Analysis, 9(6), 566-578.
[45] Ratan, R., Sharma, S., & Sharma, S. K. (2009). Multiparameter segmentation and quantization of brain
tumor from MRI images. Indian Journal of Science and Technology, 2(2), 11-15.
[46] Salman, S. D., & Bahrani, A. A. (2010). Segmentation of tumor tissue in gray medical images using
watershed transformation method. Int. J. Adv. Comp. Techn., 2(4), 123-127
[47] Bleau, A., & Leon, L. J. (2000). Watershed-based segmentation and region merging. Computer Vision and
Image Understanding, 77(3), 317-370.
[48] Toennies, K., Rak, M., & Engel, K. (2014). Deformable part models for object detection in medical
images. Biomedical engineering online, 13(S1), S1.
[49] He. L (2008). A Comparative study of deformable contour methods on medical image segmentation. Image
and Vision Computing, 26(2), 141-163.
[50] Liang J, T. Melnerney, D. Terzopoulos (2006). United snakes. Medical Image Analysis, 10(2), 215-233.
[51] Germond L (2000). A Cooperative framework for segmentation of MRI brain scans. Artificial Intelligent
Medicine, 20(1), 77-93.
[52] Adiga P S, Chaudhuri B B (2000). Region based techniques for segmentation of volumetric
histopathological images, Computational Methods Programs Biomedicine, 61(1), 23-47.
[53] Khotanlou, H., Colliot, O., Atif, J., & Bloch, I. (2009). 3D brain tumor segmentation in MRI using fuzzy
classification, symmetry analysis and spatially constrained deformable models. Fuzzy sets and
systems, 160(10), 1457-1473.
[54] Malladi, R., Sethian, J. A., & Vemuri, B. C. (1995). Shape modeling with front propagation: A level set
approach. IEEE transactions on pattern analysis and machine intelligence, 17(2), 158-175.
[55] Bishop, C. M. (2006). Pattern recognition and machine learning. springer.
[56] Mitchell, T. M. (2006). The discipline of machine learning (Vol. 9). Pittsburgh: Carnegie Mellon
University, School of Computer Science, Machine Learning Department.
[57] Duda, R. O., Hart, P. E., & Stork, D. G. (2012). Pattern classification. John Wiley & Sons.
[58] Bezdek, J. C., (1981). Pattern Recognition with Fuzzy Objective Function Algorithms. Kluwer Academic
Publishers.
[59] Hall, L. O., Bensaid, A. M., Clarke, L. P., Velthuizen, R. P., Silbiger, M. S., & Bezdek, J. C. (1992). A
comparison of neural network and fuzzy clustering techniques in segmenting magnetic resonance images of
the brain. IEEE transactions on neural networks, 3(5), 672-682.
[60] Clark, M. C., Hall, L. O., Goldgof, D. B., Velthuizen, R., Murtagh, F. R., & Silbiger, M. S. (1998).
Automatic tumor segmentation using knowledge-based techniques. IEEE transactions on medical
imaging, 17(2), 187-201.
[61] Fletcher-Heath, L. M., Hall, L. O., Goldgof, D. B., & Murtagh, F. R. (2001). Automatic segmentation of
non-enhancing brain tumors in magnetic resonance images. Artificial intelligence in medicine, 21(1-3), 43-
63.
[62] Lin, G. C., Wang, W. J., Kang, C. C., & Wang, C. M. (2012). Multispectral MR images segmentation
based on fuzzy knowledge and modified seeded region growing. Magnetic resonance imaging, 30(2), 230-
246.
[63] Szilagyi, L., Benyo, Z., Szilágyi, S. M., & Adam, H. S. (2003, September). MR brain image segmentation
using an enhanced fuzzy c-means algorithm. In Proceedings of the 25th Annual International Conference
of the IEEE Engineering in Medicine and Biology Society (IEEE Cat. No. 03CH37439) (Vol. 1, pp. 724-
726). IEEE.
[64] Cai, W., Chen, S., & Zhang, D. (2007). Fast and robust fuzzy c-means clustering algorithms incorporating
local information for image segmentation. Pattern recognition, 40(3), 825-838.
[65] Szilágyi, L., Szilágyi, S. M., & Benyó, Z. (2007, August). A modified fuzzy c-means algorithm for MR
brain image segmentation. In International Conference Image Analysis and Recognition (pp. 866-877).
Springer, Berlin, Heidelberg.
[66] Forouzanfar, M., Forghani, N., & Teshnehlab, M. (2010). Parameter optimization of improved fuzzy c-
means clustering algorithm for brain MR image segmentation. Engineering Applications of Artificial
Intelligence, 23(2), 160-168.
[67] Maintz, J. A., & Viergever, M. A. (1998). A survey of medical image registration. Medical image
analysis, 2(1), 1-36.
[68] Cuadra, M. B., Pollo, C., Bardera, A., Cuisenaire, O., Villemure, J. G., & Thiran, J. P. (2004). Atlas-based
segmentation of pathological MR brain images using a model of lesion growth. IEEE transactions on
medical imaging, 23(10), 1301-1314.
[69] Vapnik, V. (2013). The nature of statistical learning theory. Springer science & business media.
[70] Zhou, J., Chan, K. L., Chong, V. F. H., & Krishnan, S. M. (2006, January). Extraction of brain tumor from
MR images using one-class support vector machine. In 2005 IEEE Engineering in Medicine and Biology
27th Annual Conference (pp. 6411-6414). IEEE.
[71] H. Cai, R. Verma, Y. Ou, S.-K. Lee, E. R. Melhem, and C. Davatzikos (2007). Probabilistic segmentation
of brain tumorsbased on multi-modality magnetic resonance images. InBiomedical Imaging: From Nano to
Macro,600-603.
[72] R. Verma, E. I. Zacharaki, Y. Ou, H. Cai, S. Chawla,S.-K. Lee, E. R. Melhem, R. Wolf, and C. Davatzikos
(2008).Multiparametric tissue characterization of brain neoplasmsand their recurrence using pattern
classification of MR images, Academic Radiology, 15(8), 966-977.
[73] Ruan, S., Lebonvallet, S., Merabet, A., & Constans, J. M. (2007, April). Tumor segmentation from a
multispectral MRI images by using support vector machine classification. In 2007 4th IEEE International
Symposium on Biomedical Imaging: From Nano to Macro (pp. 1236-1239). IEEE.
[74] Ruan, S., Zhang, N., Liao, Q., & Zhu, Y. (2011, March). Image fusion for following-up brain tumor
evolution. In 2011 IEEE International Symposium on Biomedical Imaging: From Nano to Macro (pp. 281-
284). IEEE.
[75] Zhang, N., Ruan, S., Lebonvallet, S., Liao, Q., & Zhu, Y. (2009, November). Multi-kernel SVM based
classification for brain tumor segmentation of MRI multi-sequence. In 2009 16th IEEE International
Conference on Image Processing (ICIP) (pp. 3373-3376). IEEE.
[76] Raja, P. S. (2020). Brain tumor classification using a hybrid deep autoencoder with Bayesian fuzzy
clustering-based segmentation approach. Biocybernetics and Biomedical Engineering, 40(1), 440-453.
[77] Rehman, Z. U., Zia, M. S., Bojja, G. R., Yaqub, M., Jinchao, F., &Arshid, K. (2020). Texture based
Localization of a Brain tumor from MR-images by using a Machine learning Approach. Medical
Hypotheses, 109705.
[78] Usman, K., Rajpoot, K (2017). Brain tumor classification from multi-modality MRI using wavelets and
machine learning. Pattern Analysis and Applications, 20, 871–881.
[79] Soltaninejad, M., Yang, G., Lambrou, T., Allinson, N., Jones, T. L., Barrick, T. R., ... & Ye, X. (2017).
Automated brain tumour detection and segmentation using superpixel-based extremely randomized trees in
FLAIR MRI. International journal of computer assisted radiology and surgery, 12(2), 183-203.
[80] Amiri, S., Rekik, I., &Mahjoub, M. A. (2016, March). Deep random forest-based learning transfer to SVM
for brain tumor segmentation. In 2016 2nd International Conference on Advanced Technologies for Signal
and Image Processing (ATSIP) (pp. 297-302). IEEE.
[81] Rao, B. S., & Reddy, E. S. (2016). An Efficient Anti-noise Fast FCM clustering for Glioblastoma
Multiforme Tumor segmentation. International Journal of Computer Science and Information
Security, 14(4), 126.
[82] Szilagyi, L., Lefkovits, L., &Benyo, B. (2015, August). Automatic brain tumor segmentation in
multispectral MRI volumes using a fuzzy c-means cascade algorithm. In 2015 12th international
conference on fuzzy systems and knowledge discovery (FSKD) (pp. 285-291). IEEE.
[83] Tustison, N. J., Shrinidhi, K. L., Wintermark, M., Durst, C. R., Kandel, B. M., Gee, J. C., ... &Avants, B. B.
(2015). Optimal symmetric multimodal templates and concatenated random forests for supervised brain
tumor segmentation (simplified) with ANTsR. Neuroinformatics, 13(2), 209-225.
[84] Wu, W., Chen, A. Y., Zhao, L., & Corso, J. J. (2014). Brain tumor detection and segmentation in a CRF
(conditional random fields) framework with pixel-pairwise affinity and superpixel-level
features. International journal of computer assisted radiology and surgery, 9(2), 241-253.
[85] Ciresan, D., (2012). Deep neural networks segment neuronal membranes in electron microscopy images.
Advances in neural information processing systems, 2843–2851.
[86] Urban, G., (2014). Multi-modal brain tumor segmentation using deep convolutional neural networks.
MICCAI Multimodal Brain Tumor Segmentation Challenge, 31–35.
[87] Zikic, D., Ioannou, Y., Brown, M., &Criminisi, A. (2014). Segmentation of brain tumor tissues with
convolutional neural networks. Proceedings MICCAI-BRATS, 36-39.
[88] Raju, K., &Chiplunkar, N. N. (2018). A survey on techniques for cooperative CPU-GPU
computing. Sustainable Computing: Informatics and Systems, 19, 72-85.
[89] Havaei, M., Davy, A., Warde-Farley, D., Biard, A., Courville, A., Bengio, Y., ... & Larochelle, H. (2017).
Brain tumor segmentation with deep neural networks. Medical image analysis, 35, 18-31.
[90] Davy, A., (2014). Brain tumor segmentation with deep neural networks. MICCAI Multimodal Brain Tumor
Segmentation Challenge,1–5.
[91] Pereira, S, Pinto A, Alves V, Silva, C A (2016). Brain tumor segmentation using convolutional neural
networks in MRI images. IEEE Transactions on Medical Imaging,35(5), 1240–1251.
[92] Dvorak, P, Menze B (2015). Structured prediction with convolutional neural networks for multimodal brain
tumor segmentation. MICCAIMultimodal Brain Tumor Segmentation Challenge, 13–24.
[93] Rao, V., Sarabi, M S., Jaiswal A. Brain tumor segmentation with deep learning. MICCAI Multimodal Brain
Tumor Segmentation Challenge, 56–59.
[94] Bal, A., Banerjee, M., Sharma, P., &Chaki, R. (2020). A Multi-class Image Classifier for Assisting in
Tumor Detection of Brain Using Deep Convolutional Neural Network. In Advanced Computing and
Systems for Security (pp. 93-111). Springer, Singapore.
[95] Nema, S., Dudhane, A., Murala, S., & Naidu, S. (2020). RescueNet: An unpaired GAN for brain tumor
segmentation. Biomedical Signal Processing and Control, 55, 101641.
[96] Zhou, Z., He, Z., & Jia, Y. (2020). AFPNet: A 3D Fully Convolutional Neural Network with Atrous-
convolution Feature Pyramid for Brain Tumor Segmentation via MRI Images. Neurocomputing.
[97] Chen, S., Ding, C., & Liu, M. (2019). Dual-force convolutional neural networks for accurate brain tumor
segmentation. Pattern Recognition, 88, 90-100.
[98] Mlynarski, P., Delingette, H., Criminisi, A., & Ayache, N. (2019). Deep learning with mixed supervision
for brain tumor segmentation. Journal of Medical Imaging, 6(3), 034002.
[99] Sajid, S., Hussain, S., & Sarwar, A. (2019). Brain tumor detection and segmentation in MR images using
deep learning. Arabian Journal for Science and Engineering, 44(11), 9249-9261
[100] Wang, Y., Li, C., Zhu, T., & Zhang, J. (2019). Multimodal brain tumor image segmentation using WRN-
PPNet. Computerized Medical Imaging and Graphics, 75, 56-65.
[101] Hussain, S., Anwar, S. M., & Majid, M. (2018). Segmentation of glioma tumors in brain using deep
convolutional neural network. Neurocomputing, 282, 248-261.
[102] Zhao, X., Wu, Y., Song, G., Li, Z., Zhang, Y., & Fan, Y. (2018). A deep learning model integrating FCNNs
and CRFs for brain tumor segmentation. Medical image analysis, 43, 98-111.
[103] Kalaiselvi, T., Kumarashankar, P., &Sriramakrishnan, P. (2019). Three-Phase Automatic Brain Tumor
Diagnosis System Using Patches Based Updated Run Length Region Growing Technique. Journal of
digital imaging, 1-15.
[104] Kalaiselvi, T., &Sriramakrishnan, P. (2018). Rapid brain tissue segmentation process by modified FCM
algorithm with CUDA enabled GPU machine. International Journal of Imaging Systems and
Technology, 28(3), 163-174.
[105] Dhage, P., Phegade, M. R., & Shah, S. K. (2015, January). Watershed segmentation brain tumor detection. In 2015
International Conference on Pervasive Computing (ICPC) (pp. 1-5). IEEE.
[106] Hameurlaine, M., & Moussaoui, A. (2019). Survey of Brain Tumor Segmentation Techniques on Magnetic
Resonance Imaging. Nano Biomedical Engineering, 11(2), 178-191.
[107] Ortiz, A., Górriz, J. M., Ramírez, J., Salas-Gonzalez, D., & Llamas-Elvira, J. M. (2013). Two fully-
unsupervised methods for MR brain image segmentation using SOM-based strategies. Applied Soft
Computing, 13(5), 2668-2682.
... Tumor diagnosis at early stages enhances the probability of curing patients. Brain tumors are of two major types: non-cancerous (so-called benign), and cancerous (so-called malignant) 1 . In surgical processes, assigning tumor boundaries is very sensitive and must be identified accurately, as well as the border between the normal and abnormal tissues 1 . ...
... Brain tumors are of two major types: non-cancerous (so-called benign), and cancerous (so-called malignant) 1 . In surgical processes, assigning tumor boundaries is very sensitive and must be identified accurately, as well as the border between the normal and abnormal tissues 1 . Magnetic resonance imaging (MRI) provides a reliable source of information for highquality medical imaging. ...
Article
Full-text available
يعزز تشخيص أورام الدماغ المبكر احتمالات شفاء المرضى. مع تقدم استخدام الذكاء الاصطناعي في مجال الطب، أصبح اكتشاف أورام الدماغ أحد اهتمامات الباحثين. لا تزال مسألة تحديد أي شريحة في تسلسل الرنين المغناطيسي تعطي أفضل رؤية للورم تحديًا. قدم هذا البحث نهجًا إحصائيًا جديدًا لاستخراج الورم من فحص رنين الدماغ للمريض (تسلسل). في البداية، تم حساب دالة الكتلة الاحتمالية (PMF) لكل صورة في التسلسل. ثم تم تطبيق تقنية انحراف كولباك-ليبلر لتحديد الصورة أو الصور التي انحرفت عن الصور الصحية ذات الصلة. تم تحديد أفضل شريحة رؤية للورم باستخدام مقياس جذر متوسط الخطأ. بالإضافة إلى ذلك، تم تطبيق نهج تجميع لتجزئة الصور الورمية. تم استخدام طريقة التجميع الكمي (VQ) لتجميع الصور في 16 مجموعة مختلفة، في حين تم استخدام تقنية VQ عكسية لإنتاج صور ثنائية التنغيم. في النهاية، تم استخدام طريقة Teager ثنائي الأبعاد لاكتشاف الحواف لتحديد الورم. تم استخدام مجموعة بيانات خاصة تتضمن عشرين فحصًا بالرنين المغناطيسي (تسلسلًا) للاختبار وتقييم النظام.
... In previous studies, CNN-based encoder-decoder architectures have been recognized as having a significant role in various medical image segmentation tasks, encompassing CT [11], MRI [12], and x-ray [13]. However, medical image data exhibit low resolution, poor image quality, intricate tissue structures, noise interference, and pathological features associated with diseases. ...
Article
Full-text available
The precise delineation of glioma tumors is of paramount importance for surgical and radiotherapy planning. Presently, the primary drawbacks associated with the manual segmentation approach are its laboriousness and inefficiency. In order to tackle these challenges, a deep learning‐based automatic segmentation technique was introduced to enhance the efficiency of the segmentation process. We proposed ADT‐UNet, an innovative algorithm for segmenting glioma tumors in MR images. ADT‐UNet leveraged attention‐dense blocks and Transformer as its foundational elements. It extended the U‐Net framework by incorporating the dense connection structure and attention mechanism. Additionally, a Transformer structure was introduced at the end of the encoder. Furthermore, a novel attention‐guided multi‐scale feature fusion module was integrated into the decoder. To enhance network stability during training, a loss function was devised that combines Dice loss and binary cross‐entropy loss, effectively guiding the network optimization process. On the test set, the DSC was 0.933, the IOU was 0.878, the PPV was 0.942, and the Sen was 0.938. Ablation experiments conclusively demonstrated that the inclusion of all the three proposed modules led to enhanced segmentation accuracy within the model. The most favorable outcomes were observed when all the three modules were employed simultaneously. The proposed methodology exhibited substantial competitiveness across various evaluation indices, with the three additional modules synergistically complementing each other to collectively enhance the segmentation accuracy of the model. Consequently, it is anticipated that this method will serve as a robust tool for assisting clinicians in auxiliary diagnosis and contribute to the advancement of medical intelligence technology.
... For the past few years, U-net [1]-based image segmentation techniques have a wider application than object detection in both 2D and 3D aspects of medical images [2][3][4]. ...
Article
Full-text available
Brain tumors are highly lethal, representing 85–90% of all primary central nervous system (CNS) tumors. Magnetic resonance imaging (MRI) images are employed to identify and assess brain tumors. However, this process has historically relied heavily on the expertise of medical professionals and necessitated the involvement of a substantial number of personnel. To optimize the allocation of medical resources and improve diagnostic efficiency, this work proposes a DETR-based RPC–DETR model that utilizes the Transformer. We conducted comparative experiments using RPC–DETR with other traditional detectors on the same equipment to test the performance exhibited by the model with equal computational resources. Tested on the Br35H brain tumor dataset with 701 MRI images (500 training sets and 201 test sets), RPC–DETR surpasses the YOLO models in accuracy while utilizing fewer parameters. This advancement ensures more reliable and faster diagnosis. RPC–DETR achieves a 96% mAP with just 14M parameters, offering high accuracy in brain tumor detection with a lighter model, making it easier to implement in various medical settings.
... 8 This decade has seen the proposal of completely autonomous, highly trained deep learning techniques for the segmentation of brain tumours. 9 It has many advantages like adaptability, optimal feature extraction, selflearning, utilising the parallel architecture and providing accurate results. Convolutional neural networks (CNNs) have been applied to many segmention problems to achieve high accuracy. ...
Article
Full-text available
Glioma brain tumour is one of the life‐threatening diseases in the world. Tumour substructure segmentation by physicians is a time‐consuming task with the magnetic resonance imaging (MRI) technique due to the size of clinical data. An automatic and well‐trained method is essential to detect and segment the tumour which increase the survival of the patients. The proposed work aims to produce high accuracy on glioma substructures segmentation with less computation time using deep learning. From the literature survey, the following challenges are found: (i) computing complex spatial boundaries between normal and tumour tissues, (ii) feature reduction and (iii) overfitting problems. Hence, we proposed a fully automatic glioma tumour segmentation using a residual‐inception block (RIB) with a modified 3D U‐Net (RIBM3DU‐Net). It includes three phases: pre‐processing, modified 3D U‐Net segmentation and post‐processing. From the results, RIB with U‐Net enhances the segmentation accuracy. GPU parallel architecture reduces the computation time while training and testing. For quantitative analysis, comprehensive experiment results were computed and compared with state‐of‐the‐art methods. It achieves better Dice scores on enhancing tumour, tumour core and complete tumour of 87%, 87% and 94%, respectively. GPU speedup folds yield up to 48× when compared with CPU. Quantitatively, 3D glioma volume is rendered from the obtained segmented results and estimated using the Cavalieri estimator.
Article
Focused ultrasound-mediated opening of the blood-brain barrier offers a great opportunity to deliver therapeutics into hard-to-treat brain tumors such as glioblastoma multiforme or diffuse midline glioma. However, the potential of the technique to offer a time window for efficient nanomedicine delivery has not been thoroughly studied. Non-invasive and targeted delivery of large drug-loaded nanocarriers, such as liposomes, could offer a safe and scalable method of personalized therapy for the treatment of brain pathologies. Additionally, it is essential to monitor the safety and efficacy of such treatments, tracking drug delivery in real-time through quantitative medical imaging. In this study, liposomes were modified to have an MRI contrast agent (i.e., Gd) in both lipid membrane and core, while an infrared dye (i.e., CW800) was coupled to lipids introduced in the lipid bilayer for bimodal detection and treatment verification. Targeted delivery of 110 nm-in-diameter liposomes to the brain was quantified using 9.4-T MRI and near infrared fluorescence imaging. The spatiotemporal distribution of liposomes in vivo was assessed up to 4 h post treatment using T1 weighted MRI. In vivo MRI signal co-localized with NIRF signal from excised brains ex vivo. Passive acoustic detection during treatments revealed a correlation between acoustic signal and MRI contrast, providing a scalable metric for assessing clinical treatment efficacy in real-time. In conclusion, therapeutic ultrasound exposure can enhance delivery of large trackable nanoparticles into the brain, while enabling real-time treatment monitoring and verification.
Chapter
In the realm of pediatric brain tumors, the integration of artificial intelligence (AI) has catalyzed a paradigm shift, enriching various facets of clinical care, from detection and diagnosis to treatment response assessment. This book chapter meticulously explores the multifaceted applications of AI in pediatric brain tumor care, encompassing detection, segmentation, classification, differentiation, prognosis, and treatment response evaluation. By delving into the dynamic landscape of biomarkers, spanning imaging, molecular, genetics, and beyond, the chapter illuminates a road toward precision medicine, offering a foundation for refined diagnostics and tailored therapeutic interventions. It is important to highlight that although AI has delivered remarkable advantages to patient care, its widespread integration into healthcare systems presents a myriad of challenges.
Article
Purpose of review To provide an updated overview of artificial intelligence (AI) applications in neuro-oncologic imaging and discuss current barriers to wider clinical adoption. Recent findings A wide variety of AI applications in neuro-oncologic imaging have been developed and researched, spanning tasks from pretreatment brain tumor classification and segmentation, preoperative planning, radiogenomics, prognostication and survival prediction, posttreatment surveillance, and differentiating between pseudoprogression and true disease progression. While earlier studies were largely based on data from a single institution, more recent studies have demonstrated that the performance of these algorithms are also effective on external data from other institutions. Nevertheless, most of these algorithms have yet to see widespread clinical adoption, given the lack of prospective studies demonstrating their efficacy and the logistical difficulties involved in clinical implementation. Summary While there has been significant progress in AI and neuro-oncologic imaging, clinical utility remains to be demonstrated. The next wave of progress in this area will be driven by prospective studies measuring outcomes relevant to clinical practice and go beyond retrospective studies which primarily aim to demonstrate high performance.
Article
Full-text available
This work is integrated in the MICCAI Grand Challenge: MR Brain Image Segmentation 2013. It aims for the automatic segmentation of brain into Cerebrospinal fluid (CSF), Gray matter (GM) and White matter (WM). The provided dataset contains patients with white matter lesions, which makes the segmentation task more challenging. The proposed algorithm uses multi-sequence MR images to extract meaningful features and learn a Random Decision Forest that classifies each voxel of the image. The results show that it is robust to the presence of the white matter lesions, and the metrics show that the overall results are competitive.
Article
Full-text available
We have developed six convolutional neural network (CNN) models for finding optimal brain tumor detection system on high‐grade glioma and low‐grade glioma lesions from voluminous magnetic resonance imaging human brain scans. Glioma is the most common form of brain tumor. The models are constructed based on the different combinations and settings of hyperparameters with conventional CNN architecture. The six models are two layers with five epochs, five layers with dropout, five layers with stopping criteria (FLSC), FLSC and dropout (FLSCD), FLSC and batch normalization (FLSCBN), and FLSCBN and dropout. The models were trained and tested with BraTS2013 and whole brain atlas data sets. Among them, FLSCBN model yielded the best classification results for brain tumor detection. Experimental results revealed that our deep learning approach was better than the conventional state‐of‐art methods.
Article
Full-text available
Convolutional Neural Networks (CNNs) is a deep learning model used for image classification. The objective of our work is to analyze the different CNN models in the classification of benign tumor slices from MRI brain volumes and select a suitable CNN model for brain tumor detection. In this paper, we have developed six CNN models and trained using BraTS2013 dataset and tested with the WBA dataset. We have inferred the results for all the six models. The best model for classification of tumor slices is found among the six models. The accuracy of each and every model is recorded. Our models have attained about 96–99% of accuracy.
Article
Full-text available
In medical image processing, brain tumor detection and segmentation is a challenging and time-consuming task. Magnetic Resonance Image (MRI) scan analysis is a powerful tool in the recent technology that makes effective detection of the abnormal tissues from the brain. In the brain image, the size of a tumor can be varied for different patients along with the minute details of the tumor. It is a difficult task to diagnose and classify the tumor from numerous images for the radiologists. This paper developed a brain tumor classification using a hybrid deep autoencoder with a Bayesian fuzzy clustering-based segmentation approach. Initially, the pre-processing stage is performed using the non-local mean filter for denoising purposes. Then the BFC (Bayesian fuzzy clustering) approach is utilized for the segmentation of brain tumors. After segmentation, robust features such as, information-theoretic measures, scattering transform (ST) and wavelet packet Tsallis entropy (WPTE) methods are used for the feature extraction process. Finally, a hybrid scheme of the DAE (deep autoencoder) based JOA (Jaya optimization algorithm) with a softmax regression technique is utilized to classify the tumor part for the brain tumor classification process. The proposed scheme is implemented in a MATLAB environment. The simulation results are conducted by the BRATS 2015 database which proved that the proposed approach obtained the high classification accuracy (98.5 %) when compared to other state-of-art methods.
Article
Full-text available
Deep learning has attracted great attention in the medical imaging community as a promising solution for automated, fast and accurate medical image analysis, which is mandatory for quality healthcare. Convolutional neural networks and its variants have become the most preferred and widely used deep learning models in medical image analysis. In this paper, concise overviews of the modern deep learning models applied in medical image analysis are provided and the key tasks performed by deep learning models, i.e. classification, segmentation, retrieval, detection, and registration are reviewed in detail. Some recent researches have shown that deep learning models can outperform medical experts in certain tasks. With the significant breakthroughs made by deep learning methods, it is expected that patients will soon be able to safely and conveniently interact with AI-based medical systems and such intelligent systems will actually improve patient healthcare. There are various complexities and challenges involved in deep learning-based medical image analysis, such as limited datasets. But researchers are actively working in this area to mitigate these challenges and further improve health care with AI.
Article
Background The automatic segmentation of brain tumour from MRI medical images is mainly covered in this review. Recently, state-of-the-art performance is provided by deep learning-based approaches in the field of image classification, segmentation, object detection, and tracking tasks. Introduction The core feature deep learning approach is the hierarchical representation of features from images and thus avoiding domain-specific handcrafted features. Methods In this review paper, we have dealt with a Review of Deep Learning Architecture and Methods for MRI Brain Tumour Segmentation. First, we have discussed basic architecture and approaches for deep learning methods. Secondly, we have discussed the literature survey of MRI brain tumour segmentation using deep learning methods and its multimodality fusion. Then, the advantages and disadvantages of each method analyzed and finally concluded the discussion with the merits and challenges of deep learning techniques. Results and Conclusion The review of brain tumour identification using deep learning Techniques may help the research to have a better focus on it.
Article
Traditional deep convolutional neural networks for fully automatic brain tumor segmentation have two problems: spatial information loss caused by both the repeated pooling/striding and weak ability of multi-scale lesion processing. To overcome the first problem, we use a 3D atrous-convolution with a single stride to replace pooling/striding and build the backbone for feature learning. For the second problem, a 3D atrous-convolution feature pyramid is designed and added to the end of the backbone. By integrating with contextual features, this structure improves the discriminating ability of the overall model to segment tumors with various sizes. Finally, a 3D fully connected Conditional Random Field is constructed as a post-processing step for the network’s output to obtain structural segmentation of both the appearance and spatial consistency. Abundant ablation experiments carried on Magnetic Resonance Imaging datasets demonstrate that lossless feature computation and multi-scale information fusion driven by our method are feasible to address the above problems. Compared with the state-of-the-art methods on the public benchmarks, our method achieves competitive performance and can be efficiently implemented into the clinical medical application.
Research
Even with proper acquisition of brain tumor images, the accurate and reliable segmentation of tumors in brain is a complicated job. Automatic segmentation become possible with development of deep learning algorithms that brings plethora of solutions in this research prospect. In this paper, we designed a network architecture named as residual cyclic unpaired encoder-decoder network (RescueNet) using residual and mirroring principles. RescueNet uses unpaired adversarial training to segment the whole tumor followed by core and enhance regions in a brain MRI scan. The problem in automatic brain tumor analysis is preparing large scale labeled data for training of deep networks which is a time consuming and tedious task. To eliminate this need of paired data we used unpaired training approach to train the proposed network. Performance evaluation parameters are taken as DICE and Sensitivity measure. The experimental results are tested on BraTS 2015 and BraTS 2017 [1] dataset and the result outperforms the existing methods for brain tumor segmentation. The combination of domain-specific segmentation methods and general-purpose adversarial learning loomed to leverage huge advantages for medical imaging applications and can improve the ability of automated algorithms to assist radiologists.