Review
Survey of Brain Tumor Segmentation Techniques on Magnetic Resonance Imaging
Messaoud Hameurlaine 1*, Abdelouahab Moussaoui 2*
1 MESD Laboratory, Elwancharissi University Center, Tissemsilt, Algeria.
2 Faculty of Sciences, Ferhat Abbas University, Setif, Algeria.
* Corresponding author. E-mail: hamessainf@yahoo.fr; moussaoui_abdel@yahoo.fr
Received: Jan. 3, 2019; Accepted: Apr. 26, 2019; Published: Jun. 13, 2019
Citation: Messaoud Hameurlaine, Abdelouahab Moussaoui, Survey of Brain Tumor Segmentation Techniques on Magnetic Resonance Imaging. Nano Biomed. Eng., 2019, 11(2): 178-191.
DOI: 10.5101/nbe.v11i2.p178-191.
Abstract
Brain tumor extraction is challenging task because brain image and its structure are complicated that can be analyzed only by expert physicians or radiologist. Brain tumor detection and segmentation is one of the most challenging and time consuming task in medical image processing. The image segmentation is a very difficult job in the image processing and challenging task for clinical diagnostic tools. MRI (Magnetic Resonance Imaging) is a visualization medical technique, which provides plentiful information about the human soft tissue, which helps in the diagnosis of brain tumor. Accurate segmentation of the MRI images is extremely important and essential for the exact diagnosis by computer aided clinical tools. There are different types of segmentation algorithms for MRI brain images. This paper is to check existing approaches of Brain tumor segmentation techniques in MRI image for Computer aided diagnosis.
Keywords: Magnetic resonance imaging; Medical imaging; Brain tumor; Segmentation; Image processing; Computer aided diagnosis
Introduction
The National Brain Tumor Foundation (NBTF) for research in the United States estimated the death of 13000 patients while 29000 underwent primary brain tumor diagnosis every year [1]. Depending on the origin and growth, brain tumor is classified into two types: The primary brain tumor develops at the original site of the tumor, and the secondary brain tumor is the cancer that spreads to the other parts of the body. Nowadays, biomedical imaging has been very important for many applications for radiologist to diagnose the patient treatment related problems. At present, imaging technology is a must for patient diagnosis. The various medical images like magnetic resonance imaging (MRI), ultrasound, computed tomography (CT), X-ray, etc. play an important role in the process of disease, diagnosing and treating. The recent revolution in medical imaging results from techniques such as CT and MRI, can provide detailed information about disease, and can identify many pathologic conditions to give an accurate diagnosis. For the diagnosis and treatment of patients suffering from brain tumor, specialists take the assistance of MRI scans of the brain. In any case, the analysis of MRI scan is done manually by the specialist, which is tedious, and the precision of the outcome depends on the experience of the specialist. The conclusions may differ from one doctor to another. Thus, there is a need to overcome these issues and to automate or robotize the investigative procedure of brain tumor in MRI images. For this purpose, biomedical image processing techniques are applied to MRI scans. Thus, the segmentation and further characterization of brain tumor from MRI scans remain a broad range of research in the field of medical science. Computer aided diagnosis system has been developed for automatic detection of brain tumor through MRI. Improving the ability to identify early-stage tumors is an important goal for physicians, because early detection of class of disease is a key factor in producing successful treatments. Segmentation is a prime task of MRI processing. It is a process of dividing an image into multiple parts: Different tumor tissues (solid tumor, edema and necrosis) and normal brain tissues; Gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF). Moreover, the tumor cell characteristics, such as complex shape, heterogeneous intensity distribution, variability of the position of the tumor, and artifacts in the tumor also have a significant effect on diagnosis. Tumor heterogeneity describes the observation that different tumor cells show distinct morphological and phenotypic profiles, including cellular morphology, gene expression, metabolism, motility, proliferation, and metastatic potential. The heterogeneity of cancer cells introduces significant challenges in designing effective treatment strategies. Modalities, like T1-weighted, T2-weighted, or proton density (PD) images are utilized for various segmentation methods. Due to the good contrast of the T1-weighted images, they have been widely tested for different segmentation methods [2]. In recent years, many research publications describe the different segmentation approaches for medical image analysis which are reported by Al-Tamimi et al. [3]. Fig 1 shows the number of research papers published between 2000 and 2018. In the current survey, we present briefly different stages in brain tumor detection, and we explicitly analyze the developed automated MS lesion segmentation approaches through a comprehensive up-to-date state-of-the-art review. To this end, the approaches are categorized, in terms of their main features and properties. Furthermore, a qualitative and quantitative comparison of the state-of-the-art approaches is provided, while their strengths and weaknesses are illustrated. The ultimate goal of this survey is to provide aid in identifying the most promising research directions in the field. There are very extensive reviews on the methodologies published. Table 1 summarizes the number of methodologies reviewed by previous state-of-the-art as well as the current surveys. The current survey is focused mostly on automated MS lesion segmentation techniques published since 2010. Thus, a lot of categorizations of the related techniques are provided. The bibliography is very rich and a relatively large number of new methods have been published since 2013.
Fig. 1 Number of papers on brain tumor segmentation approaches published in 2000 - 2018 (edusol 2018 [4]).
Table 1 Survey since 2012
Survey papers |
Year |
Number of methods reviewed in each survey papers |
[5] [6] [7] [8] [3] [9] [10] [11] [12] [2] [13] Current survey |
2012 2012 2013 2013 2014 2014 2015 2018 2018 2018 2018 2019 |
44 34 55 39 31 49 32 45 31 36 61 72 |
Methodology
Detection of brain tumor from MRI images involves various phases such as preprocessing, feature extraction, segmentation and classification. Fig 2 shows different stages in brain tumor detection. One of the most important tasks for the tumor detection is preprocessing. Usually medical images appear inhomogeneous and of poor contrast, and require preprocessing for image enhancement. This stage is used for reducing image noise, highlighting edges, or displaying digital images. The removal of unwanted parts from the brain MR image, finding edge position for removing labels and smoothing the image will be processed by using innovatively new pre-processing methods. A wide variety of pre-processing techniques like linear, non-linear, fixed, adaptive, pixel-based or multi-scale, are applicable in different circumstances[14]. Segmentation methods have the ability to detect or identify the abnormal portion from the image, which is useful for analyzing the size, volume, location, texture and shape of the extracted image. Segmentation guides the result of the whole analysis, because the proceeding steps depend on the segmented regions. The main principle of segmentation algorithms is the intensity or texture variations of images using region growing, deformable templates, thresholding, and pattern recognition techniques like fuzzy clustering and neural networks. Also, techniques like region-based and edge segmentation, adaptive and global thresholding, gradient operators, watershed segmentation, hybrid segmentation and volumetric segmentation, supervised and unsupervised segmentation exist. Several researchers are currently working on this medical image segmentation area [15]. Feature extraction can be defined as the process of transforming or converting an image into its group of features. The different methods employed for feature extraction include texture features, co-occurrence matrix, Gabor features, wavelet transform based features, decision boundary feature extraction, minimum noise fraction transform, nonparametric weighted feature extraction and spectral mixture analysis. For feature reduction principal component analysis, linear discriminant analysis and independent component analysis are used. Integration of the feature extraction with the feature reduction algorithms leads to accurate systems that uses less number of features that can be extracted with less computational cost [8]. Feature selection algorithms popularly used are genetic algorithm, sequential backward selection (SBS), sequential forward selection (SFS), and particle swarm optimization (PSO), while principal component analysis (PCA), kernel PCA and ICA help in dimensionality reduction. Many methods used for feature selection are appropriate for biomedical image classification [16]. Three different techniques are often used, namely multiple kernel learning, a GA based approach having an SVM as decision function, and recursive feature elimination using many classifiers. The biomedical image classification is a very important stage for automated CAD system. In some approaches, segmentation problem is transformed into a classification problem and a brain tumor is segmented by training and classifying. Generally, a machine learning classification method for brain tumor segmentation requires large amounts of brain MRI scans with known ground truth from different cases to train on. Mainly, artificial intelligence and prior knowledge are combined to solve the segmentation problem. Currently, high segmentation performances are obtained by deep learning methods [13]. The brain MRI classification is achieved using supervised techniques like ANN, SVM, k-NN and unsupervised classification techniques such as self-organizing map (SOM) and FCM. Several algorithms and techniques have been developed for segment brain tumor regions from MR images. The most commonly used techniques were the C-means and fuzzy sets combined with other techniques to achieve better performance with the MR images uncertainties and regions. In fact, hybrid techniques and soft computing techniques as fuzzy logic, neural network and genetic algorithms have found wide applications in image segmentation. Also, PCNN and its modification forms are widely applied to image segmentation.
Fig. 2 Stages in brain tumor detection.
Current Trends in MRI-CAD Scheme
Methods
A wide variety of brain tumor segmentation techniques has been proposed. However, there is no standard segmentation technique that can produce satisfactory results for all imaging applications. Quite often, methods are optimized to deal with specific imaging modalities such as magnetic resonance imaging. Although it is hard to explicitly categorize the state-of-the-art MRI based MS lesion segmentation techniques because of large overlaps between them, the classification of methods is based on the following characteristics that all methods have: Input data handling, main strategy, and existence of supervision. Four categorizations are presented below. The first categorization is detailed by Danelakis and co-workers [11]. The categorization is based more on supervised and unsupervised methods, 2D and 3D volume based, than on the strategy used. Table 2 illustrates the proposed categorization of the state-of-the-art MS lesion segmentation techniques. The second categorization is presented by Mohan et al. [13]. The methodologies are subdivided into categories as fully automatic (fuzzy logic, adaptive neuro-fuzzy inference system, support vector machines, artificial neural networks, self-organizing maps, particle swarm optimization, random forest, miscellaneous methods, etc.) and semi-automatic (FCM, SVM, ANN) for 2D & 3D user interaction methods. This survey reveals that brain MRI segmentation method scan are classified into 6 major categories: Threshold based segmentation, region based segmentation, edge detection, clustering (hard and soft clustering, algorithms used FCM and K-means), statistical models (EM algorithm, MRF model) and ANN [17]. The third categorization is described by El-Dahshan et al. [9]. This study illustrates the classification of human brain in MRI is possible via supervised techniques such as artificial neural networks and support vector machine, and unsupervised classification techniques such as SOM and fuzzy C-means. Other supervised classification techniques, such as k-NN can be used to classify the normal/pathological T2-weighted MRI images. Also, hybrid intelligent systems using soft computing techniques are used for classifier design. Soft computing consists of several intelligent computing paradigms, including fuzzy logic, neural networks, and bio-inspired optimization algorithms (genetic algorithm and genetic programming), which can be used to produce powerful hybrid intelligent classification systems. Table 3 summarizes the segmentation techniques [9]. The studies on feature extraction and classification of brain MRI were suggested by El-Dahshan et al. [9] for a comparative study, where it can be seen that:
In the fourth categorization, the segmentation techniques have been divided by Gordillo et al. [18] into four major classes:
The first categorization can be adapted, based on input data handling, main strategy and existence of supervision, while the others use only main strategy. The advantages and disadvantages of the most used classifiers for human brain MR images are summarized in Table 4 [9, 18]. Hybrid techniques that combine two or more techniques and soft computing techniques like NN, fuzzy logic and GA have found wide applications in image segmentation. Kumar and co-workers [17] presented a review on the various hybrid segmentation methods, revealing that K-means had better performance and less computational complexity. Hence by applying K-means in conjunction with other methods, it is possible to increase the segmentation performance.
Table 2 Categorization proposed in [11]
Supervised |
Unsupervised |
||
3D volume |
2D image |
3D volume |
2D image |
Feature-based Data-driven Atlas-based Statistical |
Feature-based Data-driven Atlas-based Statistical |
Lesion- based Tissue-based Data-driven Atlas-based |
Lesion- based Tissue-based Data-driven Atlas-based |
Validation and comparison
Validation and comparison of the state of the art is crucial for any newly developed method. We would like to briefly cover the possibilities and challenges for evaluating and validating methods in the brain tumor image analysis. It would be optimal to compare any method against the real case. However, this is a big challenge in this field, if not impossible. In the lack of a well-accepted ground truth, the current gold standard for the evaluation is to compare with manual segmentations by an expert. However, this is an extremely time-consuming and tedious task; additionally, it is not objective. Another possibility for a first sanity check is to assess results on a synthetic dataset including ground truth. Generally, synthetic data lack important characteristics of real images. In the lack of a brain tumor database with ground-truth segmentations, that is available to a broad community of clinicians and researchers, so far most authors validated their algorithms on a limited number of cases from their own data. This makes it difficult to compare the performance of different methods against each other in an unbiased way. Therefore, and due to the different metrics used, the accuracy and speed of the individual methods, which have been collected from the respective publications, cannot be directly compared with each other. Table 5 summarizes a variety of databases and modalities used in the validation [13]. According to Mahesh et al. [2], analysis based on datasets and modalities utilized for experimentation of tumor recognition system is deliberated. Fig. 3 presents analysis based on the modality of MRI image used for validation of segmentation and classification techniques adopted for brain tumor recognition. Fig. 4 depicts pie chart for analysis based on datasets utilized. The evaluation of performance can be done by comparison between automated segmentations and ground truth. And it can be accomplished by either comparing each voxel in each lesion (voxel-to-voxel), or using the whole detected lesion (lesion-to-lesion). The voxels and lesions can be classified as a true positive (TP), false positive (FP), true negative (TN) or false negative (FN). The objective is to obtain the maximum of TPs and TNs, and at the same time reduce FPs and FNs. In practice, we must find the best trade-off between these values. In fact, there is permanent debate about the best method to evaluate performance of results. Table 6 summarizes the most common measures used to evaluate the MS lesion segmentation algorithms [11, 19]. Table 7 summarizes some evaluation results in the MS lesion segmentation algorithms.
Table 3 Overview of the most commonly used segmentation techniques in CAD [9, 10]
Groups of the methods |
General principles |
Methods with example papers |
Contour and shape based techniques |
Boundary and edge based image segmentation. |
Deformable model [20] ; Active contour [21, 22] Level set [23, 24] ; Atlas-based [25] |
Region based techniques |
Its classified as a pixel-based method (initially select the seed points ) |
Thresholding [26]; Edge based [27] Watershed [28]; Seed region [29] |
Statistical based techniques |
Label pixel determine based on intensity distribution of the image. |
EM (Expectation-maximization) [30, 31] MRF (Markov random field) [32] GMM (Gaussian mixture model) [33] |
Multiresolution based techniques |
Multiple scales based segmentation. |
DWT (Discrete wavelet transform)[34, 35] |
Machine learning based |
The algorithms that can learn and make predictive data. |
Supervised classifiers based ANN (artificial neural network) [36, 37] SVM (support vector machine) [38] DT (decision tree) [39] KNN (k-nearest neighbors) [40] LVQ (learning vector quantization) [41] Deep learning (CNN, U-Net) [37, 42-44] |
Unsupervised clustering based FCM (fuzzy C-mean) [45, 46] SOM (self organizing map) [47] K-mean [48] PCNN (pulse-coupled neural network) [44] FPCNN (feedback PCNN) [49] |
||
Hybrid based techniques |
Two or more approaches used to segment the image. |
EM+PCNN [50] FFT(fast fourier transform) + EM - GMM [51] FCM + LVQ [52] DWT + FCM [53] AC (active contour) + SVM [54] DWT + SOM [55] GA + SVM [56, 57] Level set + ANN [58] GA + ANN [59] GR (generalized rough) + FCM [60] DWT + GA + SVM [61] SOM + LVQ [55] PCM (probabilistic CM) + FCM [62] FCM + Level set [63] DWT + PNN [64, 65] Level set + RG (region growing) [66] fully convolutional neural network (FCNNs) + Conditional random fields (CRFs) [67] |
Fig. 3 Analysis based on the modality used.
Fig. 4 Analysis based on the dataset source used.
Table 4 The advantages and disadvantages of the most used classifiers for human brain MR images are summarized
Groups of the methods |
Techniques |
Advantages |
Disadvantages |
Threshold-based |
Global and local thresholding |
Simple and computationally fast. |
Limited applicability to enhancing tumor areas. |
Region-based |
Region-growing |
Simple and capable of correctly segmenting regions that have similar properties and generating connected region. |
Partial volume effect. Noise or variation of intensity may result in holes or over-segmentation. |
Watershed |
Segments multiple regions at the same time. It produces a complete contour of the images and avoids the need for any kind of contour joining. |
Over-segmentation. |
|
Pixel-based |
Fuzzy C means |
Unsupervised. Always converges the boundaries of tumor. it determines a membership degree of data to each class. |
Long computational time, sensitivity to noise . The requirement for initialization of several initial parameters. |
Artificial neural networks |
Ability to model non-trivial distributions and non-linear dependences. self-adaptive methods. |
Gathering training samples is not straightforward and learning phase is slow. |
|
Markov random fields |
Are able to represent complex dependencies among data instances. |
Difficulty when selecting the parameters that control the strength of spatial interactions. Usually require algorithms computationally intensive. |
|
k-nn |
It is a simple and powerful. |
The choice of k affects the performance of the k-NN algorithm. severely degraded by the presence of noisy. |
|
SVM |
It minimizes the number of misclassifications. it offers a possibility to train generalizable, nonlinear classifiers in high dimensional spaces using a small training set. |
Depends on the kernel that has been used. |
|
SOM |
The advantages of SOM are simple and easy to understand and good for visualization |
The trained network may converge to some local optimum. |
|
EM |
The main advantages of this algorithm are its simplicity and speed which allows it to run on large datasets. |
Sensitive to noise and intensity in-homogeneities. |
|
Model-based |
Parametric Deformable Models |
Capable of accommodating to the variability of biological structures over time and across different individuals. |
The model may converge to wrong boundaries in case of inhomogeneities. |
Level Sets |
Topological changes are naturally possible. |
Computationally expensive . |
|
Hybrid techniques |
|
Hybrid methods aim at combining the advantages of different paradigms within a single system. Hybrid methods which combined the relative strengths from the different classifiers and applied them in a sequence in such a way that the overall accuracy was maximized. |
High computational costs. |
Discussion
Critical review
In this review, we gave an overview of the state of the art in the MRI-based medical image analysis for brain tumor studies. The focus was on segmentation methods. The first attempts in this field were made almost two decades ago, but it can be observed that in recent years, the methods are becoming mature and an increase of their use in clinical practice is expected. Threshold-based techniques offer the possibility of conducting a simple and fast segmentation when good threshold values are defined. Although with restrictions, these techniques are generally used as a first step in the segmentation process. Region-based techniques for brain tumor segmentation are mainly used as refinement step for defining a connected boundary of the tumor. Some region-based approaches such as watershed transform, have reported very accurate results in segmenting tumors, but generally these approaches are constrained to be semi-automatic. Pixel classification techniques for brain tumor segmentation are limited to clustering nevertheless they are the most frequently used for brain tumor segmentation. The unsupervised technique of FCM, which is the most popular for medical image segmentation permits the use of vague concepts in the definition of clusters, and gives highly accurate results in cases of non-homogeneous tumors. Model-based techniques have been widely used for its sensitivity in searching the boundary of brain tumors. However, as in the case of region-based methods, these models are mainly used as refinement step in brain tumor segmentation. Segmenting tumors by making use of geometric deformable models or level sets, permits the development of fully automatic and highly accurate segmentation approaches. Unfortunately, these methods are still computationally expensive. The majority of segmentation approaches operate on multi-sequence MRI data, employing classification methods using different features and taking spatial information in a local neighborhood into account. The trend is not to segment the tumor only, but also to delineate tumor sub-compartments and different healthy regions on images from standard clinical acquisition protocols. This provides the physician with a more comprehensive information on which diagnosis, tumor monitoring and therapy planning can be based. Apart from the evaluation of accuracy and robustness, an important criterion is the computation time. Eventually, it would be useful to test any new method on a standard database of brain tumor images to allow for a fair comparison against the state of the art. The MICCAI BraTS dataset would be one candidate for such a database. It is also necessary to select some relevant measures that will be used as common evaluation measures to compare various methods.
Table 5 The database used in CAD
Data base source |
Data set used |
Modalities |
Paper |
Year |
Harvard * |
30 normal, 20 abnormal |
axial T2w |
[68] |
2010 |
320 slices |
T2w, T2c |
[69] |
2014 |
|
6 normal brains, 46 abnormal brains |
axial T2w |
[70] |
2006 |
|
70, 60 abnormal, 10 normal |
axial T2w |
[71] |
2010 |
|
75 transaxial image slices (39 normal brains, 36 pathological brain) |
axial T2w |
[56] |
2015 |
|
22 normal and 44 abnormal |
axial T2w |
[72] |
2017 |
|
BRATS ** (MICCAI) |
30 patient (20 HG, 10 LG) |
T1, T2, T1C, FLAIR |
[9] |
2014 |
5 different slices of 22 high grade and 15 low-grade tumors and 20 synthetic data |
T1w, T2w, T1c, FLAIR |
[73] |
2016 |
|
10 |
T1w, T2w, T1c |
[74] |
2012 |
|
255 |
T1, T2, PDw, FLAIR, T1c |
[57] |
2011 |
|
660,000 data points from 11 cases |
T1w, T2w, T1Cw, Flair |
[35] |
2006 |
|
250 brain tumor MRI images |
NA |
[65] |
2015 |
|
BRATS 2013- 65 MR scans + BRATS 2015- 327 MR scans |
T1, T1c, T2, FLAIR |
[37] |
2015 |
|
30 glioma patients, 10 LG, 20 HG |
FLAIR, T1w, T1c, T2w |
[32] |
2012 |
|
30 patient (20 HG, 10 LG), 30 simulated subjects |
T1c, T2, FLAIR |
[75] |
2015 |
|
30 patient (20 HG, 10 LG) |
T1, T1C, T2 and Flair |
[43] |
2017 |
|
220 (HG) and 54 (LG) |
T1, T2 and FLAIR |
[42] |
2017 |
|
IBSR *** |
172 |
T1c |
[76] |
2014 |
65 images of 40 Brain web data, 25 IBSR V2.0 |
T1w |
[77] |
2010 |
|
IBSR 1.0- 20 images, IBSR 2.0- 18 images |
T1w volumetric images |
[21] |
2012 |
|
20 normal people |
T1-w |
[78] |
2015 |
|
San Raffaele Hospital, Milan |
15 patients (9 LG, 6 HG) and 6 healthy patients |
T2w FSE (Fast Spin Echo), T1w FFE(fast field echo) |
[59] |
2012 |
PSG IMSR & Hospitals, Coimbatore, Tamilnadu, India |
35 patients; 12 meningiomas and 23 gliomas |
T1w, T1c, 1H-MRSI |
[79] |
2014 |
General Hellenic Airforce Hospital, MRI Unit, Katehaki, Athens, Greece |
67 MR images |
T1w |
[80] |
2014 |
Hua–Shan Hospital in Shanghai of China. |
DS1-280 glioma (169 LG, 111 HG), DS2- 154 cases (85 LG, 69 HG) |
T1, T2 |
[81] |
2016 |
Dr.Shajis MRI & Medical Research 709 Centre Pvt.Ltd, Puthiyara, Calicut |
200 images- 164 trainig set (82 LG, 82HG), 36 testing set (18 LG, 18HG) |
T2w |
[82] |
2011 |
Brain web tumor repository Harvard medical school Pakistan Institute of Medical Sciences |
25 patients with gliomas |
T1w, T2w, PDw |
[83] |
2013 |
* http://www.med.harvard.edu/aanlib/
** https://www.med.upenn.edu/sbia/brats2017/data.html; http://www.braintumorsegmentation.org/
*** https://www.nitrc.org/projects/ibsr/
Table 6 Summary of the most common measures used to evaluate the MS lesion segmentation algorithms
Categories |
Metrics |
Symbols |
Calculation |
Description |
Overlap based metrics |
Sensitivity, Overlap fraction, Recall, True positive rate.
Specificity, True negative rate.
Accuracy
Dice similarity coefficient, F1score Positive predictive value, Precision, Reliability.
Fallout, False positive rate, False alarm ratio.
Jaccard index, tanimoto |
SEN, TPR
SPE, TNR
ACC
DSC, Dice
PPV
FALL, FPR
JI, JAC |
|
TP: True positive. FN: False negative.
TN: True negative. FP: False positive.
|
Volume based metrics |
Volume difference
Pearson’s r coefficient |
VD
PRC |
|
: Volume of ground truth. : Volume of automatic segmentation. : Means of respectives volumes. N: Number of time points. |
Pair counting based metrics |
Rand index
Adjusted rand index |
RI
ARI |
|
For each tuple ) of the volume. Sa: Automatique segmentation. Sg: Ground truth. : The number ) that are in the same class by Sa and in the same class in Sg. : The number ) that are in the different class by Sa and in the different class by Sg. The number ) that are in the same class by Sa and in the different class by Sg. The number ) that are in the different class by Sa and in the same class by Sg. |
Information theoretic based metrics |
Mutial information
Variation of information |
MI
VOI |
|
: Marginal entropy of automatique segmentation. : Marginal entropy of ground truth. : Joint entropy. |
Probabilistic metrics |
Intra-Class correlation |
ICC |
|
: Variance of differences between the segmentations. : Variance ofdifferences between the points in the segmentations. |
Spatial distance based metrics |
Hausdorf distance
Average distance |
HD(A,B)
AD(A,B)
|
|
A,B: Two finite sets. : The euclidean distance. N: Number of elements of the finite sets. |
Critical approach: Deep learning
More recently, deep learning techniques have been adopted in brain tumor segmentation studies following their success in general image analysis fields. Convolutional neural networks (CNNs), an outstanding branch of deep learning applications to visual purposes, have earned major attention in the last years. With the time, large annotated training datasets and more powerful graphics processing units (GPUs) have been created, enabling researchers to continue working in the area. Nowadays, deep CNN architectures are widely used in brain MRI for preprocessing data, detecting and segmenting lesions and segmenting tumors. CNNs take patches extracted from the images as inputs and use trainable convolutional filters and local subsampling to extract a hierarchy of increasingly complex features. CNNs are the most popular machine learning algorithm in image processing. CNNs and recurrent neural networks (RNNs) are examples of supervised machine learning algorithms, which require significant amounts of training data. Unsupervised learning algorithms have also been studied for use in medical image analysis. These include autoencoders, restricted boltzmann machines (RBMs), deep belief networks (DBNs), and generative adversarial networks (GANs) [84-86].
Clinical applicability
Although a lot of research has been done in this field over the last few years, application in the clinics is still limited. Many tools developed so far are pure research tools, which are not easy to handle for clinicians. This is probably mostly due to a lack of communication between researchers and clinicians. So far, in most commercial workstations, only very simple methods have been implemented, for example, thresholding. Recently, more researchers have tried to consider standard clinical acquisition protocols when developing their methods, instead of focusing on feasibility studies that employ pure research data as image material. This will hopefully aid in spreading the application more quickly. Another important aspect is the computation time: the real-time segmentation will be hard to achieve, but computation times which are beyond a few minutes are unacceptable in clinical routine. In order to be able to make the best possible use of automatic methods for the medical image analysis, it is essential to have image data, which have been acquired according to a well-defined protocol across different clinical sites. We expect that such standardization would aid significantly in improving the applicability and spread the use of automatic.
Table 7 Evaluation of methods
Methods used |
Paper |
Year |
Metrics (%) * |
|
|||||
DSC, Dice PPV |
SEN, TPR |
SPE, TNR |
FALL, FPR ** |
ACC |
Jaccard |
HD |
|||
GA + SVM |
[56] |
2010 |
|
92 |
100 |
|
95 |
|
|
Rule-based + Level set + SVM |
[49] |
2010 |
77 |
|
|
|
|
|
|
GA + FCM PSO + FCM |
[87] |
2010 |
|
|
|
|
75 92 |
|
|
SVM + Region growing |
[66] |
2011 |
|
95 |
100 |
|
97 |
|
|
Fuzzy clustering and deformable model |
[20] |
2011 |
|
|
|
|
|
82 |
|
GA - SVM |
[57] |
2011 |
|
|
|
|
92 |
|
|
PCA + SVM |
[88] |
2012 |
|
89 |
84 |
|
95 |
|
|
Content-based active contour |
[21] |
2012 |
|
|
|
|
87 |
|
5 |
MRF + Ant colony optimization (ACO) |
[32] |
2012 |
76 |
|
|
|
|
|
|
ANN-controlled level-set |
[58] |
2012 |
49 |
75 |
|
|
|
|
|
GA + Fuzzy ANN |
[59] |
2012 |
|
97 |
96 |
|
95 |
|
|
PCA + ANN |
[89] |
2013 |
|
|
|
|
85 |
|
|
Graph - cut |
[90] |
2013 |
84 |
87 |
83 |
|
|
74 |
|
KNN + MRF |
[40] |
2014 |
75 |
|
|
|
|
|
|
Back propagation neural network |
[69] |
2014 |
|
|
|
|
83 |
|
|
particle swarm optimization + neural network |
[80] |
2014 |
|
98 |
95 |
|
99 |
|
|
Random forest |
[39] |
2015 |
70 |
61 |
13 |
|
|
|
|
Region growing and level set evolution |
[23] |
2015 |
81 |
82 |
|
|
|
70 |
|
Competitive EM and graph cut |
[31] |
2015 |
70 |
|
|
|
|
|
|
Convolutional neural networks |
[37] |
2015 |
|
73 |
73 |
|
|
|
|
Intensity features of multimodality MRI |
[75] |
2015 |
77 |
85 |
98 |
|
|
|
|
Self-organizing map + Fuzzy K means |
[73] |
2016 |
92 |
87 |
97 |
|
|
32 |
|
Enhanced watershed segmentation |
[81] |
2016 |
|
97 |
86 |
|
96 |
|
|
binary decision trees and random forest technique |
[39] |
2016 |
67 |
|
|
|
|
|
|
Round Randomized Learning Vector Quantization |
[41] |
2016 |
|
|
|
|
86 |
|
|
Improved Fuzzy c-Means Improved Watershed |
[45] |
2016 |
89 93 |
|
|
|
|
82 89 |
|
Cell density patterns/tumor growth modeling |
[25] |
2017 |
75 |
|
|
|
|
|
|
U-Net Based Fully Convolutional Networks |
[42] |
2017 |
86 |
|
|
|
|
|
|
Watershed Technique and Self organizing Maps |
[47] |
2017 |
97 |
95 |
100 |
|
96 |
|
|
K-means + Linear SVM K-means + DWT +PCA + Linear SVM |
[72] |
2017 |
|
95 95 |
89 100 |
|
93 97 |
|
|
Deep Neural Networks |
[43] |
2017 |
84 |
84 |
88 |
|
|
|
|
Densely Connected 3D CNN |
[44] |
2018 |
79 |
|
|
|
|
|
|
Fully Convolutional Neural Networks + Conditional Random Fields |
[67] |
2018 |
82 |
83 |
|
|
|
|
|
Random Forests + Multiscale Patch Driven Active Contour |
[91] |
2018 |
89 |
85 |
90 |
|
|
|
|
Information theoretic rough sets |
[92] |
2018 |
70 |
|
|
|
|
59 |
|
* The values given are obtained by calculating the averages of the results declared by the authors.
** Despite the fact that this measure exists, it is very rarely used.
Available toolboxes
Only very few implementations of the presented methods are publicly available. This impedes a comparison of new methods to existing approaches and also hinders a large-scale evaluation with clinical data from different sites. To the best of our knowledge, at the moment, there are only three methods publicly available for download, which are dedicated to the analysis of brain tumor images:
Future challenges
The first challenge is the acquisition of MRI data. In recent years, enriched MRI protocols have been developed. It is expected that their improvement will continue, in terms of standardization and optimization as well as the unification of the acquisition protocols. The comparison of different automated MS lesions segmentation techniques, based on MRI data, faces objective difficulties. For starters, not all advanced approaches are publicly available to the world of research. In addition, many methodologies are tested on proprietary MRI databases, making their comparison unreliable. A common large-scale database for research, as well as the corresponding truth on the ground, would be a very positive addition to this burgeoning field. Future segmentation techniques for MS lesions based on brain MRI should be hybrid. Combining the most promising individual strategies of the state of the art and exploiting their combined advantages should prove very useful in further improving the segmentation performance of MS lesions. In addition, it is also expected that, in the future, efforts will focus on unsupervised techniques to avoid the costly training process. Finally, in a more generic setting, robust segmentation techniques based on deep learning can be used. Future work must be based on the concept of real time in order to engineer the techniques into the computer-assisted processing process.
Conclusions
Detecting the existence of brain tumors from MRI in a fast, accurate, and reproducible way is a challenging problem. Medical image processing is a very active and fast-growing field that has evolved into an established discipline. Brain tumor segmentation techniques have already shown great potential in detecting and analyzing tumors in clinical images and this trend will undoubtedly continue into the future. Medical image analysis needs to address real-world issues that have been outside the realm of computer vision. These issues come largely from the fact that the end systems are mostly used by the physician. The human factor is essential, since any successful solution will have to be accepted by a physician and integrated into the medical procedural work flow. This puts strong constraints on the type of applicable methods. Due to it, there has been a discrepancy between the advanced frameworks presented in computer vision and the low-level methods used by researchers working on real medical application solutions. One major goal in tumor imaging research is to accurately locate the cancer. Segmentation techniques have been applied according to the characteristics that allow distinguishing tumors from normal tissues. When tumors can be distinguished from normal tissues by their image intensity, threshold-based or region growing techniques can be employed, other tumors can be identified by their shapes, so that a model-based technique can be applied for the segmentation. Although the reported accuracy on brain tumor segmentation of the proposed automated methods is quite promising, these approaches still have not gained wide acceptance among the pathologists for every day clinical practice. One of the principal reasons might be the lack of standardized procedures. Another two reasons could be the substantial differences with the traditional specialists’ way of work, and the deficiency of the existing methods in assisting medical decision in a transparent and interpretable way. The latter two are very important for computer aided medical diagnosis where the demand for reasoning and explanation is of main priority. After reviewing so many papers, we would expect to be able to distill the perfect method for brain tumor segmentation. Deep learning is now clearly the top performers in most medical image analysis competitions. Our point of view is also to try to group the best performing methods into clinical applications, unify MRI image acquisition protocols, and use a single set of learning data like MICCAI.
References
1. A. Singh, S. Bajpai, S. Karanam, et al., Malignant brain tumor detection. Int. J. Comput. Theory Eng., 2012, 4(6): 1002.
2. K.M. Mahesh, J.A. Renjit, Evolutionary intelligence for brain tumor recognition from MRI images: A critical study and review. Evol. Intell., 2018, 11(1-2): 19-30.
3. M.S.H. Al-Tamimi, G. Sulong, Tumor brain detection through MR images: A review of literature. J. Theor. Appl. Inf. Technol., 2014, 62(2.
4. Moteur de recherche d’articles scientifiques, éduscol, le site des professionnels de l’éducation. <http://eduscol.education.fr/numerique/tout-le-numerique/veille-education-numerique/juillet-2018/moteur-de-recherche-darticles-scientifiques>. Retrieved on Dec. 9, 2018.
5. D. Mortazavi, A.Z. Kouzani, and H. Soltanian-Zadeh, Segmentation of multiple sclerosis lesions in MR images: A review. Neuroradiology, 2012, 54(4): 299-320.
6. X. Lladó, Segmentation of multiple sclerosis lesions in brain MRI: A review of automated approaches. Inf Sci, 2012, 186(1): 164-185.
7. D. García-Lorenzo, S. Francis, S. Narayanan, et al., Review of automatic segmentation methods of multiple sclerosis white matter lesions on conventional magnetic resonance imaging. Med. Image Anal., 2013, 17(1): 1-18.
8. S. Bauer, R. Wiest, L.-P. Nolte, et al., A survey of MRI-based medical image analysis for brain tumor studies. Phys. Med. Biol., 2013, 58(13): R97.
9. E.-S.A. El-Dahshan, H.M. Mohsen, K. Revett, et al., Computer-aided diagnosis of human brain tumor through MRI: A survey and a new algorithm. Expert Syst. Appl., 2014, 41(11): 5526-5545.
10. N. Singh, N. Choudhary, A survey: Brain tumor detection techniques of Computer aided diagnosis through MRI image. Int. J. Comput. Sci. Issues IJCSI, 2015, 12(6): 148.
11. A. Danelakis, T. Theoharis, and D.A. Verganelakis, Survey of automated multiple sclerosis lesion segmentation techniques on magnetic resonance imaging. Comput. Med. Imaging Graph. Off. J. Comput. Med. Imaging Soc., 2018, 70: 83-100.
12. S. Iqbal, M.U.G. Khan, T. Saba, et al., Computer-assisted brain tumor type discrimination using magnetic resonance imaging features. Biomed. Eng. Lett., 2018, 8(1): 5-28.
13. G. Mohan, M.M. Subashini, MRI based medical image analysis: Survey on brain tumor grade classification. Biomed. Signal Process. Control, 2018, 39: 139-161.
14. I. Bankman, Handbook of medical image processing and analysis. Elsevier, 2008.
15. S. Yazdani, R. Yusof, A. Karimian, et al., Image segmentation methods and applications in MRI brain images. IETE Tech. Rev., 2015, 32(6): 413-427.
16. C. Fernandez-Lozano, J.A. Seoane, M. Gestal, et al., Texture classification using feature selection and kernel-based techniques. Soft Comput., 2015, 19(9): 2469-2480.
17. P.S.S. Kumar, A study of MRI segmentation methods in automatic brain tumor detection. Int. J. Eng. Technol, 2016, 8: 609-614.
18. N. Gordillo, E. Montseny, and P. Sobrevilla, State of the art survey on MRI brain tumor segmentation. Magn. Reson. Imaging, 2013, 31(8): 1426-1438.
19. A.A. Taha, A. Hanbury, Metrics for evaluating 3D medical image segmentation: Analysis, selection, and tool. BMC Med. Imaging, 2015, 15(1): 29.
20. A. Rajendran, R. Dhanasekaran, Fuzzy clustering and deformable model for tumor segmentation on MRI brain image: A combined approach. Procedia Eng., 2012, 30: 327-333.
21. J. Sachdeva, V. Kumar, I. Gupta, et al., A novel content-based active contour model for brain tumor segmentation. Magn. Reson. Imaging, 2012, 30(5): 694-715.
22. B. Tanoori, Z. Azimifar, A. Shakibafar, et al., Brain volumetry: An active contour model-based segmentation followed by SVM-based classification. Comput. Biol. Med., 2011, 41(8): 619-632.
23. I. Zabir, S. Paul, M.A. Rayhan, et al., Automatic brain tumor detection and segmentation from multi-modal MRI images based on region growing and level set evolution. Proceedings of the 2015 IEEE International WIE Conference on Electrical and Computer Engineering (WIECON-ECE). 2015: 503-506.
24. P.C. Barman, M.S. Miah, B.C. Singh, et al., MRI image segmentation using level set method and implement an medical diagnosis system. Comput. Sci. Eng., 2011, 1(5): 1.
25. L. Pei, S.M.S. Reza, W. Li, et al., Improved brain tumor segmentation by utilizing tumor growth model in longitudinal brain MRI. Proc. SPIE-- Int. Soc. Opt. Eng., 2017, 10134.
26. K. Somasundaram, T. Kalaiselvi, Fully automatic brain extraction algorithm for axial T2-weighted magnetic resonance images. Comput. Biol. Med., 2010, 40(10): 811-822.
27. A. Aslam, E. Khan, and M.M.S. Beg, Improved edge detection algorithm for brain tumor segmentation. Procedia Comput. Sci., 2015, 58: 430-437.
28. P. Dhage, M.R. Phegade, and S.K. Shah, Watershed segmentation brain tumor detection. Proceedings of 2015 International Conference on Pervasive Computing (ICPC). 2015: 1-5.
29. D. Anithadevi, K. Perumal, Rough set and multi-thresholds based seeded region growing algorithm for image segmentation. Artificial intelligence and evolutionary computations in engineering systems. Springer, Singapore, 2018: 369-379.
30. R. Donoso, A. Veloz, and H. Allende, Modified expectation maximization algorithm for MRI segmentation. Progress in pattern recognition, image analysis, computer vision, and applications, 2010: 63-70.
31. V. Pedoia, S. Balbi, and E. Binaghi, Fully automatic brain tumor segmentation by using competitive EM and graph cut. Image analysis and processing – ICIAP, 2015: 568-578.
32. S. Yousefi, R. Azmi, and M. Zahedi, Brain tissue segmentation in MR images based on a hybrid of MRF and social algorithms. Med. Image Anal., 2012, 16(4): 840-848.
33. H. Merisaari, Gaussian mixture model-based segmentation of MR images taken from premature infant brains. J. Neurosci. Methods, 2009, 182(1): 110-122.
34. A.R. Mathew, P.B. Anto, Tumor detection and classification of MRI brain image using wavelet transform and SVM. Proceedings of the 2017 International Conference on Signal Processing and Communication (ICSPC). 2017: 75-78.
35. M. Maitra, A. Chatterjee, A Slantlet transform based intelligent system for magnetic resonance brain image classification. Biomed. Signal Process. Control, 2006, 1(4): 299-306.
36. A. Ortiz, J.M. Górriz, J. Ramírez, et al., Two fully-unsupervised methods for MR brain image segmentation using SOM-based strategies. Appl. Soft Comput., 2013, 13(5): 2668-2682.
37. Y. Pan, Brain tumor grading based on neural networks and convolutional neural networks. Proceedings of the 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). 2015: 699-702.
38. D.C. Shubhangi, P.S. Hiremath, Support vector machine (SVM) classifier for brain tumor detection. Proceedings of the International Conference on Advances in Computing, Communication and Control. New York, USA, 2009: 444-448.
39. Z. Kapás, L. Lefkovits, and L. Szilágyi, Automatic detection and segmentation of brain tumor using random forest approach. Modeling decisions for artificial intelligence, 2016: 301-312.
40. M. Havaei, P. Jodoin, and H. Larochelle, Efficient interactive brain tumor segmentation as within-brain kNN classification. Proceedings of the 2014 22nd International Conference on Pattern Recognition. 2014: 556-561.
41. S. Abdullah, Round randomized learning vector quantization for brain tumor imaging. Comput. Math. Methods Med., 2016.
42. H. Dong, G. Yang, F. Liu, et al., Automatic brain tumor detection and segmentation using U-net based fully convolutional networks. Medical image understanding and analysis, 2017: 506-517.
43. M. Havaei, Brain tumor segmentation with deep neural networks. Med. Image Anal., 2017, 35: 18-31.
44. L. Chen, Y. Wu, A.M. DSouza, et al., MRI tumor segmentation with densely connected 3D CNN. ArXiv180202427 Cs Eess, 2018.
45. C.C. Benson, V. Deepa, V.L. Lajish, et al., Brain tumor segmentation from MR brain images using improved fuzzy c-means clustering and watershed algorithm. Proceedings of the 2016 International Conference on Advances in Computing, Communications and Informatics (ICACCI). 2016: 187-192.
46. B. Singh, P. Aggarwal, Detection of brain tumor using modified mean-shift based fuzzy c-mean segmentation from MRI Images. Proceedings of the 2017 8th IEEE Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON). 2017: 536–545.
47. A. Anand, Brain tumor segmentation using watershed technique and self organizing maps. Indian J. Sci. Technol., 2017, 10(44).
48. M. Ganesh, M. Naresh, and C. Arvind, MRI brain image segmentation using enhanced adaptive fuzzy K-means algorithm. Intell. Autom. Soft Comput., 2017, 23(2): 325-330.
49. D. Yamamoto, Computer-aided detection of multiple sclerosis lesions in brain magnetic resonance images: False positive reduction scheme consisted of rule-based, level set method, and support vector machine. Comput. Med. Imaging Graph. Off. J. Comput. Med. Imaging Soc., 2010, 34(5): 404-413.
50. J.C. Fu, C.C. Chen, J.W. Chai, et al., Image segmentation by EM-based adaptive pulse coupled neural networks in brain magnetic resonance imaging. Comput. Med. Imaging Graph., 2010, 34(4): 308-320.
51. R. Ramasamy, P. Anandhakumar, Brain tissue classification of MR images using fast fourier transform based expectation-maximization Gaussian mixture model. Advances in computing and information technology. 2011: 387-398.
52. M.A. Balafar, A.R. Ramli, M.I. Saripan, et al., Medical image segmentation using fuzzy C-mean (FCM), learning vector quantization (LVQ) and user interaction. Advanced intelligent computing theories and applications. With aspects of contemporary intelligent computing techniques. 2008: 177-184.
53. H. Ali, M. Elmogy, E. El-Daydamony, et al., Multi-resolution MRI brain image segmentation based on morphological pyramid and fuzzy C-mean clustering. Arab. J. Sci. Eng., 2015, 40(11): 3173-3185.
54. B. Tanoori, Z. Azimifar, A. Shakibafar, et al., Brain volumetry: An active contour model-based segmentation followed by SVM-based classification. Comput. Biol. Med., 2011, 41(8): 619-632.
55. A. Demirhan, İ. Güler, Combining stationary wavelet transform and self-organizing maps for brain MR image segmentation. Eng. Appl. Artif. Intell., 2011, 24(2): 358-367.
56. A. Kharrat, K. Gasmi, M.B. Messaoud, et al., A hybrid approach for automatic classification of brain MRI using genetic algorithm and support vector machine. Leonardo J. Sci., 2010, 17(1): 71-82.
57. J. Sachdeva, V. Kumar, I. Gupta, et al., Multiclass brain tumor classification using GA-SVM. Proceeding of the 2011 Developments in E-systems Engineering. 2011: 182-187.
58. J. Kuwazuru, Automated detection of multiple sclerosis candidate regions in MR images: False-positive removal with use of an ANN-controlled level-set method. Radiol. Phys. Technol., 2012, 5(1): 105-113.
59. M. Sharma, S. Mukharjee, Brain tumor segmentation using hybrid genetic algorithm and artificial neural network fuzzy inference system (anfis). Int. J. Fuzzy Log. Syst., 2012, 2(4): 31-42.
60. Z. Ji, Q. Sun, Y. Xia, et al., Generalized rough fuzzy c-means algorithm for brain MR image segmentation. Comput. Methods Programs Biomed., 2012, 108(2): 644-655.
61. K. Gasmi, A. Kharrat, M.B. Messaoud, et al., Automated segmentation of brain tumor using optimal texture features and support vector machine classifier. Image Analysis and Recognition. 2012: 230-239.
62. M.F. Zarandi, M. Zarinbal, and M. Izadi, Systematic image processing for diagnosing brain tumors: A Type-II fuzzy expert system approach. Appl. Soft Comput., 2011, 11(1): 285-294.
63. P. Singh, H.S. Bhadauria, and A. Singh, Automatic brain MRI image segmentation using FCM and LSM. Infocom Technologies and Optimization Proceedings of 3rd International Conference on Reliability. 2014: 1-6.
64. N.V. Shree, T.N.R. Kumar, Identification and classification of brain tumor MRI images with feature extraction using DWT and probabilistic neural network. Brain Inform., 2018, 5(1): 23-30.
65. S.B. Gaikwad, M.S. Joshi, Brain tumor classification using principal component analysis and probabilistic neural network. Int. J. Comput. Appl., 2015, 120(3).
66. N. Zhang, S. Ruan, S. Lebonvallet, et al., Kernel feature selection to fuse multi-spectral MRI images for brain tumor segmentation. Comput. Vis. Image Underst., 2011, 115(2): 256-269.
67. X. Zhao, Y. Wu, G. Song, et al., A deep learning model integrating FCNNs and CRFs for brain tumor segmentation. Med. Image Anal., 2018, 43: 98-111.
68. S. Andrews, G. Hamarneh, and A. Saad, Fast random walker with priors using precomputation for interactive medical image segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2010. 2010: 9–16.
69. B.M. Zahran, Classification of brain tumor using neural network. Comput. Softw., 2014: 673.
70. G.-Z. Li, J. Yang, C.-Z. Ye, et al., Degree prediction of malignancy in brain glioma using support vector machines. Comput. Biol. Med., 2006, 36(3): 313-325.
71. G. De Nunzio, M. Donativi, G. Pastore, et al., Automatic segmentation and therapy follow-up of cerebral glioma in diffusion-tensor images. Proceedings of the 2010 IEEE International Conference on Computational Intelligence for Measurement Systems and Applications (CIMSA). 2010: 43-47.
72. H. Mohsen, E.-S.A. El-Dahshan, E.-S.M. El-Horbaty, et al., Intelligent methodology for brain tumors classification in magnetic resonance images. Int. J. Comput., 2017.
73. G. Vishnuvarthanan, M.P. Rajasekaran, P. Subbaraj, et al., An unsupervised learning method with a clustering approach for tumor identification and tissue segmentation in magnetic resonance brain images. Appl. Soft Comput., 2016, 38: 190-212.
74. D. Zikic, Decision forests for tissue-specific segmentation of high-grade gliomas in multi-channel MR. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2012, 2012: 369-376.
75. M. Gupta, B.V.V.S.N.P. Rao, V. Rajagopalan, et al., Volumetric segmentation of brain tumor based on intensity features of multimodality magnetic resonance imaging. Proceedings of the 2015 International Conference on Computer, Communication and Control (IC4). 2015: 1-6.
76. B.H. Menze, The multimodal brain tumor image segmentation benchmark (BRATS). IEEE Trans. Med. Imaging, 2015, 34(10): 1993-2024.
77. E.-S.A. El-Dahshan, T. Hosny, and A.-B.M. Salem, Hybrid intelligent techniques for MRI brain images classification. Digit. Signal Process., 2010, 20(2): 433-441.
78. A. Nandi, Detection of human brain tumour using MRI image segmentation and morphological operators. Proceedings of the 2015 IEEE International Conference on Computer Graphics, Vision and Information Security (CGVIS). 2015: 55-60.
79. I. Jolliffe, Principal component analysis. Wiley StatsRef: Statistics Reference Online, American Cancer Society, 2014.
80. D.S. Nachimuthu, A. Baladhandapani, Multidimensional texture characterization: On analysis for brain tumor tissues using MRS and MRI. J. Digit. Imaging, 2014, 27(4): 496-506.
81. D. Aju, R. Rajkumar, T1-T2 weighted MR image composition and cataloguing of brain tumor using regularized logistic regression. J. Teknol., 2016, 78(9): 149-159.
82. N. Marshkole, B.K. Singh, and A.S. Thoke, Texture and shape based classification of brain tumors using linear vector quantization. Int. J. Comput. Appl., 2011, 30(11): 21-23.
83. A.G. van der Kolk, J. Hendrikse, J.J. Zwanenburg, et al., Clinical applications of 7 T MRI in the brain. Eur. J. Radiol., 2013, 82(5): 708-718.
84. J. Ker, L. Wang, J. Rao, et al., Deep learning applications in medical image analysis. IEEE Access, 2018, 6: 9375-9389.
85. U. Baid, Deep learning radiomics algorithm for gliomas (DRAG) model: A novel approach using 3D UNET based deep convolutional neural network for predicting survival in gliomas. Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. 2019: 369-379.
86. P. Mlynarski, H. Delingette, A. Criminisi, et al., 3D convolutional neural networks for tumor segmentation using long-range 2D context. Comput. Med. Imaging Graph, 2019.
87. N.N. Gopal, M. Karnan, Diagnose brain tumor through MRI using image processing clustering algorithms such as fuzzy C means along with intelligent optimization techniques. Proceedings of the 2010 IEEE International Conference on Computational Intelligence and Computing Research. 2010: 1-4.
88. F.G. Zöllner, K.E. Emblem, and L.R. Schad, SVM-based glioma grading: Optimization by feature reduction analysis. Z. Für Med. Phys., 2012, 22(3): 205-214.
89. J. Sachdeva, V. Kumar, I. Gupta, et al., Segmentation, feature extraction, and multiclass brain tumor classification. J. Digit. Imaging, 2013, 26(6): 1141-1150.
90. J. Jiang, Y. Wu, M. Huang, et al., 3D brain tumor segmentation in multimodal MR images based on learning population- and patient-specific feature sets. Comput. Med. Imaging Graph., 2013, 37(7): 512-521.
91. C. Ma, G. Luo, and K. Wang, Concatenated and connected random forests with multiscale patch driven active contour model for automated brain tumor segmentation of MR images. IEEE Trans. Med. Imaging, 2018, 37(8): 1943-1954.
92. K.Y. Lim, R. Mandava, A multi-phase semi-automatic approach for multisequence brain tumor image segmentation. Expert Syst. Appl., 2018, 112: 288-300.
93. M. Prastawa, E. Bullitt, and G. Gerig, Simulation of brain tumors in MR images for evaluation of segmentation efficacy. Med. Image Anal., 2009, 13(2): 297-311.
94. N. Porz, Multi-modal glioblastoma segmentation: man versus machine. PloS One, 2014, 9(5): e96873.
95. D. Kwon, R.T. Shinohara, H. Akbari, et al., Combining generative models for multifocal glioma segmentation and registration. Medical Image Computing and Computer-Assisted Intervention - MICCAI 2014. 2014: 763-770.
96. A. Gooya, GLISTR: Glioma image segmentation and registration. IEEE Trans. Med. Imaging, 2012, 31(10): 1941-1954.
Copyright© Messaoud Hameurlaine, Abdelouahab Moussaoui. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.