A Review on Brain Tumor MRI Image Segmentation

Anurag Singh1 and Prachi Chauhan2

B.N.College of Engineering and Technology
chauhananurag95@gmail.com
Department of Computer Science and Engineering, B.N.College of
Engineering and Technology Lucknow,India
Email,id : prachibncet@gmail.com

Abstract. recognized: benign and malignant. Malignant

tumors are characterized by rapid growth and
A brain tumor represents one of the most perilous significant danger, whereas benign tumors exhibit
medical conditions. Consequently, swift and precise slower growth rates and pose comparatively lesser
detection of brain tumors is imperative. Employing risks.
automated techniques for identifying brain tumor
tissue aids in containing the proliferation of tumor Malignant tumors, primarily originating within the
cells within the system. MRI technology is utilized brain itself, swiftly infiltrate surrounding brain
to capture high-resolution images of both the tissue, exacerbating the patient's condition. This
individual's anatomy and cancerous tissues, offering category represents the rarest form of brain tumor.
superior image quality compared to other imaging Conversely, secondary tumors typically arise in
modalities. However, pinpointing brain tumors in other parts of the body, such as the lungs or breasts,
MRI images poses challenges due to the intricate and metastasize to the brain, constituting a more
nature of the brain. Processing MRI images and prevalent form of brain tumor known as metastatic
segmenting brain tumors therein constitute complex brain tumor.
tasks. Automated segmentation in image processing MRI (Magnetic Resonance Imaging) stands as the
serves as a method for partitioning an image into foremost diagnostic modality employed for the
distinct regions. This process stands as a critical and assessment and management of brain tumors.
indispensable phase in the analysis of brain tumor Offering a non-invasive approach, MRI enables
magnetic resonance imaging. Moreover, the detailed examination of brain tissues. Symptoms
utilization of advanced image processing techniques associated with tumor growth may manifest as
facilitates the enhancement of MRI images, migraines, recurring headaches, or vision
improving their clarity and aiding in the impairment. MRI equipment produces high-
identification of subtle abnormalities associated contrast, high-resolution images, facilitating precise
with brain tumors. This enhanced image quality is localization and characterization of tumors. Its
particularly crucial for detecting small or deeply superior imaging capabilities provide clinicians
situated tumors that may otherwise go unnoticed. with detailed views of tumors from various angles
and orientations, facilitating comprehensive tumor
Keywords: Brain tumor , Image segmentation, assessment and treatment planning.
MRI, pre-processing, feature extraction The most common application of MRI is in
diagnosing medical conditions by providing
Introduction comprehensive information through the capture of
radio signal energy pulses to create detailed brain
The brain constitutes a vital component of the images. Subsequently, medical professionals
human anatomy responsible for regulating the analyze these images to detect any abnormalities
nervous system and overseeing various essential within the organ. Diagnosing brain tumors poses
functions such as sensation, respiration, significant challenges, particularly in the early
communication, cognitive abilities, consciousness, stages, due to the diverse shapes, appearances, and
and unconsciousness. Within the realm of brain sizes of tumor cells, as well as the presence of low-
disorders, two distinct types of tumors are contrast and overlapping structures in the images.
 location, type, size, and characteristics of brain
Compared to X-ray and CT scan imaging, MRI is a tumors. Therefore, MRI interpretation becomes the
more expensive option. However, its superiority lies preferred method for
in its ability to provide precise details regarding the
Tau Inversion Recovery), and FLAIR (Fluid-
diagnostic purposes and subsequent surgical Attenuated Inversion Recovery), each offering
interventions or monitoring treatment outcomes. unique insights into the nature and composition of

brain tissues and abnormalities

Various types of MRI sequences are utilized,
including T1-weighted, T2-weighted, STIR (Short
utilized to identify structured and global features
Image Segmentation within the segmented regions. Finally, different
machine learning models, such as classification or
Image segmentation represents one of the primary clustering algorithms, are applied to distinguish
challenges in image processing, involving the between affected and unaffected areas of the brain,
partitioning of an image into distinct regions while with the output images presented to medical
preserving the inherent characteristics of each professionals for diagnosis and treatment planning.
segment, considering the overall similarity among
all segments. Estimating the extent of a tumor often Materials and Methods
relies on segmentation techniques, which naturally
delineate tumor boundaries by isolating the tumor Research on brain tumor segmentation using MRI
region. However, segmenting brain tumors is a scans has been extensively pursued by researchers,
complex task due to the irregular properties of and here we provide a summary of some notable
tumors, such as size, shape, and location, contributions.
compounded by variations in cell growth over time
and image quality. A.S. Swakshar [1] proposed a methodology
employing Artificial Neural Network for tumor
The process of brain tumor segmentation typically phase categorization. This approach involves stages
involves multiple steps. Firstly, images are acquired such as image pre-processing, segmentation, feature
through MRI scanning, followed by pre-processing extraction, and SVM classification. In the pre-
techniques such as image digitization, noise processing stage, various contrast enhancement
reduction, and image enhancement. Various techniques, including adjusted contrast, adaptive
methods are then employed for segmenting and thresholding, and histogram imaging, along with the
dividing the image into distinct regions of interest. utilization of wiener and median filter techniques,
Subsequently, features extraction techniques are are applied. The TKFCM algorithm, combining K-
means and Fuzzy c-means schemes with certain automated method utilizing AI to identify and
modifications, is employed for segmentation. segment abnormal tissue cells, such as tumors, by
analyzing MRI images. This method involves
K Uma Rani [2] presents a method employing extracting entropy, energy homogeneity, and
morphology-based pre-processing, utilizing a contrast from MRI images, which are then input
double thresholding approach to separate skull into an AI system employing a neuro-fuzzy
images from MRI brain images. This study classifier to differentiate abnormal images from
compares two techniques devised for tumor normal ones. Abnormal tissues like tumors are
detection, one based on the Level set approach depicted using the region growing method.
utilizing non-parametric fuzzy models for active
contour segmentation, and the other employing the Xiao et al [8] propose an approach to estimate
K-means segmentation algorithm. features based on the correlation between brain and
side ventricular bend with brain tumor extracted
features applied to MRI images. This approach
M.A. Ansari [3] proposed a novel method for brain addresses irregularities and unnecessary
tumor detection incorporating Histogram information in the surrounding image with the use
Normalization and the adoption of K-means of the side ventricular bend for feature extraction.
Segmentation. The initial image is pre-processed to Unsupervised segmentation methods are employed
denoise redundant signals from MRI scans using for estimating LaV bend features in brain tumor
filters such as Median, Adaptive, Averaging, and segmentation.
various Un-sharp masking and Gaussian filters.
Jafari and Shafaghi [12] present a hybrid approach
Machhale Ketan et al. [6] introduced an intelligent for brain tumor tissue identification in MRI scans
classification system to differentiate normal and based on genetic algorithm and SVM. The
irregular MRI brain scans, comprising three phases: methodology involves pre-processing stages such as
image pre-processing, feature extraction, and noise reduction and contrast enhancement,
classification. The Median Filter is utilized for MRI segmentation, head stripping via morphological
scan denoising, and Skull Masking is employed to operations, and feature selection and extraction.
separate non-brain tissues from MRI brain images. Feature selection encompasses static features,
Various machine learning techniques such as SVM, Fourier and wavelet transforms, histogram, and a
KNN, and Hybrid Classifier are adopted in the combination thereof, achieved through genetic
classification phase, with a comparative study algorithms. Finally, selected features are input into
conducted among them. the SVM classifier for normal and abnormal
detection
M.Y. Bhanumurthy et al. [7] discuss a fully
 Table 1. Evaluation of diverse method

No Author Year Method use Limitation Accuracy

1 F.Hossain, Md.A. 2017 TKFCM Seg- Trouble in 97.44%
Rafiq P.S. mentation , SVM choosing ideal
Classification and features to
ANN classi- recognize
fication distinctive classes
is time con-
suming
2 Mukambika and 2017 Level set meth- Potential of 94.12%
K Uma Rani od, k-Means misiden-
Segmentation and tification of what
SVM classi- fier should be
classified
3 G. Singh and 2016 K-Means Seg- It doesn't function 91.49
M.A. Ansari [4] mentation and admirably with
SVM &Naïve clus- ters (in the 87.23%
Bayes classifica- original input
tion data) of vari- ous
size and Differ-
ent density
4 V. Kapur and 2015 SVM & SVM- At the point when 98.00%
L. Kosta [7] KNN classifica- there is an adjust-
tion ment in the
dataset, a fresh
preparing da-
taset is required
5 M.Y. Bhanu- 2014 Neuro-fuzzy Region growing 95.65%
murthy and K. classifier, Region method is very
Anne [12] growing method cost- ly in terms
of calcu- lation of
time and memory.
6 Xiao et al[9] 2013 K nearest neigh- They incorrectly 100%
bors (KNN) & allot, a non CSF
conventional pixel to the
(FCM) cluster.

7 (Reddy et al) 2019 K-NN The result have 97.3%

high Accuracy
but CT image low
Accuracy
8 (Garg et,al) 2021 Hybrid Ensemble Very costly 97.305%
classi-fier
Conclusion & Future Work
The quest for improved brain tumor detection
Brain tumor detection involves the utilization of methods is propelled by the challenges inherent in
thresholding techniques and median filtering to accurately identifying tumors amidst the complexity
remove noise from MRI images. Training is of MRI images. With meticulous attention to
conducted using an MRI database to extract accuracy and reliability, this research endeavor
features for normal, abnormal, and various types of seeks to push the boundaries of existing techniques
brain tumors. Feature extraction employs diverse and usher in a new era of enhanced tumor detection
methodologies. Tumor detection during the testing algorithms. The evaluation of these methods
phase relies on trained classifiers. This paper through computational simulations serves as a
presents a comprehensive survey of classification critical step in assessing their efficacy. The analysis
techniques employed for MRI brain images, with a of tumor presence and characteristics demands a
comparative analysis conducted across a variety of thorough and sensitive approach, where precision is
methods. Upon evaluating the recognized paramount. Therefore, the development of
techniques, it becomes apparent that certain methodologies that shed light on novel strategies
approaches can efficiently detect tumors and yield for refining image segmentation techniques
precise results. This endeavor aims to develop new becomes imperative. In the pursuit of more
algorithms for brain tumor detection to enhance effective tumor detection algorithms, this research
accuracy and reliability in future applications. not only addresses current shortcomings but also
Computational experiments are conducted to assess anticipates future advancements. By illuminating
the effectiveness of this technique. Given the new perspectives and pathways, it endeavors to
complexity and sensitivity of tumor analysis, the elevate the standards of brain tumor diagnosis and
importance of precision and reliability cannot be treatment planning. Ultimately, the ultimate goal is
overstated. Thus, there is a pressing need for to improve patient outcomes by providing accurate
detailed. methodologies that illuminate new and reliable methods for detecting and
avenues for improving image segmentation characterizing brain tumors.
techniques in this domain.

References 6, ISBN: 978-1-4673-6984-8

1. A.S. Swakshar, Md. F.Hossain R. Ahmmed
Md.A. Rafiq, “categorization of Tumors 4. K. Kamnitsas et al., “Efficient multi-scale
and It stage in Brain MRI Using SVM and 3d with CNN fully connected crf for
ANN”, International Confer- ence on accurate brain lesion segmentation,” Med
Electrical, Computer and Communication Image Anal, vol. 36, pp. 61–78, 2016.
Engineering (ECCE), 2017 5. K. Machhale, H.B.Nandpuru,V. Kapur, L.
pp. 229 - 234 , ISBN: 978-1-5090-5627-9. Kosta, “MRI Brain Cancer Classifica- tion
Using Hybrid Classifier (SVM-KNN)”,
2. K Uma Rani P.S. Mukambika, International Conference on Indus- trial
“Segmentation and categorization of MRI Instrumentation and Control (ICIC), 2015,
Brain Tumor”, International Research pp. 60 -65, ISBN: 978-1-4799-
Journal of Engineering and Technology 6. M.Y. Bhanumurthy, K. Anne, “An
(IRJET), Vol.4, Issue 7, 2017, pp. 683 – Automated Detection and segmentation of
688, ISSN: 2395-0056 Tumor in Brain using Artificial
Intelligence”, IEEE International
3. M.A. Ansari, G. Singh “Efficient Detection Conference on Computational Intelligence
of Brain Tumor from MRI Using K- Means and Computing Research, 2014, pp. 1-6,
Segmentation and Normalized Histogram”, ISBN: 978- 1-4799-3975-6
1st India International Confer- ence on
Information Processing(IICIP), 2016, pp. 1- 7. Kai xiao,A.Lei Liang, Hai Bing Guan,
Aboul Ella Hassanien ,”Extractssion and
 Application of Deformation Based Feature A.ShenbagaRajan (2014) Feature
in Medical Images”,ELSEVIER Neu- ro Extraction of Brain Tumor Using MRI,
computing 2013 International Journal of Innovative
8. Mehdi Jafari & Reza Shafaghi,”A mixture Research in Science, Engineering and
Approach for Automatic Tumor De- tection Technology , pp. 1-6
of Brain MRI Using SVM And Genetic [10] Y. Zhang, Z. Dong, and S. Wang
Algorithm Global Journal Engineering and (2011) A hybrid method for MRI brain
Technology(ISSN:2332-2441) 2012 image classification, Expert Systems with
S. Satheesh, R. Kumar, K. Prasad, and K. Applications , vol. 38, no. 8, pp. 10049 10
Reddy (2011) Skull removal of noisy 053
magnetic resonance brain images using 11 Ketan Machhale, Hari Babu Nandpuru
contourlet transform and morphological and Vivek Kapur (2015) MRI brain cancer
operations, in Computer Science and classification using hybrid classifier (SVM-
Network Technology (ICCSNT), 2011 KNN), 2015 International Conference on,
International Conference on , vol. 4, pp. Industrial Instrumentation and Control
26272631 (ICIC) , pp. 1-5
9. J. Chiverton, K.Wells, B. Podda and D.
Johnson (2007) Statistical morphological
skull stripping of adult and infant MRI data,
Computers in Biology and Medicine , vol.
37, no. 3, pp. 342 357
10. S. SivaSankari, M. Sindhu and