279
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
SELF - ACTIVATED SEGMENTATION PRACTICES
OF BRAIN TUMEFACTION IN MR SCAN IMAGES: A
STUDY
B. Perumal
Department of Electronics and Communication Engineering, Kalasalingam Academy of
Research and Education, Krishnankoil, Virudhunagar (Dt), (India).
E-mail: palanimet@gmail.com
ORCID: https://orcid.org/0000-0003-4408-9396
R. Sindhiya Devi
Department of Electronics and Communication Engineering, Kalasalingam Academy of
Research and Education, Krishnankoil, Virudhunagar (Dt), (India).
E-mail: sindhiyadevi14@gmail.com
ORCID: https://orcid.org/0000-0002-7529-6438
M. Pallikonda Rajasekaran
Department of Electronics and Communication Engineering, Kalasalingam Academy of
Research and Education, Krishnankoil, Virudhunagar (Dt), (India).
E-mail: mpraja80@gmail.com
ORCID: https://orcid.org/0000-0001-6942-4512
Recepción:
11/11/2019
Aceptación:
26/10/2020
Publicación:
30/11/2021
Citación sugerida:
Perumal, B., Devi, R. S., y Rajasekaran, M. P. (2021). Self – activated segmentation practices of brain
tumefaction in mr scan images: a study. 3C Tecnología. Glosas de innovación aplicadas a la pyme, Edición
Especial, (noviembre, 2021), 279-291. https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
280
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ABSTRACT
To segment a tumefaction of brain images obtained from any of the imaging modalities is
a lofty goal owing to the varied shape, locality and measure of tumor. This segmentation
process can be done manually by a Doctor or otherwise can be done automatically using
computer aided diagnosis. Self – Activated segmentation of brain tumor is nothing but the
separation of the tissues that are not related with the tissues within the brain case akin to
the regions with myelinated nerve bers without dendrites, portions of nerve bers with
dendrites and the cerebrum area along with the cerebrospinal uid (CSF). The various
imaging modalities are the scans from the radioscopy with emitting positrons (PET), multiple
X – rays (CT) and through Magnetic Resonance (MRI). In this paper, an overview of recent
automatic brain tumor segmentation techniques of MRI and the advantages of multimodal
imaging techniques has been explained. The segmentation techniques such as thresholding,
edge based, morphology based, watershed, k means and markov random method are the
conventional tactics of segmentation that are addressed. Also, the advanced segmentation
methods such as region growing, genetic method, fuzzy clustering, deformation, atlas
method and articial neural network are also discussed. Moreover, the hybrid methods that
have dierent combination of genetic algorithm, articial neural networks and SVM are
also considered. Among all the methods, the hybrid methods are found to be better as they
provide the beneciary factors of every method involved. But one should be aware about
the algorithm’s robustness and accuracy.
KEYWORDS
Tumor, Segmentation, Automatic, Multimodal Imaging, MRI Images.
281
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1. INTRODUCTION
Tumor arises when the cells of certain portions of our body starts growing abnormally.
Detecting tumor in the early stage is of the greatest signicance for the survival of life. Also,
the size and shape of the brain tumor decides the treatment modality. Over 120 dierent
types of brain tumor are in existence and are possibly graded as primordial and metastatic
brain tumor. The primary tumors do not spread to another part of the body and stay within
the brain. Statistically, it is found that primary tumor is found to be developed more in older
adults and children. The meta-static tumor taints the extant organs of the body from the
region of its origination. It is more prevalent in fully – grown people in comparison with
ospring.
Based on the characteristics, tumors can be categorized as benignant and virulent. The
benignant tumors are leisurely developing and are destructive to a lesser extent. The
malignant tumor is rapid growing and life threatening.
The ways and means of dissociating an image into multitudinal portions are said to be
segmentation. And it is seen that the pixels within the region have same characteristics.
During the preprocessing stage like segmentation separation of dierent tumor tissues
from normal tissues is done. In practical life, segmentation of brain tumor is performed
laboriously. But this laborious technique of splitting the tumor yields extended time span
and may sometimes create impreciseness in its results. With the intension of doing a favor
for medicos in the diagnosis and treatment of tumor the research in automatic segmentation
techniques of brain tumor is getting more important.
2. MATERIALS AND METHODS
2.1. COVENTIONAL SEGMENTATION TECHNIQUES
MRI is a noninvasive diagnosing imaging modality. It aords certain benevolent lineaments
like multi – planar capabilities. Separation methodologies are abundant in the assessment
surveys. A few among the prevailing conventional tactics for segmenting brain tumor are
addressed henceforth.
282
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.1.1. THRESHOLDING METHOD
Thresholding is one of the segmentation techniques which correlate the brightness of the
pixels with the thresholds which may be one or many. The local and global thresholding
(Lin et al., 2012) are their prominent categories.
We can distinguish the tumor region from its backdrop by considering the selected threshold
and we can utilize this bifurcation for segmentation. The object portion with tumor is
allocated the binary value 1 so that those picture elements satisfy the selected threshold
value. Others that are insucient when compared with the threshold value are allocated
with the binary value 0 and they are the background pixels (Khan & Ravi, 2013).
If an image possesses homogeneity in its intensity levels, then splitting up with the global
thresholding paves a better way. Over segmentation and under segmentation are possible
with thresholding means of separation, which may be a major drawback. The global
thresholding may cause brighter and darker patches on the grounds of inhomogeneous
intensity levels.
2.1.2. EDGE – BASED METHOD
Edge based approaches are many in number that includes Canny, Sobel and Prewit. Canny
method (1987) is obtained by simply adding up some alteration in the Sobel method. Canny
edge detector implements Gaussian in its criteria so that the consequence of noise could be
reduced. It indulges enhanced sharp edges in comparison with other methodologies.
In the segmentation of medicating imagery, we have an alternative which is the Chan and
Vese Active contour model (2001). We can call it in other words as Active contour model.
The interior contours and the undened boundary objects are competently recognized by
this methodology.
The steepness of edge of the image is detected by the snake model, which is a customary
edge indicator. The images with inhomogeneous intensity that need the betterment results
use this model. Generally, the edge – based segmentation approach is uncomplicated. The
edge-based segmentation produces open contour and also the edges are sensitive to the
threshold values.
283
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2.1.3. MORPHOLOGICAL BASED METHOD
The process in which the anatomical features of the image are employed is the morphology-
based methodology. The utilization of this method is to extract the data from the image
centered on their shape presentation. The most basic morphological operations used are
erosion and dilatation. They are contrast to each other. It means that the resultant pixels of
these processes are least possible and the uttermost values of the total pixels that appears
nearby the input picture element. In other words, we can say that, in a binary image, if any
pixel is allotted the value 1, then the dilation makes the resultant to be ‘1’ and it is ‘0’ in
case of erosion.
The manipulation of the erosion algorithm is for shrinking while dilation is for developing
the image size. The morphological based technique has the tendency to segment tumor
even in images with lesser intensity. This process incorporates some steps that include
image enrichment, generating new samples, extrication of color plane and morphological
methodology to get the Region of Interest (RoI) from the image (Sudharania, Sarma,
& Prasad, 2016). These are employed in order to eliminate picture elements of reduced
frequency and border portions.
2.1.4. WATERSHED METHOD
The working of the watershed algorithm is similar to the action of water on the landscape.
We know that the partitioned landscape is of separated portions by dams and reservoirs.
The water ows from various troughs at a centered region where the dam is present. The
ow of water stops at once when the water touches the tip of the landscape. Therefore,
every portion of landscape has a dam and resembles the watershed technique. So, an
exhaustive silhouette is formed which needs no jointure.
Over – segmentation is the principal drawback of this method. In order to subjugate this
connement, images should be processed both at the pre – stage and the post – stage of
segmentation to get satisfactory results. Pandav (2014), in her paper, propose Marker –
Controlled Watershed Segmentation which utilize oods and image gradient initiated from
the notches instead of localized least for its dissection which brings about superior outcome
284
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
from heterogeneous zones simultaneously. It generates the entire lineament of the segments
and hence it needs no jointure which is an added advantage.
2.1.5. K – MEANS CLUSTERING
The clustering technique which is the simplest and the merest mean of segmenting image is
the K-means clustering algorithm (Kamble & Rathod, 2015). In this process of clustering,
k bands are initialized rst and then for each band, we develop a barycenter or centroid
in a random manner. The centroid values are changed depending on the values of the
neighborhood bands until the stability is reached. The new centroid is assigned in such a
way that it is taken from the mean of every object in every barycenter.
2.1.6. MARKOV RANDOM FIELD METHOD
The model which converts the structural data into the procedure of crowding is the Markov
random eld (MRF) model. In this model, the overhanging of split – up and the impact of
noise are diminished. Hence, it is a user – friendly technique in segmenting medical images.
A novel method that utilizes the conjoint spatial features to extricate the conguration with
Gabor decomposition, for the fragmentation of tumor from the backdrop is explored in
Zhang, Brady, and Smith (2001). After this, the outcomes were additionally puried with this
methodology as it categorizes every sub – class of the image. But it is a perplexing procedure.
Despite that, it is an eectual method to work with images having inhomogeneous intensity.
2.2. ADVANCED SEGMENTATION APPROACHES
There are some advanced methods of segmentation. Some of those advanced techniques
are discussed in the following section.
2.2.1. REGION GROWING METHOD
The process which involves the grouping of regions according to the similarity of the pixels
until each pixel is allocated to a group is called the Region growing method. We start this
operation by electing a seed point. The seed selection is done either automatic or manual.
Similar neighbors of the chosen seed are added to the region. We repeat this growing
of regions each seed is allocated to a region. The region growing when used along with
285
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
fuzzy logic to get a knowledge – based region growing is always protable (Lin et al., 2012).
Here, dierent objects are grouped with the geographical data and equivalence from the
multimodal images, fuzzy logic and homogeneity of the images are employed in nding out
the initial seed.
The region growing methodology on the basis of its image texture is explained in Charutha
and Jayashree (2014). We can consider the threshold value of both the intensity and texture
for the extraction of brain tumor in medical images. As in every method, it also has its own
merits and drawbacks. Its merit is that the similarity of pixels could be easily measured and
tracked, while it has a demerit that it is more complex in selecting initial seed point and is
susceptible to noise.
2.2.2. GENETIC METHOD
The method used for solving both the guarded and unguarded problems for nding
optimization and it is a selection process gleaned from the natural transformation (Fan,
Jiang, & Evans, 2002). We use chromatins in the genetic algorithm with the purpose of
expressing the population. We can rejuvenate the population of individuals repetitiously
by the usage of terms like alteration and crossover with a selection operator. The function
that is meant for assessing population in the genetic method for the optimization is called
tness function. We use this genetically dependable algorithm for eective optimization of
the segmenting portions of MRI images (Chandra & Rao, 2016).
In the methodology described in Chandra and Rao (2016), they used K – means to get the
bands for the population at the initial stage. The barycenter or the centroid is estimated by
a specic tness function. The exchange of healthy chromosomes with the weaker ones is
done by using crossover and mutation. If the problem is of more diculty, then the genetic
algorithm provides a better solution.
2.2.3. FUZZY CLUSTERING
Fuzzy clustering is one of the advanced clustering techniques in which the grouping is
mainly based on the membership function (Oliveira & Pedrycz, 2007). The membership
function in the fuzzy logic owns a value within the bounds of 0 to 1. This value of the factor
shows the homogeneity amongst the picture element and its barycenter. The process of fuzzy
286
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
clustering allots a membership function for each and every picture element determined by
its characteristics. The pixel is said to be located near the barycenter if it possesses the
value 1. The vicinage fascination with reference to the locale and the pixel’s connection to
adjacency pixels surge the potential of Fuzzy C Means (Selvakumar, Lakshmi, & Arivoli,
2012).
The extent of the most appropriate value along with the scope of connection between the
pixels are established by making use of the combined Genetic algorithm ad Particle Swarm
algorithm. Normally, there are two main stages in the system that includes the analysis and
evocation. The main disadvantage of the fuzzy based clustering is that the non – usage of
geometrical data for segmenting tumors (Ain, Jaar, & SunChoi, 2014).
2.2.4. DEFORMATION MODEL
The deformation model better suits for locally changing environment since it befalls in the
variable image object. This model resembles a closed curve in 2D while a closed surface in
3D images. This model is of two major types. They are the parametric deformation where
the model is snakes which are the active outliers and the geometric deformation.
A method that uses Chan Vese Active contour model also called as Active contour model is
specied in Chan and Vese (2001). It recognizes the innards and the other objects without
gradient undened boundaries. A traditional snake model edge which depends on the
gradient value of the image is further improved by Chan Vese model. This model gives
better yield for images with similar inhomogeneity.
2.2.5. ATLAS METHOD
An atlas is nothing but a reference image that is previously segmented by an expert for the
tumor extraction in unobserved images. In the atlas dependable method, we rst register
both the atlas model and the goal image to be segmented. Then we have to map the model
with the target image for eective segmentation. A novel technique without any mesh to
design the atlas for well – conditioned brain imagery along with a reformed atlas for a
diseased or morbid brain image has been projected in Bauer et al. (2010).
287
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
In order to embellish its accuracy, the tumor’s locale along with the reason that provokes the
tumor growth should be equipped with the atlas. In case of segmentation in multiple regions,
(Al-Shaikhli, Yang, & Rosenhahn, 2014) used an adapted multileveled set of articulation
with the atlas data of the statistical provost graph. Diaz and Boulanger (2015) used Total
Lagrangian Explicit Dynamic (TLED) method to tackle a massive disguration devoid of
giving up. The atlases for the deformed brain image employ the genuine pattern of the
tumor rather than employing erratic pattern (Dia & Boulanger, 2015). The initial seed point
need not be initialized in this technique which in turn multiplies the vibrancy. Also, the
processing time is diminished because of the parallel processing system. Its performance
shows the merit while its demerit is that more time span will be taken for the construction
of atlas models.
2.2.6. ARTIFICIAL NEURAL NETWORK
The main idea of the articial neural network (ANN) classier is machine learning. They
are the brain – stimulated systems in such a way that it resembles the same way by which the
human beings learn. It comprises various nodes including the insertion node, transitional
nodes and the unexposed nodes.
We need to train the machine to determine the worth of the parameter factors. The region
of interest (RoI) is systematized by employing the modied probabilistic neural network
(PNN) with linear vector quantization (LVQ) modeling process as shown in Song et al.
(2007). Every RoI is designated with a combo of countenance and a counterweight which
are meant for deriving an interconnected structure with reference to LVQ. It also has its
own restriction that the size of the network holds the diculty criteria. It means that the
complexity increases with the loftiness in its size, since it requires surplus tutoring.
2.2.7. HYBRID METHOD
The best characteristics of dierent methodologies are combined together to get the
hybridized method. Those hybrid methods are faster with greater accuracy. The pulse-
coded neural network (PCNN) if enhances, it will improve its reality feature of simulation.
The primary ring involves the selection of neurons as its seed points for region growing
and the secondary ring grows the region by summing up the seeds with feed forward back
288
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
neural network (FFBNN). The feed forward process is done to reach the uniformity in its
input. Stationary wavelet transform (SWT) is exercised for getting the sub images with the
data of multiple resolutions (Ortiz et al., 2013).
Spatial ltering along with LVQ is done with the purpose of extracting numerical features
and regulating of the resultant image. Here, the segmentation is performed with Content
– Based active contour model (Sachdeva et al., 2012). Genetic Algorithm helps in reducing
parameters with higher dimensions. While comparing the hybrid process of GA and SVM
with GA and ANN, the prior is good in speed while the latter is better in its accuracy. The
hybrid method slips in the step that it possesses higher computative expenses.
3. DISCUSSION
Segmentation methods are becoming more growing day by day and gaining importance
in clinical research. In spite of the numerous contradictions in the process of segmentation
over MRI imagery, they are blooming nowadays. One such contradiction is the need of
transpicuousness. Another question is about its accountability to understand the need.
Some other conicts include laboring antiquity and movement antiquity. The impact of
sectional measure and the inhomogeneous intensity are also the pervasive problems in the
splitting up of the brain tumor. The deviation in its structure such as shape, size and the
locale of tumor also aects segmentation results. The value of Signal to Noise Ratio should
be low so that the resolution will be elevated. So, the relics must be eliminated by using the
appliances with higher resolution and worthier ltrating which causes no loss in its anatomy.
The tumor has its impact not only in the aected area but also the surrounding portions.
This shows the fact that the segmentation of every aected region is necessary. Hence,
further awareness should be paid on the robustness and accurateness of the algorithm. The
algorithm’s eectiveness can be validated by utilizing ground truth images.
REFERENCES
Ain, Q., Jaar, A., & SunChoi, T. (2014). Fuzzy anisotropic diusion based segmentation
and texture based ensemble classication of brain tumor. Applied Soft Computing, 21,
330–340. https://doi.org/10.1016/j.asoc.2014.03.019
289
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Al-Shaikhli, S., Yang, M., & Rosenhahn, B. (2014). Multi region labelling and
segmentation using a graph topology prior and atlas information in brain images.
Computerized Medical Imaging And Graphics, 38(8), 725–734. https://doi.org/10.1016/j.
compmedimag.2014.06.008
Bauer, S., Seiler, C., Bardyn, T., Buechler, P., & Reyes, M. (2010). Atlas-based
segmentation of brain tumor images using a Markov Random Field-based tumor
growth model and non-rigid registration. In Annual International Conference of the
IEEE Engineering in Medicine and Biology, 4080-4083. https://doi.org/10.1109/
IEMBS.2010.5627302
Canny, J. (1987). A computational approach to edge detection. Readings In Computer Vision,
184-203. https://doi.org/10.1109/TPAMI.1986.4767851
Chan, T. F., & Vese, L. A. (2001). Active contours without edges. IEEE Transactions On
Image Processing, 10(2), 266-277. https://doi.org/10.1109/83.902291
Chandra, R., & Rao, K. R. H. (2016). Tumor detection in brain using genetic algorithm.
Procedia Computer Science, 79, 449-457. https://doi.org/10.1016/j.procs.2016.03.058
Charutha, S., & Jayashree, M. J. (2014). An ecient brain tumor detection by integrating
modied texture based region growing and cellular automata edge detection.
International Conference On Control, Instrumentation, Communication And Computational
Technologies (ICCICCT), 1193-1199. https://doi.org/10.1109/iccicct.2014.6993142
Diaz, I., & Boulanger, P. (2015). Atlas to patient registration with brain tumor based
on a mesh – free method. In 37th Annual International Conference Of The IEEE
Engineering In Medicine And Biological Society, 2924-2927. https://doi.org/10.1109/
EMBC.2015.7319004
Fan, Y., Jiang, T., & Evans, D. J. (2002). Volumetric segmentation of brain images using
parallel genetic algorithms. IEEE Transactions On Medical Imaging, 21(8), 904-909.
https://doi.org/10.1109/TMI.2002.803126
Kamble, S. T., & Rathod, M. R. (2015). Brain tumor segmentation using k-means
clustering algorithm. International Journal of Current Engineering and Technology, 5(3),
290
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1521–1524. https://inpressco.com/brain-tumor-segmentation-using-k-means-
clustering-algorithm/
Khan, A. M., & Ravi, S. (2013). Image segmentation methods: a comparative
study. International Journal Of Soft Computing And Engineering (IJSCE), 3(4).
D1760093413/2013©BEIESP
Lin, G. C., Wang, W. J., Kang, C. C., & Wang, C. M. (2012). Multispectral MR images
segmentation based on fuzzy knowledge and modied seeded region growing. Magnetic
Resonance Imaging, 30(2), 230–246. https://doi.org/10.1016/j.mri.2011.09.008
Oliveira, J. V. D., & Pedrycz, W. (2007). Advances in fuzzy clustering and its applications. John
Wiley And Sons Ltd.
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. https://doi.org/10.1016/j.
asoc.2012.11.020
Pandav, S. (2014). Brain tumor extraction using marker controlled watershed segmentation.
International Journal Of Engineering Research & Technology (IJERT), 3(6), 2020-2022.
https://www.ijert.org/brain-tumor-extraction-using-marker-controlled-watershed-
segmentation
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. https://doi.org/10.1016/j.mri.2012.01.006
Selvakumar, J., Lakshmi, A., & Arivoli, T. (2012). Brain tumor segmentation and its area
calculation in brain mr images using k-mean clustering and fuzzy c-mean algorithm.
IEEE-International Conference On Advances In Engineering, Science And Management, 186-
190. Corpus ID: 9928458
Song, T., Jamshidi, M. M., Lee, R. R., & Huang, M. (2007). A modied probabilistic
neural network for partial volume segmentation in brain mr image. IEEE Transactions
On Neural Networks, 18(5), 1424-1432. https://doi.org/10.1109/TNN.2007.891635
291
https://doi.org/10.17993/3ctecno.2021.specialissue8.279-291
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Sudharania, K., Sarma, T. C., & Prasad, K. S. (2016). Advanced morphological
technique for automatic brain tumor detection and evaluation of statistical parameters.
Procedia Technology, 24, 1374–1387. https://doi.org/10.1016/j.protcy.2016.05.153
Zhang, Y., Brady, M., & Smith, S. (2001). Segmentation of brain mr images through a
hidden markov random eld model and the expectation-maximization algorithm.
IEEE Transactions On Medical Imaging, 20(1), 45-57. https://doi.org/10.1109/42.906424