See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/350820310

A Review of Medical Image Segmentation Algorithms

Article in EAI Endorsed Transactions on Pervasive Health and Technology · April 2021
DOI: 10.4108/eai.12-4-2021.169184

CITATIONS READS

159 4,653

5 authors, including:

K K D Ramesh KIRAN KUMAR GUTHIKONDA

University of Hyderabad Raghu Engineering College Visakhapatnam AP India
11 PUBLICATIONS 186 CITATIONS 110 PUBLICATIONS 3,598 CITATIONS

SEE PROFILE SEE PROFILE

Debabrata Datta
Heritage Institute of Technology
145 PUBLICATIONS 1,093 CITATIONS

SEE PROFILE

All content following this page was uploaded by Suman Rajest on 12 April 2021.

The user has requested enhancement of the downloaded file.

EAI Endorsed Transactions
on Pervasive Health and Technology Research Article

A Review of Medical Image Segmentation Algorithms

K.K.D. Ramesh1,2, *, G. Kiran Kumar3, K. Swapna1, Debabrata Datta4,5, and S. Suman Rajest6

1
Department of Physics, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Guntur, AP, India.
2
Department of Radiation Oncology, CIMS Hospital, Ahmedabad, Gujarat, India. [email protected].
3
Department of Physics, Raghu Engineering College, Dakamarri, Visakhapatnam, AP, India. [email protected]
4
SRMIST, Kattankulathur, Chennai, Tamil Nadu, India. [email protected]
5
Department of Mechanical Engineering, GLA University, Mathura, UP, India.
6
Researcher, Vels Institute of Science, Technology & Advanced Studies (VISTAS), Tamil Nadu, India.

Abstract
INTRODUCTION: Image segmentation in medical physics plays a vital role in image analysis to identify the affected
tumour. The process of subdividing an image into its constituent parts that are homogeneous in feature is called Image
segmentation, and this process concedes to extract some useful information. Numerous image segmentation techniques
have been developed, and these techniques conquer different restrictions on conventional medical segmentation
techniques. This paper presents a review of medical image segmentation techniques and statistical mechanics based on the
novel method named as Lattice Boltzmann method (LBM). The beauty of LBM is to augment the computational speed in
the process of medical image segmentation with an accuracy and specificity of more than 95% compared to traditional
methods. As there is not much information on LBM in medical physics, it is intended to present a review of the research
progress of LBM.
OBJECTIVE: As there is no review paper on the research progress of the LB method, this paper presents a review with an
objective to give some thought regarding the different segmentation for medical image and novel LB method to advance
interest for future investigation and exploration in medical image segmentation.
METHODS: This paper in attendance a short review of medical image segmentation techniques based on Thresholding,
Region-based, Clustering, Edge detection, Model-based and the novel method Lattice Boltzmann method (LBM).
CONCLUSION: In this paper, we outlined various segmentation techniques applied to medical images, emphasize that
none of these problem areas has been acceptably settled, and all of the algorithms depicted are available for broad
improvement. Since LBM has the benefits of speed and adaptability of modelling to guarantee excellent image processing
quality with a reasonable amount of computer resources, we predict that this method will become a new research hotspot
in image processing.

Keywords: Segmentation, Medical Physics, Radiation Therapy, Computed Tomography (CT), Magnetic Resonance Imaging (MRI),
Radiotherapy treatment planning systems (RTPS), Image processing, Image analysis, Thresholding, Edge detection, Clustering, lattice
Boltzmann method (LBM)

Received on 20 November 2020, accepted on 07 April 2021, published on 12 April 2021

Copyright © 2021 K.K.D. Ramesh et al., licensed to EAI. This is an open access article distributed under the terms of the Creative
Commons Attribution license, which permits unlimited use, distribution and reproduction in any medium so long as the original work
is properly cited.

doi: 10.4108/eai.12-4-2021.169184

Corresponding author. Email: [email protected]

*

and further hotspot and focal point in image processing

techniques. Image segmentation is a technique of
1. Introduction partitioning an image into sub constituent parts,
remarkably homogenous in feature and this process
Image processing techniques have become progressively concede to extract some useful information. Image
significant in a wide assortment of applications with segmentation accepts a part prominent position in image
sophisticated techniques and instruments. Image analysis. Medical images take part in health care as the
segmentation is an exemplary subject in image processing

EAI Endorsed Transactions

1 on Pervasive Health and Technology
Online First
K.K.D. Ramesh et al.

medical image's quality influences the diagnosis and

treatment. Segmentation is significant in medical image
analysis and intends to draw out some details from the
images. Those images can be utilized for high-level image
understanding. Scientifically speaking, segmentation is a
hypothetical middle-level vision task performed by
neurons between low-level and high-level cortical areas,
as schematically shown in figure 1[1]. Segmentation of
clinical images intends to find anatomic structures and
portray their limits on a computerized source. Especially Figure 1. Image Engineering and Image
in Radiotherapy (RT), Imaging is an essential part of the Segmentation [1]
therapy routine since it utilizes to perceive the therapy
target and ordinary structures to abstain from irradiation.
Hence, Radiotherapy treatment planning systems (RTPS) 2. Segmentation techniques
require anatomical data to be portrayed on CT scans. The
treatment target and ordinary structures are physically Of late, a lot of endeavours have been concentrating on
depicted on CT scans by clinicians. Segmented clinical the segmentation process. Distinct segmentation methods
images are moved to an RTPS for computing radiation or algorithms have been discussed in the literature. These
dose. Hence, precise segmentation is crucial for the methods conquer different restrictions on conventional
patient treatment result. For the reasons of radiotherapy medical division techniques. Yet, there isn't so much as
treatment, segmentation quality incorporates spatial one single strategy to be considered a better technique for
precession and dose computation exactness, which are different kinds of images; those techniques are only
firmly related [2, 3]. Image segmentation aims to partition suitable for particular images and applications. Image
the medical image into elemental parts or objects in the division techniques can be grouped into [6] Thresholding,
Region growing & region merging and splitting,
medical image with uniform and similar features. One of
Clustering, Edge detection and Model-based methods. All
them is intensity, a colour that relates to anatomical
segmentation techniques depend on two essential things
structures, tumours, etc. The degree of segmentation is of intensity values, discontinuity, and similarity. The
application dependent, and there is no universal theory of discontinuity proceeds towards segment the image based
image segmentation. Distinct segmentation methods or on an instantaneous change in intensity levels in an image
algorithms have been proposed in the literature. These or grey levels of an image. In this technique, our interest
techniques conquer different restrictions on conventional predominantly focuses on the recognition of isolated
medical segmentation techniques. Choice of specific points. The second approach is based on the pixels, which
techniques or algorithm over the other depends on the are similar in some range as per the preset criteria upon
problem's image type and nature. Recent advances in which images are segmented, includes approaches like
image segmentation techniques have frequently been Thresholding, Region growing, and region splitting and
reviewed [4, 5]. These review articles are driven by merging.
classifying the methods utilized for processing pixel data
voxel data and their applications in diagnosis, treatment 3. Thresholding approach
planning and follow up studies; However, methods that
have been applied in practice, a shortcoming still exists, Thresholding is the primary method of image
an obvious one is a computational speed. This paper in segmentation, in which one threshold value is used to
attendance a short review of medical image segmentation change a greyscale image into a binary image. This
techniques based on Thresholding, region-based, procedure's important thing is to choose the threshold
Clustering, edge detection, model-based, and the novel value (T), pixels with intensity over and above the
method Lattice Boltzmann method (LBM) to augment the foreground region's threshold value, and all other pixels in
computational speed, which is based on the microscopic the background region [7]. A few famous strategies
description of the macroscopic physical process. As there utilized in the industry are "The maximum entropy
is no review paper on the LB method's research progress, method" [8], Otsu's method (maximum) variance [9]. K-
this paper presents a review to give some thought means clustering [9] can also be used in this regard. The
regarding the different segmentation for medical images presence of noise and obscure boundaries influences
and novel LB method to advance interest for future Thresholding segmentation and performs well for the
investigation and exploration in medical image images with sharp edges [9]. To invalidate the impact of
segmentation. noise on Thresholding smoothing Image and
Thresholding with Edge detection are two regular
strategies.

EAI Endorsed Transactions

2 on Pervasive Health and Technology
Online First
 A Review of Medical Image Segmentation Algorithms

4. Region-based methods flooding. In the process of rainfall algorithm

neighbourhood, lowest is established all over the image.
There are three essential techniques for Region-based Every neighbourhood lowest is given a distinctive
Segmentation of the Image. marking, close to neighbourhood lowest is joined with
this distinctive marking. A theoretical water drop is kept
Region growing at each unmarked pixel. The drop makes headway, its
negligible sufficiency neighbour until it comes to a
At the region growing algorithms beginning of the marked pixel, where it presumes a marked value. In the
specific pixel, the region's growth will be dependent on process of flooding, single theoretical pixels are kept at
the connectivity with the neighbouring cells depending on each neighbourhood lowest. An ocean of pixels continues
similarity criteria corresponding to the greyscale intensity, to drench the neighbouring area of the pixel located at the
shape size, or colour defined by the thresholds to expand neighbourhood lowest. If there is a flood of pixels, the
the growth [10]. The choice of the seed point and the additional pixels are disposed of, and later one is speedier
similarity criteria reflects the segmentation results by than earlier. Yet, it isn't reasonable for frail boundaries
region growing. Statistical information and prior images for the segmentation. Standard watershed methods
knowledge assimilated in algorithms to depend on starting are vulnerable to excessive segmenting of the images
seed points and make the algorithm adaptive [11]. when it contains noise or the objects themselves having a
low signal to noise ratio [14]. Excessive segmentation can
be minimized by selecting a suitable filtering technique,
Region split and Merge approach thereby eliminating the unrelated neighbourhood lowest
[15, 16]. The power watershed algorithms unravel the
The second method based on the region is the split and issues in regular watershed algorithms, thereby delivering
merge image segmentation. The said method depends on a the ideal outcome [17]. Incorporates the benefits of
quadtree information depiction where splitting of an unsupervised neural network (NN) classification and
image takes place into four quadrants with due morphological watershed segmentation to draw out the
consideration of the non-uniform feature of the original exact contours of breast tumours from ultrasound images
image segment. If splitting of neighbouring images is [18]. The new variant stochastic watershed was applied to
discovered as uniform, they are converged by a solitary increase the accuracy of the selected contour [19].
image made out of the four neighbouring images [11]. Optimal parameters of the method are tuned utilizing
The cycle ends when no further merges are conceivable. training. Afterwards, they were applied to 17 data sets,
This technique wipes out the high recurrence artefacts and the demonstrated results show that it is a robust tool
with the seed point selection depend on nearby statistics, for automatic liver segmentation compared to the other
utilized for evaluating the breast and cyst mass [11]. methods. Extraction of tumour in brain MR images was
These algorithms predominantly depend on the image done using marker-controlled watershed algorithms and
intensity information to deal with the partial volume distinctive feature combinations [20].
effects and control the leakage.

Watershed approach 5. Clustering approach

The other technique that depends on the region is the A gathering of homogeneous data into clusters depend on
Watershed method. It views the image as a topographic analogy criteria is called Clustering. "K-means clustering
surface. The idea is that low-intensity pixels are (hard clustering)" is the fundamental clustering
interpreted as valleys of the surface, while high-intensity segmentation algorithms in which the dataset's
pixels appear as hills or peaks [12]. The algorithm starts components belong to only one cluster at a time. Soft
by filling the valleys from local minima if we have seeds clustering approach (FCM) in which the pixel can have a
that are water sources. Every seed can colour with place with greater than one group and the fuzzy
different coloured water. Those seeds are used to flood community function value is the decisive power to fit in
such local minima to increase the water in each part. with a pixel in the group, and the value lies in the span of
When different water in terms of colour meets, a barrier is 0 to 1. The pixel position is exceptionally near the
built to prevent them from merging from one region to centroid, and it very well may be obliged in that class if
another, the watershed lines [13]. This process of filling the community function value is one. The summation of
with water and barrier construction continues until all the squared Euclidean separation connecting to each input
peaks are underwater, then the barrier they were found sample and relating to the cluster centre with the fuzzy
during the process represents the segmentation. In such a community's distances is called the objective function
way, this algorithm integrates the information of the [21]. FCM algorithm is used to partition both grey and
gradient and grey levels of the image. The gathering of colour images; we can set the number of clusters in
water towards the neighbourhood of local lowest called advance. A difference from the anticipate value may assist
catchment basins. The pixels in the greyscale images are us in finding out any shortcomings in the segmentation. In
clone to water beads, and the outcome of the catchment FCM, the objective function can be balanced and
basins are the part of the locale of the image having a like stretched out according to the necessities. The
class of pixels. There are two basic algorithm processes to convenience of the variability of the objective function
deal with the image using watershed: (a) rainfall and (b) makes FCM appropriate for various types of images. The
FCM algorithm itself has the facility for the validity

EAI Endorsed Transactions

3 on Pervasive Health and Technology
Online First
K.K.D. Ramesh et al.

measure, thus minimize the computational exertion yields of the shape and existence of the structure. This technique
palatable outcomes. Refinement of the FCM algorithms is sturdier against the artefacts associated with the images
was in practice. The outcome of the customary FCM than the conventional algorithms.
algorithms is subtle to noise in compensating MR images'
intensity heterogeneities. It replaces the Euclidean Markov Random field models
separation with the kernel induced separation called
Kernelized fuzzy C-means (KFCM)[22]. The dimensional It is a simple stochastic process in which future states'
penalty was added to the objective function. Presented distribution relies only on the current state and not how it
Fast generalized fuzzy, c-means (FGFCM) clustering arrived in the current state. A random sequence with the
algorithm by assimilating the local spatial and grey data in Markov property if its present status dictates its
conjunction to suppress the noise and detail preserving for distribution is called Markov random fields (MRF),
an image [23]. A portion of the upgrades in the regular inspired by the Ising model [29]. MRF methods have been
FCM algorithms was made. It was tested these algorithms broadly utilized to segmentation the images, and the
in the presence and absence of noise on CT brain images image's revamped since they can protect the edges by
to identify the abnormality region and with a bacteria parameter approximation [30]. Presented Hidden Markov
image to separate the bacteria from the background. They random field (HMRF) [31], based on a random process
concluded that T2FCM (TYPE-II FUZZY C-MEANS) created by MRF model whose state of arrangement cannot
has efficiently removed the noise with the cost of an be watched straightforwardly, which can be seen through
increase in the object's size. IFCM (INTUITIONIST an examination. It can be shown mathematically "FM
FUZZY C-MEANS) was efficient in segmenting the model" is a decadent version of the "HMRF model". By
images in contrast to other methods. Proposed "Fuzzy incorporating the HMRF model and an "Expectation-
based artificial bee colony (FABC)", where they maximization (EM)" algorithm into an "HMRF-EM
combined "Artificial bee colony optimization (ABC) framework", precise and sturdy segmentation achieved
"with the "Fuzzy C-Means (FCM)"[24]. They used the and was exhibited by correlation experiments with the
Fuzzy community function to search an optimum cluster "FM model-based segmentation". Discussed the
centre using ABC and tested on synthetic and medical combination of MRF and SOFM (self-organizing feature
images; the efficiency results were shown in contrast with map) by incorporating the spatial constraints, which
the other methods. Proposed ARKFCM, a customizable improves the smoothness of partitioning regions [32].
regularized kernel-based algorithm for brain MRI Discussed mass breast segmentation using proclaimed
segmentation, upgrades the robustness to preserve image PRF (Pickard random fields), an "Unsupervised MRF
subtleties with the advent of the regularised kernel model". The PRF model was discovered to be productive
adaptability [26]. compared to the regular MRF in terms of computational
complexity [33].
6. Edge detection Atlas-based approach

It is the most traditional methodology for the detection of Atlas consists of images, including anatomical subtleties
irregularity in an image. The separation between two to join earlier data for partitioning and outcome changes
areas having distinct intensity levels or grey level is called depending on the Atlas' particulars. The informational
Edge. The edges are remarkable in any applications and indexes of clinically debilitated subjects and ordinary
image enhancement to enhance the details of the image. subjects are used to advance an atlas. Atlas can be a
Derivative operations are used for the Edge's existence in physically marked picture; it possesses a close connection
an image, and primarily a convolution function is done on with the picture to be partitioned. The image enrolment
a picture with an appropriate mask [11]. The canny is an takes part in an essential job, and numerous atlases
efficient edge detector that enhances the Edge; it uses the enhance the partition precession [34]. In the image
gradient extent threshold to locate the possible Edge. It enrolment course, the physically marked Atlas was
stifles them through the system of non-maximal changed by a mapping strategy frequently named
suppression and hysteresis thresholding [27]. Noise in the "LABEL PROPAGATION" to precise partitioning the
Image strongly influences edge detections leads to the earmark object. The exactness of enrolment is essential
identified edges comprised of discrete pixels, might be since mistakes might happen if there is a topographic
inadequate or irregular; hence before the edge detection, distinction between the Atlas and the examination object
the image has to smooth using the Gaussian operator. The [35]. Choice of source images for an Atlas development
images pre-processing may erstwhile lead to fallacious was made by either deciding the population's true mean or
edge detection and can get rid of multi-resolution Edge collecting the sample closer to the mean. Depending upon
detection and Edge tracing techniques [28]. several parameters, variable numbers of iterative
algorithms could diminish the atlas development's bias
effect. On account of numerous atlas partitions, an
7. Model-based algorithms enormous database is required, and the determination of a
proper atlas of the query image is to be done [36]. Labels
Model-based approaches have been set up as probably the of Atlas for the earmark depiction was confined by
best strategies for picture analysis to go with a model. dimensionally changing the fusion weights obtained from
This model contains information about the look forward neighbourhood evaluation of the enrolment activity [37].

EAI Endorsed Transactions

4 on Pervasive Health and Technology
Online First
 A Review of Medical Image Segmentation Algorithms

They presented average -shape atlas-based segmentation Graph cut approach

strategy makeover the preferred outcome over a single
atlas-based strategy of cardiac and aortic Segmentation in The graph cut algorithm's fundamental idea is to borrow
CT scan images [37]. In this study, authors utilized rib- tools from the graph theory to partition the image into
bone atlases to automatically take out the patient rib bone foreground and background. In graph theory, every pixel
on conventional chest x-ray (CXR) utilizing the will be a node, and the edges are the links that connect
physically alongside the mimicked models from CT and those nodes. Connecting links are the probability of that
bone images acquired from the dual-energy x-ray machine node being foreground or background to connect the
[38]. During the enrolment, they compute the source (S) or sink (T) with the weight correlated to the
transformation mapping between the model x-ray and the probability, i.e., the Edge's weight. The weight that
patient x-ray by the regional similarities between the x- promotes the pixels that are similar to stay together in the
rays, and they apply the result mapping to the rib masks. same segment and the different pixels promotes them to
The average of enrolled models establishes the rib bone become different segments [46]. After constructing the
likelihood map for the patient-X-ray. graph, partition that graph by creating a minimal cut with
the least effort that partition the foreground and
Artificial neural networks background. With some hard constraints, the cost function
represents the boundary and region properties considered
It is a mathematical model of neurons, just like human as soft constraints for the segmentation. When the hard
brain cells inspired by biological neural networks. The constraints are changed, the global optimization
node is nothing but a replica of neurons, which may be recomputed as per the new constraints when the cost
implemented with specific functional units. These nodes function is mentioned. In the graph, cut pixels represent
are connected through communication links with synaptic the nodes, and edges are the weighed connecting nodes.
weights. The inputs of these synaptic weights are They compute the global optimal minimum cut to obtain
processed to classify or identify the object with the the object & background from the image. They
activation function's help [39]. Training and learning are demonstrated this technique with photo, video editing and
the two significant features of neural networks. Neural Medical image processing [46]. They proposed two
networks are prepared with attributes, perhaps statistical algorithms to minimize the energy function, which is
in a feature like mean, standard deviation, kurtosis, based on the graph cuts. In the first, the labelling is done
skewness, or transfer, depending upon the attributes with among the arbitrary set of pixels so that there is a
the application of Wavelet transform or Curvelet movement among the labelled pixels so that energy is
transform. During the training phase, the introductory minimized. The second algorithm requires smoothing.
phase of neural networks is named a speculating phase They used three quadratic energy functions first one is
and progresses until the study state achieved. Moreover, with truncated quadratic energy function. The second and
the neural network's training, the more acceptable the third energy functions correspond to the Potts model, and
outcome related to the examination image. Learning is a truncated distance as the penalty of smoothness and the
customizable activity in which the weights related to the results are compared with the different annealing variants.
interlinking neurons changed to give appropriate [47]. The graph cut algorithm gives an optimum outcome
feedback. The Learning process involves neural networks than the conventional graph cut algorithms with earlier
grouped as supervised learning and Unsupervised learning information about the foreground's shape. Multiregional
[40]. Neural network algorithms have significant graph cut partitioning is done through kernel mapping of
difficulty determining the architecture, network size, type, the image data, and the objective function contains
number of layers, architecture, network size and original data to assess the deflection of the changed
geographies. The choice of the above said components image. The optimization algorithm iterated two
influence the exhibition of dealing with the problem. consecutive steps: graph cut optimization and fixed point
Identification of the lungs from the clinical images was iterations for refreshing the region's parameters. This
described by the "Group Method of Data Handling strategy bears a powerful option in contrast to the
(GMDH)" [41]. Fuzzy neural networks yield better complex modelling of the original data while exploiting
segmentation result, which was insensitive to noise [42]. the computational advantages of graph cut images. This
Proposed separation measurements utilized by the SVM method was quantified and comparatively validated with
classifier and also investigated the importance of synthetic, natural, and medical images. Kernel mapping
hyperparameter selection [43]. This method proved that yields good results in the multi-region partition of brain
the segmentation accuracy is good apart; it takes less MR images [48].
processing time and less memory usage. Worked on
segmentation of liver malignancy out of MR images Lattice Boltzmann method (LBM)
utilizing a combination of "3D fast marching algorithm"
and the "Single hidden layer "feed-forward neural They are powerful techniques with a high degree of
network. The results are validated with those physically accuracy. Lattice Boltzmann method (LBM) is one such
delineated by a radiologist utilized as ground truth [44]. simulation technique that depends on the microscopic
The time unpredictability was extraordinarily decreased, elucidation of the macroscopic physical process, which
and precise outcomes were acquired with other semi- has applied widely in kinetic theory to simulate various
automatic segmentation techniques [44]. A deep systems [49]. In LBM, an attempt has been made to close
convolution Neural network (DCNN) was used to identify the gap between the macroscopic and microscopic scales
the brain MR images' glioblastomas [45]. by taking the behaviour of a group of particles supposed

EAI Endorsed Transactions

5 on Pervasive Health and Technology
Online First
K.K.D. Ramesh et al.

to behave similarly, rather than a single particle Utilized LBM to tackle the Level Set Equation (LSE),
behaviour. Each such collection of particles is given as a they proposed a Region-based stop function [71-76].
distributive function, which represents the collection of Unsigned Pressure Force (UPF) in light of regional
particles. In LBM, the solution area is separated into attribute can adequately and effectively stop the contour
lattices. At every lattice node, the dissemination of at feeble obscured edges [57]. They utilized the
particle resides. Some of these particles advance along a parallelizable lattice Boltzmann method (LBM) to solve
specific direction to the adjacent node. The number of the level set equation(LSE). This technique is quicker
directions, linkage relies upon the lattice alignment. The since it is unravelled in histogram space instead of pixel
normal phrasing utilized in LBM DnQm, where 'n' speaks
domain [77-81]. The time problem is extensively
the dimension of the problem, 'm' alludes to the speed
diminished since the quantity of grey levels is commonly
model, the number of linkages. The crucial element in
LBM is the equilibrium distributive function (few) with a a lot littler than the size of the picture. The strategy is
relaxation time (τ) ascertaining the sort of issue that productive, profoundly parallelizable, and quicker than
should have to be tackled. LBM has an alternate tool to those dependent on the LSM [58]. Edge of object division
the conventional mathematical methods for solving partial can be obtained by the Lattice Boltzmann Anisotropic
differential equations (PDE). LBM is faster, easier, and it Diffusion Model (LBADM) and indicated their
has the facility to assimilate large parallel computation calculation could precisely fathom the convection and
and has the advantage of occupying less memory during diffusion condition [59]. The algorithm incredibly lessens
the simulations as it considers the distribution of particles the calculation of division. They proposed a new variation
rather than tagging each particle. The general LBM multiphase level set approach to clinical segmentation,
comprises of two stages as follows: a streaming stage in calculated multiphase level set conditions and entropy for
which particles (or particle densities) moves from one the LBM model for D2Q9[60]. Demonstrated their
node to another node on a lattice and a collision stage in method for MR breast images reveals it is more efficient
which particles (or particle densities) are reorganized at and faster [81-84]. The LBM's merit is a pixel-based
each node [50]. algorithm. It deals with the particles, and the pixels are
The two stages are governed by the LBM evolution the replica of particles, therefore at any resolution, we can
equation, where parameter relaxation time(τ) and source tune the algorithm using various kinds of lattice points.
term (α) decide the particles' movement. The state of each With the increase of lattice points, computational load
node at the next moment is only related to the state of its also will be increasing, and it may be a demerit of the
neighbouring nodes because the particles move along the LBM in the view of computation. Since it is particle-
links, and the lattice number and lattice speed govern the based in the microscopic domain, all the macro
links. LBM can be efficiently used in image analysis parameters can be redefined, but since LBM is dealt with,
techniques such as (i)image smoothening [51-53], (ii) density has to define in a regular structure. In a medical
image inpainting [54], (iii) image segmentation [55- image, getting a regular structure is very difficult;
60]etc. In image processing, each pixel value is therefore, it may have the LBM's demerit.
considered as particle densities, and changes in pixel
value can be considered as a redistribution of particles
that are decided by relaxation time (τ) source term (α) in
8. Conclusion
which image information such as gradient and curvature
are embedded. LBM in image processing is done, which
can be applied easily to complex domains and could be This paper presents a review to explain the different
used to serve multiphase and multicomponent flows. segmentation for medical images and novel LB method to
Introduced an anisotropic diffusion model dependent on advance interest for future investigation and exploration
LBM for picture division and showed the adequacy of the in medical image segmentation. Emphasize that none of
calculation in clinical pictures [55], [61-65]. Proposed a these problem areas has been acceptably settled, and all of
novel LBM method using the D2Q19 lattice arrangement the algorithms depicted are available for broad
model for the Segmentation of MR and clinical images, improvement. Segmentation of medical images is a yet
which is similar to anisotropic diffusion as shown in complicated issue in real-time applications like diagnosis
figure 2 [56], [66-71]. and radiotherapy treatment where segmentation algorithm
can accurately recognize different tissues encompassing
the tumour site and tumour boundaries, henceforth more
inventive work is required augment the computational
speed. LBM has the benefits of speed and adaptability of
modelling to guarantee excellent image processing quality
with a reasonable amount of computer resources. The
LBM has an exact physical meaning in image processing,
the image's pixel value is considered particle densities,
and changes in the pixel value can be considered
Figure 2. Segmentation of MR image by LBM [56]. redistribution of particles. It is governed by the relaxation
time, which decides the kind of problem needed to be
(a) MR Image (b) LBM addressed, and the addition of source term is straight

EAI Endorsed Transactions

6 on Pervasive Health and Technology
Online First
 A Review of Medical Image Segmentation Algorithms

forward. The quality of the image segmentation and the [13] Belaid LJ, Walid M. "Image segmentation: a watershed
transformation algorithm,".Image Analysis Stereology.
computational speed utilizing LBM with high 2009;28:93-102.
dimensionality and more lattice vectors are expected to [14] Hanbury A." Image segmentation by Region-based and
comprehend, so we predict that the LB method will watershed algorithms,".Wiley Comp Sci Eng.2009;543-
become a new research hotspot field Medical image 1552.
segmentation. [15] Biernacki W ." Over segmentation avoidness in watershed
based algorithms for color images,".Proc IEEE
International Conference Modern Problems Radio Eng
Acknowledgements Telecommuni Comp Sci,28 Feb. 2004, Lviv-
Authors KKDR, GKK and KS would like to thank the SlavskoUkraine,IEEE, 30 Dec 2004;169-172.
HOD and division of Radiation Oncology team of CIMS [16] Grau V Mewes AUJ, Alcaniz M, Ron K Simon KW."
Improved watershed transform for medical image
Hospital, Ahmedabad and KLEF, Guntur, for their segmentation using prior information,".IEEE Transactions
constant support and encouragement. The authors also Medical imaging..2004;23:2606-2614.
wish to acknowledge the Department of Science and [17] Hamid R.Arabnia," Advances in computational
Technology (DST), Govt. of India, for the DST-FIST biology".Springer Science Business Media, 2010;4-18.
Level-1 (SR/FST/PS-1/2018/35) scheme of Physics, [18] Huang YL, Dar RC." Watershed segmentation for breast
KLEF. tumour in 2D sonography,".Ultrasound MedBiol.
2004;30:625-632.
[19] Lopez MF, Valery N, Jesus A, Alcaniz M Luna L ." Liver
segmentation in MRI: A fully automatic method based on
References stochastic partitions,".Comp Methods Prog Biomed.
2014;114,11-28.
[1] Zhang, Y.J, "An overview of Image and Video [20] Benson CC, Lajish VL, Kumar R." Brain tumour extraction
Segmentation,".Beijing, China, IGIGlobal;2006.chapter-1, from MRI brain images using marker-based watershed
An overview of Image and Video Segmentation in the last algorithm,".International Conference Advances Comp
40 years; p.1-16. Communications Informatics( ICACCI).2015;318-323.
[2] Gregory Sharo, Karl D Fritscher, Vladimir Pekar, Marta [21] Ahmed MN, Sameh MY, NevinM, AlyAf, Thomas M." A
Peroni, Nadya Shusharina, Harini Veeraraghavan and modified fuzzy c-means algorithm for bias field estimation
Jinzhong yang, “ Vision 20/20: Perspectives on automated of Segmentation of MRI data,".IEEE Trans Med Imaging.
image segmentation for radiotherapy,”.Med. Phys.2014; 2002;21:193-199.
41(5):p.050902-13.
[22] Zhang DQ, Song CC ." A novel kernelized fuzzy c-means
[3] Najeeb Chowdhury, Robert Toth, Jonathan Chappelow, algorithm with application in medical image
Sung Kim, Sabin Motwani, Salman Punekar, Haibolin, segmentation,".Artificial Intel Medicine. 2004;32:37-50.
Stefan Both, Neha Vapliwala, Stephen Hahn and Anant
Madabhushi, "Concurrent segmentation of the prostate on [23] Cai W SongeanC, Daoqiang Z ." Fast and robust fuzzy c-
MRI and CT via linked statistical shape models for means clustering algorithms incorporating local
radiotherapy planning,”.Med.Phys.2012;39(4):,2214-28 . information for image segmentation,".PatternRecogn.
[4] Nida M Zaitoun, Musbah J.Aqel," Survey on Image 2007;40:825-838.
Segmentation Techniques, "Procedia Computer Science [24] Anjea D, TarunKR." Fuzzy Clustering algorithms for
.2015;65:p.797 – 806. effective medical image segmentation,".Int J Intel Sys
[5] Kumar SN, Lenin Fred A, Muthukumar S, Ajay Kumar H Appl.2013;5:55.
and Sebastian Varghese P,".A Voyage on Medical Image [25] AnkitaB, Kalyani M." Fuzzy-based artificial bee colony
segmentation algorithms,” Biomedical Research. optimization for grey image segmentation,".Signal and
2017;Special Issue,Res.p.1-12. Video Proc.2016;10:1089-1096.
[6] D. L. Pham, C. Xu, and J. L. Prince, "Current methods in [26] Ahmed E, ChangmiaoW, Fucang J, Jianhuang W, Guanglin
medical image segmentation,".Annual Review of L Qingmao H ." segmentation of brain tissues from MRI
Biomedical Engineering. 2000;vol.2:pp.315-337. images using adaptively regularized Fuzzy c-Means
[7] W. X. Kang, Q. Q. Yang, R. R. Liang, "The Comparative Clustering,".Hindawi Publishing Corporation
Research on Image Segmentation Algorithms,”.IEEE Computational Mathematical Methods Medicine. 2015;1-
Conference on ETCS. 2009; pp. 703-707. 12.
[8] R.C.Gonzalez and R.E.Woods," Digital image processing,” [27] Nadeem M, Asadullah 88uhnjgbS, AdamuA, Ahafia K
2-ed.NewYork: Prentice Hall; 2002.1-813. Syed BZ." Image segmentation methods and edge
detection: an application to knee joint articular cartilage
[9] R.M Haralick and L.G.Shapiro,”Image Segmentation edge detection, ".J Theor Appl Inform Technol. 2015;71.
Techniques,“.Computer Vision ,Graphics,and Image
processing .1985;Vol-29,Issue 1, p.100-132. [28] Lopez-MolinaC, DeBates B, Bustince H, SanzJ,
Barreneche E." Multiscale edge detection based on
[10] Wu J, Skip P, Michael DN, Marked VK." Texture feature- Gaussian smoothing and edge tracking, ".Knowledge-
based automated seeded region growing in abdominal MRI Based Systems. 2013;44:101-111.
segmentation,".Proc IEEE international conference
Biomedical Engineering and Informatics,27-30May [29] Kindermann, Ross; Snell, J. Laurie." Markov Random
2008,Sanya, China,IEEE,2008;2:263-267. Fields and their applications,".1-ed. Providance, Rhode
island, American Mathematical Society.1980.1-147.
[11] Thakur A Radhey SA. "A local statistics based region
growing segmentation method for ultrasound medical [30] Held K Rota KE, Bernd JK, William MW, Ron K, Muller
images".World Academy Science Engineering and Gartner HW." Markov random field segmentation of brain
Technology, Int J Med Health Biomed, Boeing Pharm Eng. MR images,".IEEE Trans Medical Imaging. 1997;16:878-
2007;1:564-56. 886.
[12] T.Pavlidis, "Algorithms for Graphics and Image [31] Zhang Y, Brady M, Smith S. "Segmentation of brain MR
Processing, "Computer science press, Rockville, MD,1982. images through a hidden markov random field model and
expectation-maximization algorithm,".IEEE Trans Med
Imaging .2001;20.

EAI Endorsed Transactions

7 on Pervasive Health and Technology
Online First
K.K.D. Ramesh et al.

[32] Li Y, ZheruC ." MR Brain image segmentation based on [50] A.A. Mohamad "Lattice Boltzmann method: Fundamentals
self -organizing map network,".Int J Inform Technol. and Engineering Applications with Computer codes,"
2005;11:45-53. Springer–Verlg London Limited: Springer;2011.1-73.
[33] Segbedji RTJG, YvesG, Hichem M. "Unsupervised [51] B Jawerth, P Lin and Sinzinger E "Lattice Boltzmann
malignant mammographic breast mass segmentation models for anisotropic diffusion of images,".Journal of
algorithm based on Pickard Markov random field,". IEEE Mathematical Imaging and Vision. 1999;11:231-237.
Int Conference Image Proc,.2016;156-164. [52] Y Chen, Z.Z.Yan and Y.H. Qian, "An AnistropicDifussion
[34] Hyunjin P, Peyton HB Charles RM," Construction of an Model for Medical Image smoothing by using the Lattice
abdominal probabilistic atlas and its application in Boltzmann method,".7th Asian-Pacific Conference on
segmentation ".IEEE Trans Med Imaging .2003;22:483- Medical and Biological Engineering. IFMBE Proceeding
492. of APCMBE;22-25 April 2008; Beijing, China.China:
[35] Kang WX, Qing QY, Run PL ." The comparative research Springer; 2008. 255-259.
on image segmentation algorithms,".Proc First Int [53] Wenchuan Zhang, Baochang Shi," Application of Lattice
workshop Education Technology Computer Science Boltzmann method to Image filtering ", Mathematical
.2009;1:703-707. Imaging and Vision. 2012;43:135-142.
[36] CauadraMB, Claudio P, Anton B, Olivier C, Villemure JG, [54] Y Chen" A lattice-Boltmann method for Image inpainting",
ThiranJP." Atlas-based Segmentation of pathological MR 3rd international congress on imge and signal processing
brain images using a model of lesion growth ".IEEE Trans (CISP) ;Yantai, China.China:IEEE;2010.vol 3,pp.1222-
Med Imaging .2004;23:1301-1314. 1225.
[37] Ivana I, Marius S, Annenarieke R, Mathias P Max AV, [55] Y Chen, Z Yan and Shi J, "Application of Lattice
Bram van G "Multi -atlas-based segmentation with local Boltzmann method to image segmentation", 29th Annual
decision fusion application to cardiac and aortic international conference of the IEEE engineering in
segmentation in CT scans,".IEEE Trans Med Imaging. medicine and biology society;22-26 Aug. 2007; Lyon,
2009;28:1000-1011. France.France:IEEE;2007.pp.6561-6564.
[38] Sema C,Stefan J ,Sameer A UlasB,Les RF,Ziyue X,George [56] Y Chen, Z.Z.Yan and Y.H. Qian," Lattice Boltzmann
T.” Atlas based rib-bone detection in chest x-rays,“.Comp Method based Medical Image Segmentation.".2009 2nd
Med Imaging Graphics. 2016;51:32-39. International Congress on Image and Signal Processing;17-
[39] Dayhoff JE, DeL, James M," Artificial neural networks 19 Oct .2009; Tianjin, China. China:IEEE;2009.pp.1-5.
".Cancer John Wiley Sons Inc.2001;91:1615-1635. [57] S Balla-Arabe, B Wang and X Gao," Level set region-
[40] Egmont P, Michael, Heinz H ."Image processing with based image segmentation using lattice Boltzmann
neural networks -a review,''.Pattrecogn.2002;35:2279-2301. method,".Seventh International Conference on
computational intelligence and security(CIS);3-4 Dec.
[41] Tadashi K AbhijitSP, Jack MZ." GMDH type neural 2011; Hainan, China. China: IEEE;2011.pp,1159-1163.
networks and their application to the medical image
recognition of the lungs".Proc 38th SICE Annual [58] S Balla-Arabe, X Gao," Image multi-thresholding by
Conference;30-30 July 1999; Morioka, Japan.Japan:IEEE; combining the lattice Boltzmann model and a localized
1991;1:1181-1186. level set algorithm,".Neurocomputing.2012; 93:106-114.
[42] WongchomphuP, Narissara E ." Enhance neuro-fuzzy [59] Q Wu, Y Chen and G Teng, "Lecture Notes on Electrical
system for classification using dynamic Clustering,".Proc Engineering 107 ".Computer, informatics, cybernetics and
4th Joint international Conference Information applications,1-ed.Berlin: Springer.Chapter 61,” Lattice
communication Technology Electronics and Electrical Boltzmann anisotropic diffusion model-based image
Engineering;5-8 March 2014; Chiang Rai, Thailand: segmentation;p.577-589.
IEEE;2014;1-6. [60] S Balla-Arabe, Wang B and X Gao, "A Multiphase
[43] Mohmmad H Hugo L, Phillippe P." Within Brain entropy-based level set algorithm for MR breast image
classification for brain tumour segmentation,". International segmentation using lattice Boltzmann model".Intelligence
Journal of Computer Assisted Radiology and Surgery. science and intelligent data engineering,2013;7751:8-16.
2015;11:777-788. [61] Barot, V., Kapadia, V., & Pandya, S., QoS Enabled IoT
[44] Trong –Ngoc L, Pham The B, HieuTH," Livertumour Based Low Cost Air Quality Monitoring System with
segmentation from MR images using 3D fast marching Power Consumption Optimization, Cybernetics and
algorithm and single hidden layer feed word neural Information Technologies, 2020, 20(2), 122-140. doi:
network,".Hindawi Publ Corp Bio Med Research Int. https://doi.org/10.2478/cait-2020-0021.
2016;1-8. [62] Sur, A., Sah, R., Pandya, S., Milk storage system for
[45] Mohammad H Axel D, David WF, Antoine B Aaron C, remote areas using solar thermal energy and adsorption
YoshuaB, ChrisP, Pierre-MJ, HugoL." Brain tumour cooling, Materials Today, Volume 28, Part 3, 2020,
segmentation with deep neural networks,".Medical Image Elsevier, Pages 1764-1770, ISSN 2214-7853,
Analysis. 2017;35:18-31. https://doi.org/10.1016/j.matpr.2020.05.170.
[46] Boykov YY, Jolly MP," Interactive graph cuts for optimal [63] H. Ghayvat, Pandya, S., and A. Patel, "Deep Learning
boundary and region segmentation of objects in ND Model for Acoustics Signal Based Preventive Healthcare
images,”.Proc Eighth IEEE International Conference Monitoring and Activity of Daily Living," 2nd International
Computer Vision. 2001;1:105-112. Conference on Data, Engineering and Applications (IDEA),
Bhopal, India, 2020, pp. 1-7, doi:
[47] Boykov Y Olga V, Ramin Z. "Fast approximate energy 10.1109/IDEA49133.2020.9170666
minimization via graph cuts,".IEEE Trans Patt Anal Mac
Intel .2001;23:1222-1239. [64] Pandya, S., Shah, J., Joshi, N., Ghayvat, H.,
Mukhopadhyay, S.C. and Yap, M.H., 2016, November. A
[48] Ben Salah M, Mitiche A, Ben AyedI." Multiregional image novel hybrid based recommendation system based on
segmentation by parametric kernel graph cuts".IEEE Trans clustering and association mining. In Sensing Technology
Image Process. 2011;20:545-557. (ICST), 2016 10th International Conference on (pp. 1-6).
[49] D.Grunall, S.Chen and K. Eggert." A Lattice Boltzmann IEEE.
model for multiphase fluid flows,".Phys [65] Pandya, S., W. Patel, H. Ghayvat, “NXTGeUH: Ubiquitous
Fluids.1993;5:2557-2562. Healthcare System for Vital Signs Monitoring & Falls
Detection”, IEEE International Conference, Symbiosis
International University, December 2018.

EAI Endorsed Transactions

8 on Pervasive Health and Technology
Online First
 A Review of Medical Image Segmentation Algorithms

[66] Ghayvat, H., Pandya, S., “Wellness Sensor Network for Computing, International Journal of Cloud Computing
modeling Activity of Daily Livings Proposal and Off-Line (IJCC), Vol. 9, No. 2/3, 2020.
Preliminary Analysis” IEEE International Conference, [81] Haldorai, A. Ramu, and S. Murugan, “Social Aware
Galgotias University, New Delhi, December 2018. Cognitive Radio Networks,” Social Network Analytics for
[67] Mehta, P., Pandya, S., A review on sentiment analysis Contemporary Business Organizations, pp. 188–202.
methodologies, practices and applications, International doi:10.4018/978-1-5225-5097-6.ch010
Journal of Scientific and Technology Research, 2020, 9(2), [82] M. Suganya and H. Anandakumar, “Handover based
pp. 601–609 spectrum allocation in cognitive radio networks,” 2013
[68] Shah, J., Pandya, S., N. Joshi, K. Kotecha, D. B. Choksi, International Conference on Green Computing,
Load Balancing in Cloud Computing: Methodological Communication and Conservation of Energy (ICGCE),
Survey on Different Types of Load Balancing Algorithms‖, Dec. 2013.doi:10.1109/icgce.2013.6823431.
IEEE International Conference on Trends in Electronis and doi:10.4018/978-1-5225-5246-8.ch012
Informatics, Tamilnadu, India, May 2017. [83] A. J. Obaid, "Critical Research on the Novel Progressive,
[69] D.S. Hooda and D.K. Sharma (2008), Generalized R-Norm JOKER an Opportunistic Routing Protocol Technology for
information Measures-Journal of Appl. Math, Statistics & Enhancing the Network Performance for Multimedia
informatics (JAMSI), Vol. 4 No.2 , 153-168. Communications," in Research in Intelligent and
[70] Ravi Manne, Snigdha Kantheti, Sneha Kantheti, Computing in Engineering. Advances in Intelligent
(2020),"Classification of Skin cancer using deep learning, Systems and Computing, vol 1254. Springer, Singapore.,
ConvolutionalNeural Networks - Opportunities and Springer, Singapore, 2021, pp. 369-378.
vulnerabilities- A systematic Review", International Journal [84] S. Sharma and A. J. Obaid, "Mathematical modelling,
for Modern Trends in Science and Technology, ISSN: analysis and design of fuzzy logic controller for the control
2455-3778, Vol. 06, Issue 11, pp. 101- 108. of ventilation systems using MATLAB fuzzy logic
https://doi.org/10.46501/IJMTST061118 toolbox," Journal of Interdisciplinary Mathematics, vol. 23,
[71] Dilip Kumar Sharma, “Some Generalized Information no. 4, pp. 843-849, 2020.
Measures: Their characterization and Applications”,
Lambert Academic Publishing, Germany, 2010. ISBN:
978-3838386041.
[72] S. Suman Rajest, D.K. Sharma, R. Regin and Bhopendra
Singh, “Extracting Related Images from E-commerce
Utilizing Supervised Learning”, Innovations in Information
and Communication Technology Series, pp. 033-045, 28
February, 2021.
[73] Ganguli S., Kaur G., Sarkar P., Rajest S.S. (2020) An
Algorithmic Approach to System Identification in the Delta
Domain Using FAdFPA Algorithm. In: Haldorai A., Ramu
A., Khan S. (eds) Business Intelligence for Enterprise
Internet of Things. EAI/Springer Innovations in
Communication and Computing. Springer, Cham
[74] Singla M.K., Gupta J., Nijhawan P., Ganguli S., Rajest S.S.
(2020) Development of an Efficient, Cheap, and Flexible
IoT-Based Wind Turbine Emulator. In: Haldorai A., Ramu
A., Khan S. (eds) Business Intelligence for Enterprise
Internet of Things. EAI/Springer Innovations in
Communication and Computing. Springer, Cham
[75] Rao, A. N., Vijayapriya, P., Kowsalya, M., & Rajest, S. S.
(2020). Computer Tools for Energy Systems. In
International Conference on Communication, Computing
and Electronics Systems (pp. 475-484). Springer,
Singapore.
[76] R. Arulmurugan and H. Anandakumar, “Region-based seed
point cell segmentation and detection for biomedical image
analysis,” International Journal of Biomedical Engineering
and Technology, vol. 27, no. 4, p. 273, 2018.
[77] Dr.S. Suman Rajest Dr. Bhopendra Singh, P. Kavitha, R.
Regin, Dr.K. Praghash, S. Sujatha, “Optimized Node
Clustering based on Received Signal Strength with Particle
Ordered-filter Routing Used in VANET” Webology,
Vol.17, No.2, pp. 262-277, 2020.
[78] D Datta, S Mishra, SS Rajest, (2020) “Quantification of
tolerance limits of engineering system using uncertainty
modeling for sustainable energy” International Journal of
Intelligent Networks, Vol.1, 2020, pp.1-8,
https://doi.org/10.1016/j.ijin.2020.05.006
[79] Leo Willyanto Santoso, Bhopendra Singh, S. Suman Rajest,
R. Regin, Karrar Hameed Kadhim (2021), “A Genetic
Programming Approach to Binary Classification Problem”
EAI Endorsed Transactions on Energy, Vol.8, no. 31, pp. 1-
8. DOI: 10.4108/eai.13-7-2018.165523
[80] Dr. Laxmi Lidiya. S. Suman, Rajest, “Correlative Study
and Analysis for Hidden Patterns in Text Analytics
Unstructured Data using Supervised and Unsupervised
Learning techniques” in International Journal of Cloud

EAI Endorsed Transactions

9 on Pervasive Health and Technology
Online First

View publication stats