Professional Documents
Culture Documents
Current Methods in Medical Image Segmentation: A Review: Ramandeep Kaur and Jaswinder Kaur
Current Methods in Medical Image Segmentation: A Review: Ramandeep Kaur and Jaswinder Kaur
Current Methods in Medical Image Segmentation: A Review: Ramandeep Kaur and Jaswinder Kaur
Segmentation: A Review
Ramandeep Kaur1 and Jaswinder Kaur2
1,2
Department of Electronics & Comm. Engineering
Shaheed Bhagat Singh State Technical Campus, Ferozepur, Punjab
E-mail: 1ramancheema92@gmail.com, 2jaswinder.ece@gmail.com
Abstract—In this paper, different approaches are on this approach. Canny edge detector uses the threshold
discussed for medical image segmentation. These are of gradient magnitude to find the potential edge pixels
based on thresholding, learning, modeling and automatic and suppresses them through the procedures of the non -
fuzzy method. Segmentation techniques, discussed under maximal suppression and hysteresis thresholding. In this,
these approaches are used in different applications. In
identification of brain lesions, vessel lumen segmentation
the detected edges are consisted of discrete pixels, may
and histopathology cancer image segmentation. Further be incomplete or discontinuous. So, it is necessary to
used in tissue segmentation based upon image processing apply post-processing like morphological operation to
chain optimization, combining graph cut and oriented connect the breaks or eliminate the holes. Pixels inside a
Active Appearance Model (AAM) and in brain image structure tend to have similar intensities and from this
segmentation by using fuzzy symmetry. These techniques observation idea of region-based algorithms developed.
overcome various limitations of conventional medical Region growing algorithm is a typical algorithm of this
image segmentation techniques. type. In this algorithm firstly initial seeds are selected
Keywords: CTA, MRA, MRI, Segmentation, then it search for the neighbored pixels whose intensities
Thresholding
are inside the intervals defined by the thresholds and then
I. INTRODUCTION merge them to expand the regions. Statistical information
and a priori knowledge can be incorporated to the
Image segmentation is the process of partitioning a algorithms to eliminate the dependence on initial seeds
digital image into multiple segments or sets of pixels, and make the algorithm automatic. For example, a
which are also known as super pixels. Basically homogeneity criterion was introduced in [3], which made
segmentation is used to simplify and/or analyze images the region growing algorithms adaptive for the different
[1] [2]. Image segmentation is typically used to locate locations of initial seeds.
objects and boundaries (lines, curves, etc.) in images. In These algorithms mainly rely on the image
image segmentation process a label is assigned to every intensity information, so they are hard to handle the
pixel in an image and pixels with the same label share partial volume effects and control the leakage.
certain characteristics. Set of segments obtained as a Watershed algorithms are typical example of this
result of image segmentation and these segments algorithm [4], which combines the image intensity with
the gradient information. In this algorithm, gray scale
collectively cover the entire image. Image segmentation
images are considered as reliefs and the gradient
using thresholding was not satisfactory in medical
magnitude of each pixel is treated as elevation.
imaging. Due to the high dimensionality of the image Watershed lines are defined to as the pixels with local
relative to smaller sample sizes direct estimation of the maximum of gradient magnitude. Segmentation
statistical variation of the entire volumetric image was procedure is used to construct watersheds during the
challenged, vascular segmentation was not easily successive flooding of the gray value relief. Watershed
possible and automated reconstruction of cortical algorithms can achieve better results due to the
surface was also the most challenging problem in the combination of image information, but when the images
analysis of human brain Magnetic Resonance Imaging are noisy or the objects themselves have low signal-to-
(MRI). Labeling a histopathology image as having noise ratio these algorithms tend to over-segmentation.
cancerous regions or not was a critical task in cancer Hybrid threshold-based algorithms can further combine
diagnosis. Only 2D models were used, 3D models were with other techniques to perform the segmentation [5].
not compatible in medical imaging. Due to the noise influence and partial volume effect
algorithms based on threshold are seldom used alone
II. THRESHOLDING BASED SEGMENTATION because the edges of organs or structures in medical
images are usually not clearly defined.
Thresholding based segmentation, in which one
threshold value is used to select the area of interest and III. LEARNING BASED SEGMENTATION
this threshold value can be selected by using prior
knowledge or from image information. Further threshold In learning based approach, there may be use of
approach can be edge based, region based or hybrid. In statistical learning, supervised, unsupervised, can be
edge based approach, edge information is required. weakly supervised also. Techniques based upon this
Canny edge detector and Laplacian edge detectors work approach are following
Internatiional Conferennce on Comm
munication, Co
omputing & Syystems (ICCC
CS–2014)
200
Current Methods
M in Meedical Image Segmentation:
S A Review
subject, and
a this method
m provvides statistiically whicch reduces thet ambiguityy in MIL. MCIL and
significant speedup. Cortical
C segmmentation meethod contextual MCIL are able to acchieve comparrable results
LOGISMO OS-B1 consistts of four mainn steps that arre Pre in seegmentation with an apprroach of full pixel-level
processing of raw imagge to create a rough prelim minary supeervision in experiment. T This will insspire future
segmentatiion; graph constructioon; LOGISMOS reseaarch in applyiing different ffamilies of joint instance
segmentatiion; and post--processing foor removal of brain moddels to the framework oof MIL/MCIIL, as the
stem and cerebellum. An A overview of the pipeliine is indeppendence assuumption mighht be loose.
given in Fiig. 1. This meethod offers improved anattomic
accuracy and dramatiically reduced computattional IV. MOD
DEL BASED SE
EGMENTATION
N
requiremennt. A. Statistical Shaape Models foor 3D Medicall Image
D. Weakly
ly Supervised Histopatholoogy Cancer Image
I Segmentationn
Segmeentation and Classification
C Nowadays, model-based
m ssegmentation approaches
Labeliing a histopathology im mage as haaving havee been establiished as one of the mostt successful
cancerous regions or not n is a criticcal task in cancer meth hods for imaage analysis by matching g a model.
diagnosis. It is also clinnically important to segmennt the Mod del contains innformation abbout the expeected shape
cancer tisssues and clusster them intto various claasses. and appearance off the structuree of interest to o images, in
Existing suupervised appproaches for im mage classificcation this segmentationn is conductedd in a top-dow wn fashion.
and segmeentation requiire detailed manual
m annotaations This method is moore stable agaainst local imaage artifacts
for the caancer pixels, which are time-consuminng to and perturbationns than cconventional low-level
obtain. algorrithms. Whille a single template sh hape is an
A neew learning method, Multiple M Clusstered adeqquate model for f industrial applications where
w mass
Instance Learning
L (MC CIL) for histtopathology image prodduced, rigid obbjects need to be detected, this
t method
segmentatiion. MCIL method
m simultaaneously perfforms is prrone to fail in case of biological objects due to their
image-leveel classificatioon (cancer vs. non-cancer im
mage) conssiderable naturral variabilityy [10]. Informmation about
as shown inn Fig. 2. mediical image seggmentation (cancer comm mon variations thus has to be included in the
vs. non-cancer tissue)), and patchh-level clusttering mod del.
(different classes
c [9]. A straight-fforward approach to gather g this
Underr this conceppt, Multiple Instance Leaarning inforrmation is to examine a nuumber of train ning shapes
(MIL) perfforms the aboove three taskks in an integgrated by sttatistical meanns, leading to Statistical Sh
hape Models
frameworkk. In addittion, under this conteextual (SSMMs). Trainingg data for SSM Ms in the medical
m field
constraintss are introducced as a priorr in MCIL, which
w will most likelyy consist off segmented volumetric
further redduces the am mbiguity in MIL.
M Experim
mental imag ges. Dependinng on the seggmentation method used,
results on histopatholoogy colon caancer imagess and the initial represenntation can bee in form of biinary, voxel
cytology images
i demonnstrate the great
g advantagge of data,, fuzzy voxel data (e.g. from m probabilistic methods),
MCIL overr the competinng methods. or suurface meshess. Data originnating from otther sources
of acquisition, e.g. surfacee scanning might be
repreesented diffe ferently. In any case, all shape
repreesentations caan be convertted into each h other, and
the choice of shape repreesentation is the first
funddamental decission when deesigning statisstical shape
mod dels. Most of the subsequeent steps depeend on this
initiaal decision, and many m methods are technically
limitted to certainn representatioons, include constructing
c
shappe models whiich is basicallly extracting of o the mean
shappe and a num mber of moddes of variatiion from a
colleection of trainning samples. Obviously, th he methods
emplloyed stronggly depend on the cho osen shape
repreesentation. Affter this Shappe correspondeence that is
Fig. 2 Cancer and Non Cancer Images
basiccally modelinng the statistiics of a classs of shapes
The addvantages of MCIL
M are eviddent over the state- requuires a set of appropriate
a trraining shapess with well-
of the- artt methods thaat perform thee individual tasks,
t definned corresponndences.
which incllude easing thhe burden of manual annottation Depending on o the chossen represen ntation, the
in which only image-levvel label is reqquired and perrform meth hods of how to best definne these correespondences
image-leveel classificatioon, pixel-levell segmentationn and varyy. In any case, estaablishing deense point
patch-levell clustering simultaneously
s y. In additionn, we correespondences between
b all shapes of the training set
introduce contextual coonstraints as a prior for MCIL M is geenerally the most challennging part of 3D model
201
International Conference on Communication, Computing & Systems (ICCCS–2014)
construction, and at the same time one of the major then weights of this winner neurons can be adjusted by
factors influencing model quality (the other one being using different algorithms.
the local gray-value appearances).After construction the
model is fitted to new, previously unseen data. For this E. SOM and Hybrid SOM Method for Segmentation of
purpose, a model of the appearance of the structure of Brain MRI Image
interest is required to be trained from sample data, due In SOM method, firstly find the winning neuron
to the large size of the search space in 3D, most and secondly updating weight of the neuron and its
methods applied to locate an SSM in new image data neighboring pixels based upon input [12]. Hybrid SOM
use local search algorithms that require an initial combines self organization and topographic mapping
estimate of the model pose. technique.
B. Medical Image Segmentation by Combining Graph F. Markov Random Field (MRF) Model and Fast
Cuts and Oriented Active Appearance Models Fourier Transform (FFT) Based Segmentation for
This method is combination of active appearance Segmentation of Brain MRI Image
model (AAM), Live Wire (LW) and Graph Cuts (GCs) In Markov Random Field model neighborhood
for abdominal 3D segmentation of organs. This method information is used, because most neighborhood pixels
consists of two phases training and segmentation. In are in same class as a result influence of noise
training phase AAM algorithm is constructed and LW decreased [12]. FFT based segmentation used in brain
boundary cost functions and GC parameters are segmentation because in all tumors boundaries between
estimated, and segmentation phase consists two main active and necrotic part are not clear, for this radix 4
parts recognition or initialization and delineation [11]. FFT partitions Discrete Fourier Transform (DFT) into
In the recognition step, a pseudo-3-D initialization four quarter length DFT's of groups of every fourth time
strategy is employed in which the pose of the organs is sample, total computational cost reduced by these FFT
estimated slice by slice via a multi object OAAM outputs which are reused for computing the output.
(MOAAM) method. In the delineation part object shape
information generated from the initialization step is G. Tissue Segmentation in Medical Images Based on
integrated into GC cost computation. Image Processing Chain Optimization
C. Fuzzy C-Mean (FCM) Method for Segmentation of Differential evolution method is purposed to
Brain MRI Image optimize an image processing chain .In this method
training is based upon three sample images provided by
In this method, with the help of Self Organizing an expert [13]. Mainly Differential Evolution(DE)
Map(SOM) clustering algorithm initial cluster centers method is population based optimization method, idea
are selected, after many iterations of this algorithm final behind DE is generating trial parameter vectors, for
cluster centre is obtained. The winning neural units and every vector in the population, DE selects randomly
their corresponding weight vectors from each layer result two other vectors, subtract them and add the weighted
in an abstraction tree. A particular region of the image at difference to randomly chosen third vector(base vector)
a certain level of abstraction is represented with one node to produce mutant vector, cross over rate (user defined
of this abstraction tree [12]. Under this segmentation is value) is used for every vector in mutant population to
performed on demand by transverseing the abstraction control the fraction of parameter values that are copied
tree in Breadth-First Search(BFS) manner starting from from the mutant and target vector to trial vector, if trial
root node until certain criteria is satisfied. If the sum of vector have equal or lower fitness value than that of its
variances of weight vector divided by size of weight target vector, then it replaces the target vector in next
vectors is less than element of weight vector if the size of generation, otherwise target retain its place for at least
abstraction tree is expanded else the node is labeled as one more generation, steps are repeated for every vector
closed node and regions corresponds to closed nodes in population to generate new population.
constitute a segmented image. This method trying to overlap gold images by well
known images generated by experts and images
D. LVQ Method for Segmentation of Brain MRI Image processed with this technique.
Linear Vector Quantization (LVQ) technique is V. AUTOMATIC FUZZY APPROACH
supervised learning technique obtain decision FOR SEGMENTATION
boundaries based upon training data .In this method
A. MRI Brain Image Segmentation by Fuzzy Symmetry
three layers are there input, competitive and output
Based Genetic Clustering Technique
layer [12]. Input data is classified in the competitive
layer and then those classes or patterns are mapped to Automatic segmentation technique of MRI of brain
the target class in the output layer, under this winner using new fuzzy point symmetry based genetic
neuron is selected based upon the Euclidean distance clustering technique is proposed, which is able to
202
Current Methods in Medical Image Segmentation: A Review
evolve the number of clusters present in the data set [3] R. Pohle and K. D. Toennies, “Segmentation of medical images
using adaptive region growing, ” in Medical Imaging 2001.
automatically [14]. In this assignment of points to International Society for Optics and Photonics, 2001, pp. 1337–
clusters are based on point symmetry based distance 1346.
rather than the Euclidean distance and because of this, [4] L. Najman and R. Vaillant, “Topological and geometrical
proposed algorithm Fuzzy Variable string length corners by watershed, ” in Computer Analysis of Images and
Patterns. Springer, 1995, pp. 262–269.
Genetic Point Symmetry(Fuzzy VGAPS) enable to [5] S. A. Mani, W. Guo, M.-J. Liao, E. N. Eaton, A. Ayyanan, A. Y.
identify any type of cluster irrespective of its shape size Zhou, M. Brooks, F. Reinhard, C. C. Zhang, M. Shipitsin et al.,
convexity and this method automatically evolve the “The epithelial-mesenchymal transition generates cells with
clusters. properties of stem cells, ” Cell, vol. 133, no. 4, pp. 704–715,
2008.
VI. CONCLUSION [6] G. Erus, E. I. Zacharaki, and C. Davatzikos, “Individualized
statistical learning from medical image databases: Application
In this paper, various segmentation techniques to identification of brain lesions, ” Medical image analysis, vol.
18, no. 3, pp. 542–554, 2014.
applied for medical images are briefly explained. All [7] D. Lesage, E. D. Angelini, I. Bloch, and G. Funka-Lea, “A
the discussed techniques overcome various limitations review of 3d vessel lumen segmentation techniques: Models,
occurred in medical image segmentation like direct features and extraction schemes, ” Medical image analysis, vol.
estimation of the statistical variation of the entire 13, no. 6, pp. 819– 845, 2009.
[8] I. Oguz and M. Sonka, “Logismos-b: Layered optimal graph
volumetric image, vascular segmentation and in the image segmentation of multiple objects and surfaces for the
analysis of human brain magnetic resonance brain, ” IEEE transaction, vol. 33, p. 6, 2014.
imaging(MRI) automated reconstruction of cortical [9] Y. Xu, J.-Y. Zhu, E. I. Chang, M. Lai, Z. Tu et al., “Weakly
surface was the most challenging problem. Labeling a supervised histopathology cancer image segmentation and
classification, ” Medical image analysis, vol. 18, no. 3, pp. 591–
histopathology image as having cancerous regions or 604, 2014.
not was a critical task in cancer diagnosis. 3D models [10] T. Heimann and H.-P. Meinzer, “Statistical shape models for 3D
were not compatible in medical imaging. Performance medical image segmentation: A review, ” Medical image
of these medical image segmentation techniques can be analysis, vol. 13, no. 4, pp. 543–563, 2009.
[11] X. Chen, J. K. Udupa, U. Bagci, Y. Zhuge, and J. Yao, “Medical
improved in future by incorporating many more images image segmentation by combining graph cuts and oriented
from healthy subjects into the models, by the design of active appearance models, ” Image Processing, IEEE
new sequential combinations of different methods, for Transactions on, vol. 21, no. 4, pp. 2035–2046, 2012.
more optimization many algorithm can be added in pre [12] S. Bandhyopadhyay and T. U. Paul, “Segmentation of brain
MRI image–a review, ” International Journal of Advanced
and post processing phases, combination of graph cut Research in Computer Science and Software Engineering, vol.
and OAAM can be improved by introducing 2, no. 3, pp. 409– 413, 2012.
parallelization and by incorporating the spatial [13] S. Rahnamayan and Z. Mohamad, “Tissue segmentation in
information. medical images based on image processing chain optimization, ”
in International Workshop on Real Time Measurement,
REFERENCES Instrumentation and Control, Toronto, 2010, pp. 1–9.
[14] S. Saha and S. Bandyopadhyay, “MRI brain image segmentation
[1] L. Barghout and L. Lee, “Perceptual information processing by fuzzy symmetry based genetic clustering technique, ” in
system, ” 2003. Evolutionary Computation, 2007. CEC 2007. IEEE Congress
[2] G. Srinivasan and G. Shobha, “An overview of segmentation on. IEEE, 2007, pp. 4417–4424.
techniques for target detection in visual images, ” in
Proceedings of the 9th WSEAS International Conference on
International Conference on Automation and Information.
World Scientific and Engineering Academy and Society
(WSEAS), 2008, pp. 511–518
203