Survey on Brain MRI Segmentation Techniques

Scientific Journal Impact Factor (SJIF): 1.711
International Journal of Modern Trends in Engineering
and Research
www.ijmter.com
@IJMTER-2014, All rights Reserved 187
e-ISSN: 2349-9745
p-ISSN: 2393-8161
Survey on Brain MRI Segmentation Techniques
Merlyn Mary Michael1
1
Computer Science Department, Amal Jyothi College Of Engineering
Abstract: Image segmentation is aimed at cutting out, a ROI (Region of Interest) from an image. For
medical images, segmentation is done for: studying the anatomical structure, identifying ROI ie tumor
or any other abnormalities, identifying the increase in tissue volume in a region, treatment planning.
Currently there are many different algorithms available for image segmentation. This paper lists and
compares some of them. Each has their own advantages and limitations.
Keywords: Segmentation, thresholding, region based, pixel classification, model based
I. INTRODUCTION
Currently, the Medical Imaging Technology provides the clinician with a number of complementary
diagnostic tools such as Computed Tomography (CT), Ultrasound, X-ray, Magnetic Resonance
Imaging (MRI) and Positron Emission Tomography (PET). From the different existing imaging
modalities, magnetic (MRI) is the one which is most widely used, (especially for the brain). This is
because, MRI images have a high spatial resolution, high signal to noise ratio and can do an excellent
discrimination of soft tissues of the body and can thus provide detailed information about human
anatomical structure.
Radiologists perform visual and qualitative analysis of these medical images. However, for detailed
diagnosis and treatment, an elaborate quantitative analysis is required and to facilitate this assessment,
Computer aided diagnosis is inevitable. CAD systems are effective due to the below factors.
 Less impact on diagnosis results from manual mistakes / missing steps
 Availability of fast and accurate results.
 Faster communication even to distant areas is possible.
Image segmentation is the major step for image analysis and computer vision. Image Segmentation is
the process of partitioning an image into non - overlapping, constituent regions which are identical
with respect to some characteristics such as intensity, color and texture [2]. During these years many
significant advances have been made by researchers in the field of brain segmentation. Both semi-
automated and automated methods have been proposed.
Brain segmentation methods are categorized according to the required extend of human interaction as:
a) manual segmentation b) semi-automated segmentation c) fully automated segmentation [5].
Section 2 explains the most relevant existing brain MRI segmentation methods. The paper is concluded
in Section 3. Finally, in Section 4, a comparison is done between the different segmentation techniques.
II. DIFFERENT SEGMENTATION TECHNIQUES
Major existing segmentation methods are discussed in the below sections.
According to [1], the fundamental roles of segmentation are to: (1) permit quantification, (2) reduce
the dataset by focusing the quantitative analysis on the extracted regions that are of interest, and (3)
establish structural correspondences for the physiological data sampled within the regions.
In general, segmentation techniques have been divided into four major classes:
• Threshold-based techniques

International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 01, Issue 06, [December - 2014] e-ISSN: 2349-9745, p-ISSN: 2393-8161
• Region-based techniques
• Pixel classification techniques
• Model-based techniques
Threshold-based, region-based and pixel classification techniques are generally used for two-
dimensional image segmentation [6]. Model based techniques such as parametric and geometric
deformable models (level sets), are mostly employed in volumetric (3D) image segmentation [7].
A. Threshold based segmentation
Thresholding is a simple and effective region segmentation method, in which the objects of the image
are classified by comparing their intensities with one or more intensity thresholds.
These thresholds can be either global or local. If the histogram of an image expresses a bimodal pattern,
the object can be separated from the background in the image by a single threshold called global
thresholding. However, if the image contains more than two types of regions, corresponding to
different objects, the segmentation must be carried out using local thresholding. Such images are
segmented by applying several individual thresholds or by using a multithresholding technique. A local
threshold is determined adaptively in a local region around a pixel. According to Yao [4] the values of
thresholds are generally estimated based on prior knowledge. Local threshold values are usually
estimated using the local statistical properties [6] or by calculating partial volumes of each region [8]
or by the Gaussian distribution of data values as threshold in normal brain T2-weighted MRI [9].
B. Region based segmentation
Region-based segmentation approaches examine pixels in an image and form disjoint regions by
merging neighborhood pixels with homogeneity properties based on a predefined similarity criterion
[1]. These methods can be sketched in general as follows: Let X be an image that is segmented into
N regions, each of which is denoted as Ri where i = 1, 2… N. The original image can be exactly
assembled by putting all regions together and there should be no overlapping between any two regions
Ri and Rj for i ≠ j. The logical predicate L (⋅) contains a set of rules (usually a set of homogeneity
criteria) that must be satisfied by all pixels within a given region, and it fails in the union of two regions
since merging two distinct regions will result in an inhomogeneous region. The regions must satisfy
the following properties:
R 1Ụ R 2 Ụ R 3 Ụ… Ụ R i = I
where R 1 , R 2 , R 3 ,…R i are the region in the image I,
and further, R 1∩ R 2 ∩ R 3 ∩… ∩ R i = 0
This is as per the set theory of homogeneity.
Region growing and watershed methods are common region based segmentation techniques. The
simplest region-based segmentation technique is the region growing, which is used to extract a
connected region of similar pixels from an image [10]. In watershed method, to avoid over-
segmentation some pre or post processing methods are used in order to produce a more reasonable
segmentation that reflects the layout of objects [11,12]
C. Pixel classification based segmentation
Another type of segmentation method is based on pixel classification. Pixels in an image can be
represented in feature space using pixel attributes that may consist of gray level, local texture, and
color components for each pixel in the image. Segmentation methods based on pixel classification are
constrained to the use of supervised or unsupervised classifiers to cluster pixels in the feature space.

Clustering is the process of grouping similar objects into a single cluster, while objects with dissimilar
features are grouped into different clusters based on some similarity criteria. The similarity is
quantified in terms of an appropriate distance measure. Common distance measures used are:
Euclidean distance or cosine measure. Many clustering techniques like Fuzzy CMeans (FCM), Markov
Random Fields (MRF) (unsupervised), Artificial Neural Networks (ANN) (supervised) are available.
In FCM method, for brain segmentation, the tissue classes are first identified and then pixels are
assigned membership values (0-1) to the tissue classes according to its attributes (intensity, texture
etc.). MRF provides a way to integrate spatial information into the clustering process. Markov Random
Fields are widely used not only for modeling classes of segmentation, but also to model texture
properties and in homogeneities of the intensities. In ANN, the classifier feeds the features through a
series of nodes, where mathematical operations are applied to the input nodes and a classification is
made at the final output nodes. Although ANN training is complex, it has the ability to model non-
trivial distributions.
D. Model based segmentation
The segmentation of volumetric (3D) image data is a challenging problem that has been mainly
approached by model based segmentation techniques such as parametric deformable models and
geometric deformable models or level sets. In model-based segmentation, a connected and continuous
model is built for a specific anatomic structure by incorporating a priori knowledge of the object such
as shape, location, and orientation. Some models incorporate prior statistical information drawn from
a population of training datasets [4]. The statistical parameterization provides global constraints and
allows the model to deform only in ways implied by the training sets. Segmenting structures from
medical images and reconstructing a compact geometric representation of these structures are difficult
due to the sheer size of the datasets and the complexity and variability of the anatomic shapes of interest
[13].Existing deformable models can be broadly divided into two categories: parametric and
geometric. The parametric deformable models (also known as active contour models or snakes) have
the ability to segment, match, and track images of anatomic structures by exploiting constraints derived
from the image data together with a priori knowledge about the location, size, and shape of these
structures. But they are incapable of naturally handling topological changes in the image, while
geometric deformable models or level sets have that capability.
III.CONCLUSION
This paper has surveyed the principle methods for brain MRI image segmentation. A brief description
of these methods along with their key advantages and limitations has been narrated above.
Even though there is an availability of latest segmentation techniques, medical image segmentation
still remains as the most challenging area for research in the field of image processing, because medical
images are basically very complicated (example the structure of brain) and a precise segmentation is
quintessential for effective medical diagnosis and treatment. The majority of research in medical image
segmentation pertains to the use for MR images, especially in brain imaging, because of its ability to
derive contrast from a number of tissue parameters. Moreover different pulse sequences exist for
acquiring MR images. Hence to determine the optimal pulse sequence for obtaining accurate
segmentations is another important task which requires knowledge about the underlying tissue
properties of the anatomy to be segmented. Thus image segmentation is still the most challenging
research area in the field of image processing.

IV. COMPARISONS
Table 1. Comparison of Segmentation Methods based on level of human interaction
Comparison
Factor
Manual Segmentation Semi Automated Fully Automated
Level of
Human
Interaction
involved
- involves manually drawing
the boundaries of the
structures of interest, or
painting the region of
anatomic structures with
different labels
- includes the intervention of
a human operator. Role of the
human operator is to:
initialize the method, check
the accuracy of the
result,manually correct the
segmentation result.
- does not involve human
interaction, the computer
system or program determines
the segmentation
Application - is still widely used in areas
where a lot of human
knowledge and expertise is
required.
- is used as a golden standard
against which semi and fully
automated methods are
validated
- uses different strategies to
combine computers and
humans’ expertise to obtain
an efficient brain
segmentation
- can provide an efficient
segmentation for a large batch
of images
Limitation - if the person drawing ROI
is not a
radiologist/anatomist/trained
technologist who is well
versed with brain anatomy,
segmentation results will be
poor.
- may have inter or intra rater
variability
-May be subject to variations
both between expert users
and within the same user.
- not widely accepted
-lacks interpretability and
transparency in the
segmentation process.
- developing the fully
automated system is extremely
difficult(as high-level visual
processing, and specialized
domain knowledge is
required)[7]
Table 2. Comparison of major segmentation methods
Sample
technique or
Algorithm
Major Advantage Major Constraint Impact of Noise Computational
time& memory
requirement
Threshold based Global
thresholding,
- simple and fast
segmentation is
done when good
-Threshold
selection
becomes difficult
Less immune to
noise than other
techniques
Simple and
computationally
fast

Local
Thresholding
threshold values
are defined.
-Global
thresholding
performs well if
the image
contains objects
with
homogeneous
intensity or the
contrast between
the objects and
the background is
high.
with increasing
number of
regions or
noise levels, or
low contrast in
image.
- Unable to
exploit all the
information
provided
by MRI, so
mostly used as a
first step in the
segmentation
process.
Region Based Region
growing,
Watershed
Work best when
the region
homogeneity
criterion is easy
to define.
- Region growing
has inherent
dependence on
the selection of
seed region and
the order in
which pixels and
regions are
examined;
- generally these
approaches are
constrained
to be semi-
automatic
- Noise or
variation of
intensity may
result in holes or
over-
segmentation.
Are by nature
sequential and quite
expensive both in
computational time
and memory;
Pixel Based Fuzzy C
Means,
ANN, MRF
-Unsupervised
technique - FCM,
permits the use of
vague concepts
in the definition
of clusters
-Unsupervised
method - MRF
can represent
complex
FCM-The
determination of
fuzzy
membership is a
difficultl job
MRF- Difficult
to select
parameters that
control
FCM is
generally
sensitive to
noise
For ANN -Training
time is long;
All the methods are
computationally
intensive

dependencies
among data
-Supervised
clustering
method of ANN
has the ability to
model non-trivial
distributions
the strength of
spatial
interactions.
ANN- gathering
of training
samples is not
easy. It can affect
the result
Deformable
Model Based
Parametric
deformable
models,
Level Sets
-can
accommodate the
variability of
biological
structures over
time and across
different
individuals [13]
-Can be
implemented as a
fully automated
technique
The model may
converge to
wrong
boundaries in
case of in
homogeneities
[3].
Less immune to
noise
computationally
intensive
REFERENCES
[1] Wong K. Medical image segmentation: methods and applications in functional imaging. Handb Biomed Image Anal
Segmentation Models Part B 2005;2: 111–82.
[2] Bhandarkar S, Koh J, Suk M. Multiscale image segmentation using a hierarchical self-organizing map. Neurocomputing
1997;14:241–72.
[3] Luo S, Li R, Ourselin S. A new deformable model using dynamic gradient vector flow and adaptive balloon forces.
APRS Workshop on Dig Image Comp; 2003.p. 9–14.
[4] Yao J. Image processing in tumor imaging. New techniques in oncologic imaging; 2006. p. 79–102.
[5] Foo JL. A survey of user interaction and automation in medical image segmentation methods. Tech rep ISUHCI20062,
Human Computer Interaction Department, Iowa State Univ; 2006.
[6] Shanthi KJ, Sasi Kumar M. Skull stripping and automatic segmentation of brain MRI using seed growth and threshold
techniques. Conf Intelligent Adv Sys 2007:422–6.
[7] Chang H, Valentino D. An electrostatic deformable model for medical image segmentation. Comput Med Imaging
Graph 2008;32:22–35
[8] ChangSun Y, SunHan K, JunSong C, MooNoh S, WonPark J. Threshold estimation for region segmentation on MR
image of brain having the partial artifact.ICSP2000 2000:1000–9.
[9] Stadlbauer A, Moser E, Gruber S, Buslei R, Nimsky C, Fahlbusch R, et al. Improved delineation of brain tumors: an
automated method for segmentation based on pathologic changes of 1H-MRSI metabolites in gliomas. Neuroimage
2004;23:454–61.
[10] Adams R, Bischof L. Seeded region growing. IEEE Trans Pattern Anal Mach Intell 1994;16:641–7.
[11] Bleau A, Leon L. Watershed-based segmentation and region merging. Comput Vis Image Under 2000;77(3):317–70.
[12] Gies V, Bernard T. Statistical solution to watershed over-segmentation. Int Conf Image Process; 2004. p. 1863–6.
Image Technol 2004:207–11.
[13] McInerney T, Terzopoulos D. Deformable models. Handbook of medical imaging processing and analysis. San Diego,
CA: Academic Press; 2000. p. 127–45.

Survey on Brain MRI Segmentation Techniques

Survey on Brain MRI Segmentation Techniques

More Related Content

What's hot (18)

Similar to Survey on Brain MRI Segmentation Techniques (20)

More from Editor IJMTER (20)

Recently uploaded (20)

Survey on Brain MRI Segmentation Techniques