A Review on Label Image Constrained Multiatlas Selection

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 03 | Mar -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 801
A Review on Label Image Constrained Multiatlas Selection
Ms. VAIBHAVI NANDKUMAR JAGTAP1, Mr. SANTOSH D. KALE2
1PG Scholar , Department of Electronics and Telecommunication, SVPM College of Engineering, Malegaon (BK),
Maharashtra, India.
2Assistant Professor , Department of Electronics and Telecommunication, SVPM College of Engineering, Malegaon
(BK), Maharashtra, India.
--------------------------------------------------------------------------------------------------------------------------------------------
Abstract: Multiatlas selection method is a powerful and
frequently used for medical image segmentation when a
standard atlas is available. In recent era of image
segmentation manifold ranking based methods are very
useful. Magnetic resonance imaging (MRI) provides an
attractive way for confined mapping of the anatomical
structure of the thing. In multiatlas based image
segmentation, the two key factors affecting the
performance are atlas selection and combination. It is
difficult to receive the correct atlas selection result due to
prostate structure in raw images. In this paper we try to
solve the problem of manual segmentation by proposing a
new approach for MRI image segmentation using
multiatlas selection method which is automatic method
useful for diagnosis of prostate.
Keywords: Atlas selection, computer vision,
normalization, image segmentation, manifold
ranking.
1. INTRODUCTION
In American men the second cause of death is
prostate cancer. The existing methods are effective in
some cases; automated segmentation of prostate MR
image is still very challenging due to the unclear
boundary information in some areas. Atlas-based
segmentation has become a standard method for
automatically labeling regions of interest (ROIs) on MR
brain images. In the method of atlas-based segmentation,
the atlas is registered to the individual brain image by
discovering the best spatial alteration, and then mapping
the structural information in the atlas on the individual
brain image.
Fig.1. [1] Left: original MRI image with two red arrows to
demonstrate the weak border.
Right: prostate segmentation detected by an expert with
red curve.
Fig.1 shows the red curve in the MR image on
the right is a segmentation of the prostate detected by an
expert. It can be seen that the boundary is very weak in
the areas indicated in the original MR image by the two
red arrows in the left of Fig. 1. Experts fragment these
areas mainly according to their knowledge of the
anatomical structure of the prostate. Therefore, it is
important to use the anatomical knowledge in the
automated methods. The atlas essentially depicts the
shapes and locations of anatomical structures and
together with the spatial relationships between them.
Thus, atlas based segmentation is one of the most

common techniques applied to the automatic
segmentation of the prostate MRI image. Generally, an
atlas consists of a raw image and label image. In the
process of multiatlas selection method, each atlas is first
registered to the target image, which will generate
deformed atlas that close to the image to be segmented.
Based on certain of selection criteria a subset of atlases
is selected from the deformed atlases. Finally, the
selected atlases are combined into a single binary
template for segmentation. The approach of atlas
selection is one of the most vital factors affecting the
correctness of segmentation in all the three steps of
multiatlas based method. Besides that, atlas combination
is another important ingredient, where assigning the
proper weight for each selected atlas is a crucial factor.
2. RELATED WORKS
2.1 LITERATURE SURVEY:
A. LEAP: Learning embeddings for atlas propagation, R.
Wolz, P. Aljabar, J. Hajnal, A. Hammers, and D.
Rueckert,[2] proposed a novel framework for the
automatic propagation of a set of manually labeled
brain atlases to a diverse set of images of a
population of subjects. A manifold is learned from a
coordinate system embedding that allows the
identification of neighborhoods which contain
images that are similar based on a chosen criterion.
Within the new coordinate system, the initial set of
atlases is propagated to all images through a
succession of multi-atlas segmentation steps. This
breaks the problem of registering images that are
very “dissimilar” down into a problem of registering
a series of images that are “similar”. At the same
time, it allows the potentially large deformation
between the images to be modeled as a sequence of
several smaller deformations.
B. Construction and validation of mean shape atlas
templates for atlas-based brain image segmentation,
Q. Wang,[3] evaluate different schemes for
constructing a mean shape anatomical atlas for
atlas-based segmentation of MR brain images. Atlas
construction and atlas based segmentation are
performed by non-rigid intensity-based registration
using a viscous fluid deformation model with
parameters that were optimally tuned for this
particular task. Firstly, other atlas construction
schemes could have been included. In for instance, a
mean shape template was constructed for the bee
brain by first performing an affine registration of all
images to a common reference image, followed by
non-rigid warping of the same set of images to the
intensity averaged image obtained after affine
registration. This process was iterated until
convergence using the average brain from a
previous iteration as template. However, for human
brain images that show substantial inter-subject
variability, the averaging of affinely co-registered
images introduces significant blurring that is likely
to affect the accuracy of subsequent NRR steps. The
segmentation performance of each atlas scheme is
evaluated on the same database using a leave-one-
out approach and measured by the volume overlap
of corresponding regions in the ground truth manual
segmentation and the warped atlas label image.
C. Manifold learning for biomarker discovery in MR
imaging, R. Wolz, P. Aljabar, J. Hajnal, and D.
Rueckert,[4] presented a method for deriving
biomarkers from low-dimensional manifolds which
are learned from variations in inter-subject
appearance in a dataset at baseline and from intra-
subject changes over time. Laplacian eigen maps
were used to nonlinearly embed images based on
pair-wise similarities. In the embedded space, we
estimated a separating hyper plane from labeled
images to classify test data into groups of clinical
interest. Applying the framework to similarities

based on a region of interest (ROI) around the
hippocampus in images taken from a large, multi-
center AD study (ADNI) leads to classification results
as good as those obtained from automatically
determined hippocampal volume and atrophy. They
have proposed a fast and robust alternative to
classify subjects that is generic, data-driven and easy
to extend to different structures or areas in the brain
without the need for complex priors. The embedding
of subjects in a continuous manifold space, however,
allows an application beyond classification. In
particular, the correlation between clinical variables
and subjects’ condition with coordinates in the
embedded space needs to be further investigated to
potentially explore other ways of inferring a
subject’s state.
D. Optimum template selection for atlas-based
segmentation, Minjie Wu,CaterinaRosano,Pilar
Lopez-Garcia,Cameron S. Carter,and Howard J.
Aizensteind,[5] addressed atlas selection with a
different approach: instead of choosing a fixed brain
atlas, we use a family of brain templates for atlas-
based segmentation. For each subject and each
region, the template selection method automatically
chooses the ‘best’ template with the highest local
registration accuracy, based on normalized mutual
information. The region classification performances
of the template selection method and the single
template method were quantified by the overlap
ratios (ORs) and intra class correlation coefficients
(ICCs) between the manual tracings and the
respective automated labeled results.
2.2 PROPOSED SYSTEM
Proposed system is based on the idea to
constrain the computation of the affinity matrix of raw
images to employ label images in constructing the lower-
dimensional manifold space, as shown in Fig. 2(b).
Fig. 2. [1] (a) Misleading manifold projection due to the
influence of other anatomical structures in raw images.
(b) Manifold projection constrained by the label images
to reduce the influence and preserve the neighborhood
structure.
When learning the manifold ranking the region
of interest information from the label images is exploited
together with the raw images. The intrinsic similarity
between the target regions can be exposed in the lower-
dimensional manifold space due to the constraint. In this
space, the atlases are closer to the test image are
selected in terms of the regions of interest (ROI), and
then the final fused atlas template can recover the
performance of the segmentation.
A new weight assignment method for the atlas
combination is also proposed based on this manifold
subspace analysis. The label images and the raw images
of the selected atlases can be assumed to share an
identical manifold structure in the lower-dimensional
space. As a result, weight assignment for the elected
label images can be measured as calculating the weights
for the conversion of the data points of raw images in the
manifold space. Then, the computed transformation
weights can be mapped for label images for combination.
3. SYSTEM FRAMEWORK:
The block diagram of the label image constrained
atlas selection method is shown in Fig.3. The proposed
method consists of three main steps: registration,
selection and combination. The different raw MR images
of atlases within the same dynamic range is done using

normalization step. After normalization, the next stage is
registration that each raw image of atlases maps to the
test image. In transformation stage each atlas is warped
to the test image, generating the deformed atlas. The
next stage is atlas selection. The atlas selection method
should measure the similarity between only the regions
of interest across images. Thus, manifold ranking should
not only preserve the neighborhood of the original
manifold of raw images, but also consider the intrinsic
similarity between the regions of interest. Final step of
system is segmentation.
Fig.3. Block diagram of label image constrained
multiatlas selection
Normalization
In image processing, normalization is a process that
changes the range of pixel intensity values. Applications
include photographs with poor contrast, for example due
to frown. Normalization is sometimes called contrast
stretching or histogram stretching. It is referred as
dynamic range expansion in more common fields of data
processing, such as digital signal processing. The
purpose of dynamic range expansion in the various
applications is usually to get the image, or new kind of
signal, into a range that is familiar or usual to the
intellect, hence the term normalization. Often, the
motivation is to achieve consistency in dynamic series
for a set of signals, data, or images to avoid mental
disturbance or weakness. In order to make the different
raw MR images of atlases within the same dynamic
range, the classical z-score normalization is done.
Registration method
Registration is the determination of a geometrical
transformation that aligns points in one view of an object
with corresponding points in another view of that object
or another object. Atlas-based segmentation labels the
anatomical regions on individual images by registering
the template image of the brain to the individual brain
image. After image normalization, each normalized raw
image of atlases is aligned to the normalized test image.
With the parameters of transformation yielded from the
alignment, each atlas is warped to the test image,
generating the deformed atlas.
Atlas Selection
Ideally, the atlas selection method should measure the
similarity between only the regions of interest across
images. A new manifold projection method is developed
by taking the label image information into account for
selecting atlases on a lower-dimensional manifold for
image segmentation, which has been overlooked by
other existing methods. Atlas-selection is an important
step affecting the accuracy of MAS. On one hand, using
small subsets of atlases may produce less accurate
segmentations as some related information from other
atlases possibly left out. On the other hand, using large
subsets of atlases may undermine the segmentation
performance because of the high amount of irrelevant
information introduced.
Subspace Analysis for Weighted Combination
In atlas combination step, weight assignment for the
selected atlases is also an important factor affecting the
segmentation performance. In this system, the weight
computation is based on the lower-dimensional manifold
subspace. The objective of combination is to make the
result near to the ground truth as possible. Specifically,

the goal is to reduce the difference between the ground
truth and the combination result. The atlas combination
weights are computed by solving a problem of
reconstruction of data points in the manifold subspace.
3.2 ADVANTAGES
 It is an automatic method of medical image
segmentation.
 This method improves the performance of atlas selection
and final segmentation.
 It reduces the influence of surrounding anatomical
structures in raw images.
 MAS have been successfully used in a large variety of
biomedical segmentation problems
3.3 APPLICATIONS
 Biomedical segmentation problem.
 Segmentation of cortical and sub cortical regions in
structural images.
 Skull stripping and tissue classification, the
segmentation of tumors, eyes and optic nerves.
 Segmentation of brain MRI in animal studies.
 MAS have also been used in abdominal imaging.
4. CONCLUSION
A novel method is proposed for prostate cancer
identification. In order to analyze a disease, Physicians
consider MR imaging modality is the most efficient one
for identification of cancer present in a variety of organs.
Consequently, analysis on MR imaging is necessary for
efficient disease diagnosis. In this system, we proposed a
novel manifold learning based atlas selection method
and a novel weight computation algorithm for atlas
combination in multiatlas based segmentation. In atlas
selection step, it employs the label images to constrain
the manifold projection to reduce the influence of
surrounding anatomical structures in raw images.
REFERENCES
[1] Pingkun Yan, Yihui Cao, Yuan Yuan, Baris Turkbey,
and Peter L. Choyke, “Label Image Constrained Multiatlas
Selection,”IEEE TRANSACTIONS ON CYBERNETICS, VOL.
45, NO. 6, JUNE 2015.
[2] R. Wolz, P. Aljabar, J. Hajnal, A. Hammers, and D.
Rueckert, “LEAP: Learning embeddings for atlas
propagation,” Neuroimage, vol. 49, no. 2, pp. 1316–1325,
2010.
[3] Q. Wang et al., “Construction and validation of mean
shape atlas templates for atlas-based brain image
segmentation,” in Information Processing in Medical
Imaging. Berlin, Germany: Springer, 2005, pp. 689–700.
[4] R. Wolz, P. Aljabar, J. Hajnal, and D. Rueckert,
“Manifold learning for biomarker discovery in MR
imaging,” in Machine Learning in Medical Imaging.
Berlin, Germany: Springer, 2010, pp. 116–123.
[5] Minjie Wu,CaterinaRosano,Pilar Lopez-
Garcia,Cameron S. Carter,and Howard J. Aizensteind,
“Optimum template selection for atlas-based
segmentation”,© 2006 Elsevier Inc.
[6] P. Yan, S. Xu, B. Turkbey, and J. Kruecker, “Discrete
deformable model guided by partial active shape model
for TRUS image segmentation,” IEEE Trans. Biomed. Eng.,
vol. 57, no. 5, pp. 1158–1166, May 2010.
[7] Juan Eugenio Iglesias1 and Mert R. Sabuncu, “Multi-
Atlas Segmentation of biomedical Images A Survey,” June
12, 2015.
BIOGRAPHIES
Ms. VAIBHAVI NANDKUMAR JAGTAP
Pursing Master Engineering (M.E.E&TC),
From SVPM College of Engineering Malegaon (BK).

Santosh D. Kale:- Currently
working as a Assistant Professor
at college of engineering,
Malegaon (Bk), Baramati. He
received B.E. Degree in
Electronics & Telecommunication
in 2001, from North Maharashtra University of Jalgaon,
Maharashtra, India. He received M.Tech Degree in
(Electronics Instrumentation) in Electronics &
Telecommunication, from college of Engineering, Pune
(COEP), India. He guided several UG & PG projects. His
research area includes signal and image processing.

A Review on Label Image Constrained Multiatlas Selection

More Related Content

What's hot (20)

Similar to A Review on Label Image Constrained Multiatlas Selection (20)

More from IRJET Journal (20)

Recently uploaded (20)

A Review on Label Image Constrained Multiatlas Selection