Automated segmentation of breast in 3 d mr images using a robust atlas

OUR OFFICES @ CHENNAI/ TRICHY / KARUR / ERODE / MADURAI / SALEM / COIMBATORE / BANGALORE / HYDRABAD
CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
IEEE SOFTWARE | EMBEDDED | MECHANICAL | ROBOTICS PROJECTS DEVELOPMENT
Visit: www.ecwaytechnologies.com | www.ecwayprojects.com Mail to: ecwaytechnologies@gmail.com
AUTOMATED SEGMENTATION OF BREAST IN 3-D MR IMAGES
USING A ROBUST ATLAS
By
A
PROJECT REPORT
Submitted to the Department of electronics &communication Engineering in the
FACULTY OF ENGINEERING & TECHNOLOGY
In partial fulfillment of the requirements for the award of the degree
Of
MASTER OF TECHNOLOGY
IN
ELECTRONICS &COMMUNICATION ENGINEERING
APRIL 2016

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
CERTIFICATE
Certified that this project report titled “Automated Segmentation of Breast in 3-D MR Images
Using a Robust Atlas” is the bonafide work of Mr. _____________Who carried out the research
under my supervision Certified further, that to the best of my knowledge the work reported herein
does not form part of any other project report or dissertation on the basis of which a degree or
award was conferred on an earlier occasion on this or any other candidate.
Signature of the Guide Signature of the H.O.D
Name Name

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
DECLARATION
I hereby declare that the project work entitled “Automated Segmentation of Breast in 3-D MR
Images Using a Robust Atlas” Submitted to BHARATHIDASAN UNIVERSITY in partial
fulfillment of the requirement for the award of the Degree of MASTER OF APPLIED
ELECTRONICS is a record of original work done by me the guidance of Prof.A.Vinayagam
M.Sc., M.Phil., M.E., to the best of my knowledge, the work reported here is not a part of any
other thesis or work on the basis of which a degree or award was conferred on an earlier occasion
to me or any other candidate.
(Student Name)
(Reg.No)
Place:
Date:

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
ACKNOWLEDGEMENT
I am extremely glad to present my project “Automated Segmentation of Breast in 3-D MR
Images Using a Robust Atlas” which is a part of my curriculum of third semester Master of
Science in Computer science. I take this opportunity to express my sincere gratitude to those who
helped me in bringing out this project work.
I would like to express my Director,Dr. K. ANANDAN, M.A.(Eco.), M.Ed., M.Phil.,(Edn.),
PGDCA., CGT., M.A.(Psy.)of who had given me an opportunity to undertake this project.
I am highly indebted to Co-OrdinatorProf. Muniappan Department of Physics and thank from
my deep heart for her valuable comments I received through my project.
I wish to express my deep sense of gratitude to my guide
Prof. A.Vinayagam M.Sc., M.Phil., M.E., for her immense help and encouragement for
successful completion of this project.
I also express my sincere thanks to the all the staff members of Computer science for their kind
advice.
And last, but not the least, I express my deep gratitude to my parents and friends for their
encouragement and support throughout the project.

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
ABSTRACT:
This paper presents a robust atlas-based segmentation (ABS) algorithm for segmentation
of the breast boundary in 3-D MR images. The proposed algorithm combines the well-known
methodologies of ABS namely probabilistic atlas and atlas selection approaches into a single
framework where two configurations are realized. The algorithm uses phase congruency maps to
create an atlas which is robust to intensity variations. This allows an atlas derived from images
acquired with one MR imaging sequence to be used to segment images acquired with a different
MR imaging sequence and eliminates the need for intensity-based registration. Images acquired
using a Dixon sequence were used to create an atlas which was used to segment both Dixon images
(intra-sequence) and T1-weighted images (inter-sequence). In both cases, highly accurate results
were achieved with the median Dice similarity coefficient values of and , respectively.

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
INTRODUCTION:
Magnetic resonance imaging (MRI) of breast is an increasingly common approach for
monitoring and detection of breast cancer mainly due to higher sensitivity and no ionizing radiation
compared to conventional X-ray mammography. MR imaging is now also recommended for
screening women who are known to be at a higher risk of breast cancer . The segmentation of
breast tissue in MR images has a number of important applications. The removal of unwanted
structures such as the chest wall and the heart makes it possible to improve the quality of 3-D
visualization. In order to use MRI to assess breast density, a significant risk factor and an important
biomarker in determining the possibility of breast cancer, it is essential to determine the total
volume of the breast. In computer-aided diagnosis (CAD), breast segmentation allows the software
to ignore tissue outside of the breast, reducing the false positive rate
Finally, in order to plan therapeutic interventions such as radiotherapy using MRI, accurate
segmentations of the target organs will be required. Atlas-based segmentation (ABS) is a well
established and widely used technique for extracting contours from medical images of different
organs. In this method, processed images are stored in a database called a set of atlases. Each atlas
consists of an image and its gold standard (i.e., manual segmentation or label). The atlas images
are first registered to the target image and then the resulting transformations are applied to the
labels. Finally, the deformed labels are combined to achieve the final segmentation result. In
general, there are three approaches to design an ABS algorithm: multi-atlas, atlas selection, and
probabilistic atlas.
The multi-atlas approach registers a limited number of atlas images to the target image to
create multiple (deformed) labels, which are then combined using a fusion algorithm to generate
the final result. Different fusion algorithms for multi-atlas approaches have been proposed in the
literature. Majority voting is the simplest approach where each pixel is assigned to the label which
occurs with the highest frequency. More complex methods attempt to weigh the segmentation
results from each atlas based on their estimated accuracy. Langerak et al, proposed a multi-atlas
segmentation method for prostate MR images based on an iterative label fusion approach. First,
all atlas images are registered to the target image to obtain a set of deformed labels. The deformed
labels are fused together using a weighted majority voting method and then, the overlap of each
label is calculated against the fused label. Labels with low overlaps are discarded and a new fused
label is recalculated

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
As another label fusion method, the STAPLE algorithm, uses an expectation-maximization
approach in which the unknown true segmentation and the performance of individual atlas
segmentations are estimated simultaneously. The atlas selection approach proposed in uses both
STAPLE and majority voting as the postprocessing stage to fuse the generated labels. Another
approach is to assume that the performance of each segmentation is related to the similarity
between the registered atlas image and the target image. In it was found that weighing labels using
a local image similarity measure given by the mean squared difference between target and
registered atlas image gave better results than majority voting or STAPLE. Several atlas selection
methods have been proposed which try to find the subset of atlas images that best represent a
specific target image in order to improve accuracy. The mutual information (MI) between the atlas
and the target image after registration has been used as a figure of merit; for example in the target
images are subdivided into regions and for each region the atlas image with the best local MI is
selected, and in the subset of atlas images that has an MI greater than a preset threshold level is
used.
Atlas selection approaches that require all atlas images to be registered to the target image
are computationally prohibitive and limit the number of atlases that can be used. To reduce the
computational cost of atlas selection, Aljabar et al. proposed to register all atlas images and the
target image to an arbitrary reference image and thus, the best-match atlas is selected by comparing
the registered target image with the registered atlas images. In other words, the atlas registrat ion
can be performed offline during the atlas construction phase reducing the computational time. In
a recent work, Langerak et al, proposed a method that clusters the atlas images using the pairwise
registration results to group similar atlases together. For each cluster, an exemplar atlas is selected
and these are then registered to the target image. Once the best exemplar is found, all of the atlas
images belonging to that cluster are registered to the target image and the resulting labels are
combined using a simple majority voting. Although the process of clustering the atlases off-line is
extremely computationally expensive, the number of registrations required for each target image
is somewhat reduced
The probabilistic atlas approaches are usually based on either sequential models or
generative models. In sequential models,the probabilistic atlas is generated by aligning (and
averaging) multiple images and their labels in a dataset using a registration algorithm. In generative
models, the tissue classification and image deformation are performed simultaneously using an

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
objective function optimized by deforming the tissue probability maps. The main difference
between the two classes is that in the former, the images are first registered and then combined to
generate a probabilistic atlas. Next, the resulting probabilistic atlas is registered to the target image
and consequently, the probabilistic label is deformed via the registration transformation to generate
the result for the target image. In contrast, in the latter model, the target image is segmented
(classified) directly as a result of deforming the tissue maps of the image via optimizing the
objective function.
There are different variations of sequential models of probabilistic atlas algorithms in the
literature. For instance, Martin et al. proposed to create a probabilistic atlas for prostate by
registering images to a manually picked reference image. Next, a mean image is created by
averaging the registered images. Each image is registered against the mean image to produce a
deformed label of the image. The deformed labels are then averaged to generate a probability map
of the labels. To segment a target image, the mean image is registered to it and the probability map
of the labels is deformed using the registration transformation. Dowling et al. Also proposed a
similar approach. To build a probabilistic atlas using a sequential model, a reference image(s)
which has already been manually segmented and selected plays a crucial role. In practice, however,
generating a reference image is not straight forward; even with careful atlas selection, the selected
image may not represent the population well enough and hence, the registration may not produce
acceptable results. Balci et al. Extended the groupwise registration algorithm proposed in to be an
alternative to pair-wise reg istration when designing a sequential probabilistic atlas for segmenting
brain MR images. Groupwise registration aligns a set of images to a common reference image by
generating a set of transformations that map the reference image to each of the images in the group.
This makes the sequential probabilistic atlas more robust and unbiased to a particular reference
image.
Ashburner and Friston, proposed a generative model of probabilistic atlas which unifies
tissue classification, bias correction, and template registration for brain MR images. The objective
function is based on a mixture of Gaussian (MOG) models where the tissue classes are represented
by Gaussian probabilities and the bias, which is a smooth and spatially varying artifact that corrupts
the image, is taken into account by extra parameters in MOG model. In this method, a template
(spatial priors) is used in the process of optimizing the objective function where the deformation
(registration) is updated during the process. Recently, Iglesias et al. Proposed a framework for

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
multi-atlas segmentation of brain in MR images. The proposed algorithm is an extension of the
work presented in where in contrast to it builds multiple probabilistic atlases. Although the ABS
algorithms have been applied extensively to different organs such as brain and prostate, there have
been limited work on breast segmentation using ABS in the literature. Gubern et al. Proposed a
probabilistic atlas algorithm to segment different structures in breast MR (T1-weighted) images.
In this method, the probabilistic atlas is created using pairwise registration of each atlas with a
reference image. The segmentation is done in a Bayesian framework where for a given target image
voxel, the segmentation is estimated by searching for the most probable label.
The generative models of probabilistic ABS algorithms are usually computationally
expensive since they integrate multiple optimizations. The main drawback of the sequential models
of probabilistic ABS algorithms is that they usually depend on the image intensity, rather than
image key characteristics. As a result, if intensity variations exist between the target image and
atlas image, which is common when the two images are from different scanners, the ABS
algorithms may perform poorly resulting in inaccurate segmentation results. Moreover, in the
conventional ABS algorithms, the atlas and target images must be from the same image sequence
(e.g., MRI T1-weighted) in order to obtain reasonable segmentation results. One possible approach
to make the sequential models of probabilistic ABS intensity invariant is to use a registration
algorithm that is robust to intensity variations. Heinrich et al. Proposed an intensity-robust feature
for image registration based on modality independent neighborhood descriptor, which can be used
for nonrigid registration across different modalities and different sequences of one modality
In this paper, we propose an improved ABS algorithm which is novel in that it combines
the atlas selection method with a probabilistic atlas generated using a sequential model. The atlas
images are grouped into classes based on their pairwise similarities. Each class of atlases is then
used to create a probabilistic atlas using a groupwise registration. The proposed algorithm is
presented in two configurations for the intra-sequence and inter-sequence settings where in the
former, the atlas and target images are from the same MR imaging sequence and in the latter, they
are from different MR imaging sequences. In order to tackle the intensity-dependence deficiency
of conventional ABS algorithms, for the inter-sequence setting, we use phase congruency maps,
which are robust to intensity variations, to build the probabilistic atlas1. These intensity-invariant
maps are also used to register the atlas and target images. In order to refine the atlas results, the
actual edge information from the target image is also incorporated directly into the segmentation

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
pipeline by using an active contour model. To reduce the computational time, our proposed
approach performs the atlas selection in pre-nonrigid registration stage and hence, to segment a
targ

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
CONCLUSION:
In this paper, a robust ABS algorithm was presented with two configurations for intra-
sequence and inter-sequence settings. The proposed algorithm brings different ABS strategies into
a single framework. It also proposes an atlas robust to intensity variations which allows an atlas of
one particular MR imaging sequence to be used to segment images of a different MR sequence.
The proposed algorithm was applied to both intra-sequence and inter-sequence settings where in
the former the atlas and target images were both acquired using a Dixon sequence and in the latter,
the atlas and target images were from Dixon and MRI T1-weighted sequences, respectively. In
both cases, highly accurate results were achieved (median DSC of 94% and 87%, respectively).

CELL: +91 9894917187 | 875487 1111 / 2111 / 3111 / 4111 / 5111 / 6111
ECWAY TECHNOLOGIES
REFERENCES:
[1] J. H. Jonsson, M. G. Karlsson, M. Karlsson, and T. Nyholm, “Treatment planning using MRI
data: An analysis of the dose calculation accuracy for different treatment regions,” Radiat. Oncol.,
vol. 5, 2010.
[2] J. Dowling, J. Fripp, P. Greer, J. Patterson, S. Ourselin, and O. Salvado, “Automatic atlas-
based segmentation of the prostate: A MICCAI 2009 Prostate Segmentation Challenge entry,” in
Worskshop Med. Image Comput. Assist. Interv., 2009, pp. 17–24.
[3] A. Gubern-Merida, M. Kallenberg, R. Marti, and N. Karssemeijer, “Segmentation of the
pectoral muscle in breast MRI using atlas-based approaches,” Med. Image Comput. Comput.
Assist. Intervent., vol. 15, pt. 2, pp. 371–378, 2012.
[4] T. R. Langerak, U. A. van der Heide, A. N. T. J. Kotte, M. A. Viergever, M. van Vulpen, and
J. P. W. Pluim, “Label fusion in atlas-based segmentation using a selective and iterative method
for performance level estimation (SIMPLE),” IEEE Trans. Med. Imag., vol. 29, no. 12, pp. 2000–
2008, Dec. 2010.
[5] X. Artaechevarria, A. Munoz-Barrutia, and C. Ortiz-de-Solorzano, “Combination strategies in
multi-atlas image segmentation: Application to brain MR data,” IEEE Trans. Med. Imag., vol. 28,
no. 8, pp. 1266–1277, Aug. 2009.
[6] P. Aljabar, R. A. Heckemann, A. Hammers, J. V. Hajnal, and D. Rueckert, “Multi-atlas based
segmentation of brain images: Atlas selection and its effect on accuracy,” Neuroimage, vol. 46,
pp. 726–738, 2009.
[7] T. R. Langerak, F. F. Berendsen, U. A. Van der Heide, and A. N. T. J. Kotte, “Multiatlas-based
segmentation with preregistration atlas selection,” Med. Phys., vol. 40, no. 9, pp. 091701-1–8,
2013.

Automated segmentation of breast in 3 d mr images using a robust atlas

Recommended

Recommended

More Related Content

Recently uploaded

Recently uploaded (20)

Featured

Featured (20)

Automated segmentation of breast in 3 d mr images using a robust atlas