International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
Pilocytic Astrocytoma Cellular Tumor
Dissection Modus on Contrast MR
Images
K.Sudhakar#1, Dr. K.P.Yadav*2
#
Research Scholar, Department of Computer Science and Engineering,
SunRise University, Alwar, India.
*
Professor & Director, Mangalmay Institute of Engineering & Technology ,
Greater Noida,U.P, India.
Abstract— Accurate automatic extraction of a 3-D
cerebrovascular system from images obtained by Time-OfFlight (TOF) or Phase Contrast (PC) Magnetic Resonance
Angiography (MRA) is a challenging segmentation problem
due to the small size objects of interest (blood vessels) in each
2-D MRA slice and complex surrounding anatomical
structures (e.g., fat, bones, or gray and white brain matter).
We show that due to the multimodal nature of MRA data,
blood vessels can be accurately separated from the
background in each slice using a voxel-wise classification
based on precisely identified probability models of voxel
intensities. To identify the models, an empirical marginal
probability distribution of intensities is closely approximated
with a Linear Combination of Discrete Gaussians (LCDG)
with alternate signs, using our previous EM-based techniques
for precise linear combination of Gaussian approximation
adapted to deal with the LCDGs. The high accuracy of the
proposed approach is experimentally validated on 85 real
MRA datasets (50 TOF and 35 PC) as well as on synthetic
MRA data for special 3-D geometrical phantoms of known
shapes.
the brain. The cerebellum located at the back of the brain
and it consists of outer GM and internal WM. The
brainstem connects to the spinal cord consists of midbrain,
pons and medulla oblongata. The diencephalon layer is the
central structure of the brain and consists of thalamus,
hypothalamus and pituitary gland and communicated
through ventricles.
Keywords—: Cerebrovascular system, linear combination of
discrete Gaussians (LCDG), magnetic resonance angiography
(MRA), segmentation.
I.
INTRODUCTION
The Central Nervous System and peripheral nervous system
are the two main nervous systems present in the brain
structure and it consists of Gray Matter (GM) and White
Matter (WM). The Gray Matter control brain activity and
cortex region cover the brain which is made of glial cells
and the gray matter nuclei (colostrum) are located deep
within the white matter. The myelinated axons are
considered as white matter fibers that connect the cerebral
cortex with other brain regions. The cerebrospinal fluid
(CSF ) consists of nutrition rich glucose, salts, enzymes and
WBC's present between the lower part of brain and spinal
cord. The meninges is present in the intra cranial of brain
and act as protective layer. The cerebrum parts of brain is
divided into two hemisphere regions, the right and left
cerebral hemisphere and consists of four lobes including
parietal, frontal, temporal and occipital lobe at the back of
ISSN: 2231-5381
Figure 1.1 Structure of Axon in human brain
In the vertebrate animal, the brain appears to be a most
complex organ and contains many billions of neurons
depends upon the cerebral cortex, each connected through
synapses or axon to another few hundred of neurons and it
communicates very quickly in irregular patterns that
withstand fault tolerance. It carries trains of signal pulses
and secretes hormones and it differentiates from Glial cells
inside the human body and it support metabolic support and
structural support. The neurons are covered by a fatty
substance known as myelin sheath which contains rich in
nerve fibers and hence the tumor cells has tended to grow in
foreign body and gain energy and similarly, Glial cells also
involve in brain metabolism through controlling the
chemical fluids like ions and nutrients around the neurons
and this is the main reason tumor grows only in brain.
Accurate 3-D cerebrovascular system segmentation from
magnetic resonance angiography MRA) images are one of
the most important problems in practical computer assisted
medical diagnostics. Phase contrast (PC)-MRA provides
good suppression of background signals and quantifies
blood flow velocity vectors for each voxel. Time-of-flight
http://www.ijettjournal.org
Page 93
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
(TOF)-MRA is less quantitative, but it is fast and provides
images with high contrast. The most popular techniques for
extracting blood vessels from MRA data are scale-space
filtering, centerline-based methods, deformable models,
statistical models, and hybrid methods. Multiscale filtering
enhances curvilinear structures in 3-D medical images by
convolving an image with Gaussian filters at multiple scales
[1]–[4]. Eigenvalues of the Hessian at each voxel are
analyzed to determine the local shapes of 3-D structures (by
the eigenvalues, voxels from a linear structure, like a blood
vessel, differ from those for a planar structure, speckle
noise, or unstructured components). The multiscale filter
output forms a new enhanced image such that the
curvilinear structures become brighter whereas other
components (e.g., speckle noise and planar structures such
as skin) become darker [1]. Such an image can be directly
visualized, threshold, or segmented using a deformable
model. Alternatively, the obtained eigenvalues define a
candidate set of voxels corresponding to the centerlines of
the vessels [2]. Multiscale filter responses at each of the
candidates determine the likelihood that a voxel belongs to
a vessel of each particular diameter. The maximal response
over all the diameters (scales) is assigned to each voxel, and
a surface model of the entire vascular structure is
reconstructed from the estimated centerlines and diameters.
After segmenting the filtered MRA image using
thresholding, anisotropic diffusion techniques are used to
remove noise, but preserve small vessels [3]. Lacoste et al.
[4] proposed a multiscale technique based on the Markov
marked point processes to extract coronary arteries from 2D X-ray angiograms. Coronary vessels are modeled locally
as piece-wise linear segments of varying locations, lengths,
widths, and orientations. The vessels’ centerlines are
extracted using a Markov object process specified by a
uniform Poisson process. Process optimization was
achieved via simulated annealing using a reversible Markov
chain Monto Carlo algorithm. Centerline minimal pathbased techniques [5]–[7] formulate the two-point centerline
extraction as the minimum cost integrated along the
centerline path. G¨uls¨un and Tek [5] used multiscale
medialness filters to compute the cost of graph edges in a
graph-based minimal path detection method to extract the
vessels’ centerlines. A post processing step, based on the
length and scale of vessel centerlines, was performed to
extract the full vessel centerline tree. P`echaud et al. [6]
presented an automatic framework to extract tubular
structures from 2-D images by the use of shortest paths.
Their framework combined multiscale and orientation
optimization to propagate 4-D (space + scale + orientation)
paths on the 2-D images. Li and Yezzi [7] represented the
3-D vessel surface as a 4-D curve, with an additional non
spatial dimension that described the radius (thickness) of
the vessel. They applied a minimal path approach to find the
minimum path between user defined end points in the 4-D
space. The detected path simultaneously described the
vessel centerline as well as its surface. To overcome the
possible shortcut problem of minimal path techniques (i.e.,
track a false straight shortcut path instead of following the
true curved path of the vessel), Zhu and Chung [8] used a
ISSN: 2231-5381
minimum average-cost path model to segment the 3-D
coronary arteries from CT images.
II.
LITERATURE SURVEY
Many in their approach, the average edge cost is minimized
along paths in the discrete 4-D graph constructed by image
voxels and associated radii. Deformable model approaches
to 3-D vascular segmentation attempt to approximate the
boundary surface of the blood vessels [9]–[14]. An initial
boundary, called a snake [15], evolves in order to optimize
a surface energy that depends on image gradients and
surface smoothness. To increase the capture range of the
evolving boundary, Xu and Prince [16] used a gradient
vector flow (GVF) field as an additional force to drive
snakes into object concavities, which was later used to
segment the blood vessels from 3-D MRA [9]. Geodesic
active contours [17] implemented with level set techniques
offer flexible topological adaptability to segment the MRA
images [10] including more efficient adaptation to local
geometric structures represented. Fast segmentation of
blood vessel surfaces is obtained by inflating a 3-D balloon
with fast marching methods [11]. Holtzman-Gazit et al. [12]
extracted blood vessels in CTA images based on variational
principles. Their framework combined the ChanVese
minimal variance model with a geometric edge alignment
measure and the geodesic active surface model. Manniesing
et al. [13] proposed a level set-based vascular segmentation
method for finding vessel boundaries in CTA images. The
level set function is attracted to the vessel boundaries based
on a dual object (vessels) and background intensity
distributions, which are estimated from the intensity
histogram. Recently, Forkert et al. [14] used a vesselness
filter to guide the direction of a level set to extract vessels
from TOF-MRA data. It is well acknowledged that medical
image databases, which enable access to a patient’s
historical data, including multidimensional medical images
from previous examinations and the opportunity for
statistical and comparative image analyses, are key
components in preventive medicine and future diagnosis
[4], [5]. However, these medical images are primarily
indexed by text keywords that limit the image features to
textual descriptions and retrievals to text-based queries.
Although text-based retrieval is capable of supporting a
high degree of image-content semantics, it is likely that
text-based retrieval is unable to sufficiently describe the
visual features of the images [6], [7]. The content-based
image retrieval (CBIR) of medical images according to its
domain-specific image features is an important alternative
and complement to traditional text-based retrieval using
keywords [3], [6], [8]–[16]. In recent years, various CBIR
systems have been introduced for medical images. In
medical CBIR systems, visual features commonly applied
to broad images are unable to fully describe the
characteristics and diagnostic meanings of these images [3],
[8]. Furthermore, due to the differences within the medicalimaging modalities, the CBIR systems are designed and
implemented
to
domain-specific
medical-imaging
modalities. Consequently, to extract features for a particular
http://www.ijettjournal.org
Page 94
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
imaging modality, image processing algorithms that are
domain specific to the imaging applications must also be
developed.
III. METHODOLOGY
3.1 Existing Method
Comparison Compared to scale-space filtering, deformable
models produce much better experimental results, but have
a common drawback, namely, a manual initialization. Also,
both group approaches are slow when compared to
statistical approaches. Statistical extraction of a vascular
tree is completely automatic, but its accuracy depends on
the underlying probability models. The MRA images are
multimodal in that the signals (intensities or gray levels) in
each region of interest (e.g., blood vessels, brain tissues,
etc.) are associated with a particular dominant mode of the
total marginal probability distribution of signals. To the best
of our knowledge, adaptive statistical approaches for
extracting blood vessels from the MRA images have been
proposed so far only by Wilson and Noble [18] for the
TOF-MRA data and Chung and Noble [19] for the PCMRA data. The former approach represents the marginal
data distribution with a mixture of two Gaussians and one
uniform component for the stationary cerebrospinal fluid
(CSF), brain tissues, and arteries, respectively, whereas the
latter approach replaces the Gaussians with the more
adequate Rician distributions. To identify the mixture (i.e.,
estimate all its parameters) a conventional EM algorithm is
used in both cases. It was called a “modified EM” in [18],
after replacing gray levels in individual pixels considered
by their initial EM scheme with a marginal gray level
distribution. Actually, such a modification simply returns to
what has been in common use for decades for density
estimation (see e.g., [20]), while the individual pixels
appeared in their initial scheme only as an unduly verbatim
replica of a general EM framework. Different hybrid
approaches have attempted to combine the aforementioned
approaches. For instance, a region-based deformable
contour for segmenting tubular structures is derived in [21]
by combining signal statistics and shape information. Law
and Chung [22] guided a deformable surface model with the
second order intensity statistics and surface geometry to
segment blood vessels from TOF- and PC-MRA images. A
combination of a Gaussian statistical model with the
maximum intensity projection (MIP) images acquired at
three orthogonal directions [23] allows for extracting blood
vessels iteratively from images acquired by rotational
angiography. Alternatively, Hu and Hoffman [24] extracted
the object boundaries by combining an iterative
thresholding approach with region growing and component
label analysis. Mille et al. [25] used a generalized cylinder
(GC) region based deformable model for the segmentation
of the angiogram. The GC is modeled as a central planar
curve, acting as a medial axis, and a variable thickness. The
GC is deformed by coupling the evolution of the curve and
thickness using narrow band energy minimization. This
energy was transformed and derived in order to allow
ISSN: 2231-5381
implementation on a polygonal line deformed with gradient
descent. Tyrrell et al. [26] proposed a super ellipsoid
geometric model to extract the vessel boundaries from in
vivo optical slice data. Their approach predicted the
direction of the centerline utilizing a statistical estimator.
Chen and Metaxas [27] combined a prior Gibbs random
field model, marching cubes, and deformable models. First,
the Gibbs model is used to estimate object boundaries using
region information from 2-D slices. Then, the estimated
boundaries and the marching cubes technique are used to
construct a 3-D mesh specifying the initial geometry of a
deformable model. Finally, the deformable model fits the
data under the 3-D image gradient forces. Recently, Shang
et al. [28] developed an active contour framework to
segment coronary artery and lung vessel trees from CT
images. A region, competition-based active contour model
is used to segment thick vessels based on a Gaussian
mixture model of the gray-level distribution of the vessel
region. Then, a multiscale vector field, derived from the
Hessian matrix of the image intensity, is used to guide the
active contour through thin vessels. Finally, the surface of
the vessel is smoothed using a vesselness function that
selects between a minimal principal curvature and a mean
curvature criterion. Gao et al. [29] used a statistical model
to find the main cerebrovascular structure from TOF-MRA.
Then, an edge-strength function that incorporates statistical
region distribution and gradient information is used to guide
a 3-D geometric deformable model to deal with the under
segmentation. Dufour et al. [30] proposed an interactive
segmentation method that incorporates component-trees and
example-based segmentation to extract the cerebrovascular
tree from TOF-MRA data.
3.2 Disadvantages
The above mentioned overview shows the following
limitations of the existing approaches.
1) Most of them presume only a single image type
(e.g., TOF- or PC-MRA).
2) Most of them require user interaction to
initialize a vessel of interest.
3) Some deformable models assume circular vessel
cross sections; this holds for healthy people, but not for
patients with a stenosis or an aneurysm.
4)
All
but
statistical
approaches
are
computationally expensive.
5) Known statistical approaches use only
predefined probability models that cannot fit all the cases
because actual intensity distributions for blood vessels
depend on the patient, scanner, and scanning parameters.
3.3 Proposed Method
In our approach, the empirical gray level distribution for
each MRA slice is closely approximated with an LCDG.
Then, the latter is split into three individual LCDGs, one per
region of interest. These regions are associated with three
dominant modes: darker bones and fat, gray brain tissues,
and bright blood vessels, respectively. The identified
http://www.ijettjournal.org
Page 95
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
models specify an intensity threshold for extracting blood
vessels in that slice. Finally, a 3-D connectivity filter is
applied to the extracted voxels to select the desired vascular
tree. As our experiments show, more precise region models
result in significantly better segmentation accuracy
compared to other methods.
distribution of gray levels for the MRA slice Xs . In
accordance with [32], each such slice is considered as a Kmodal image with a known number K of the dominant
modes related to the regions of interest (in our particular
case, K = 3). To segment the slice by separating the modes,
we have to estimate the individual probability distributions
of the signals associated with each mode from fs .
3.4 Advantages
List of Modules
We have shown the fast and highly accurate statistical
approach to extract blood vessels obtained when the
probability models of each region of interest in TOF- or PCMRA images are precisely identified rather than predefined
as in [18] and [19]. The primary motivation of this study is
the fact that statistical approaches are faster than other
segmentation methods. The main limitation of known
statistical approaches is that they are based on predefined
probability models that cannot fit all possible cases due to
the fact that actual intensity distributions for blood vessels
depend on the patient, scanner, and scanning parameters. In
response to this problem, we propose a fast and highly
accurate adaptive statistical approach to extract blood
vessels from TOF- and PC-MRA images. In this approach,
the probability models of each region of interest are
precisely identified using an LCDG based approximation,
rather than predefined approximations as used in other
statistical approaches [18], [19]. This makes the proposed
approach suitable to work with any imaging modality, e.g.,
TOF- or PC-MRA images. Comparisons between the
proposed approach and the current state-of-the-art
segmentation approaches on both real MRA datasets (TOF
and PC) and synthetic phantoms confirm that our more
precise model yields a much higher segmentation accuracy,
In addition, qualitative analysis in Fig. 6 shows that other
approaches fail to detect sizeable parts of the brain vascular
trees, assigned by expert radiologists to the actual trees,
whereas these sections are successfully extracted when
using our approach. In terms of practicality of
computations, the execution time of our approach is faster
than deformable model-based approaches [15], [16], [29]
and slightly slower than current well-known statistical
approaches [18], [19] due to the sub models estimation step
which does not exist in these approaches. For the proposed
LCDG probabilistic model, the only manually provided
parameter is the number K of the dominant modes in the
empirical mixture (that is, the number of objects to be
separated by the segmentation).
1.
2.
3.
4.
5.
cumulative Gaussian probability
K-modal probability model
Log-likelihood for empirical data
LCDG-sub model
Object Thresholding for MRA Voxel Analysis
Cumulative Gaussian probability
In contrast to a conventional mixture of Gaussians, one per
region [20], or slightly more flexible mixtures involving
other simple distributions, one per region, as e.g., in [18]
and [19], we closely approximate Fs with LCDG. Then, the
LCDG of the image is partitioned into sub models related to
each dominant mode. The discrete Gaussian (DG) is defined
as the probability distribution Ψθ = (ψ(q|θ): q ∈ Q) on Q of
gray levels such that each probability ψ(q|θ) relates to the
cumulative Gaussian probability function Φθ (q) as follows
( | )
{
(
(
)
)
(
(
)
)
K-modal probability model
The LCDG with Cp positive and Cn negative components
such that Cp ≥ K
( )
∑
( |
)
∑
( |
has obvious restrictions on its weightsw = [
namely, all the weights are non negative and
∑
,.,
)
,. ],
∑
IV. CELLULAR BASED SEEDED SEGMENTATION
METHOD
We use the expected log-likelihood as a model
identification criterion. Let X = (Xs : s = 1, . . . , S) denote a
3-D MRA image containing S coregistered 2-D slices Xs =
(Xs (i, j): (i, j) ∈ R;Xs (i, j) ∈ Q). Here, R and Q = {0, 1, . .
., Q − 1} are a rectangular arithmetic lattice supporting the
3-D image and a finite set of Q-ary intensities (gray levels),
respectively. Let Fs = (fs (q): q ∈ Q;q∈Q fs (q) = 1, where q
denotes the gray level, be an empirical marginal probability
ISSN: 2231-5381
Generally, the true probabilities are nonnegative: , ( ) ≥
0 for all q ∈ Q. Therefore, the probability distributions
comprise only a proper subset of all the LCDGs in (1),
which may have negative components
, ( ) < 0 for
some q ∈ Q.
Log-likelihood for empirical data
Our goal is to find a K-modal probability model that closely
approximates the unknown marginal gray level distribution.
http://www.ijettjournal.org
Page 96
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
Given Fs , its Bayesian estimate F is as follows [20]: f(q) =
(|R|fs (q) + 1)/(|R| + Q), and the desired model has to
maximize the expected log-likelihood of the statistically
independent empirical data by the model parameters:
(
)
∑ ( )
( )
∈
For simplicity, we do not restrict the identification
procedure to only the true probability distributions, but
instead check the validity of the restrictions during the
procedure itself. The Bayesian probability estimate F with
no zero or unit a value in (3) ensures that a sufficiently large
vicinity of each component f(q) complies with the
restrictions. To precisely identify the LCDG-model
including the numbers of its positive and negative
components.
Object Thresholding for MRA Voxel Analysis
Eliminate artifacts from the whole set of the extracted
voxels using a connectivity filter that selects the largest
connected tree structure built by a 3-D volume growing
algorithm [33].1 The main goal of the whole procedure is to
find the threshold for each MRA slice that extracts the
brighter blood vessels from their darker background in such
a way that the vessels’ boundaries are accurately separated
from the surrounding structures that may have similar
brightness along these boundaries. The initialization at Step
1b produces the LCDG with the nonnegative starting
probabilities pw,Θ(q).While the refinement at Step 1c
increases the likelihood, and the probabilities continue to be
nonnegative. In our experiments presented as follows, the
opposite situations have never been met.
a) Collect the marginal empirical probability distribution
Fs = (fs (q): q ∈ Q) of gray levels.
b) Find an initial LCDG-model that closely approximates Fs
by using the initializing algorithm in Appendix A to
estimate the numbers Cp − K, Cn , and parameters w, Θ
(weights, means, and variances) of the positive and negative
DGs.
c) Refine the LCDG-model with the fixed Cp and Cn by
adjusting all other parameters with the modified EM
algorithm in Appendix B.
d) Split the final LCDG-model into K sub models, one per
each dominant mode, by minimizing the expected errors of
misclassification and select the LCDG-sub model with the
largest mean value (i.e., the sub model corresponding to the
brightest pixels) as the model of the desired blood vessels.
e) Extract the blood vessels’ voxels in this MRA slice using
the intensity threshold t separating their LCDG-sub model
from the background ones.
V. EXPERIMENT RESULT
The programming language used with MATLAB is usually
referred to as MATLAB script or M-script. After becoming
familiar with the basic syntax of the M-script, a number of
useful utilities are available to you that allow you to make
extended uses of MATLAB. You can, for example, write
programs that involve simulation. You can also create
graphics, web pages, and GUI applications. When you
develop programs using MATLAB, you can output the
results to a number of media, including graphics files,
HTML pages, PDF files, and Word documents. You can
also connect up MATLAB with other applications, such as
Excel or LabView to make extended uses of it. Since it is
programmed in part using Java, you can modify it in the
background using Java. The feasibility of the project is
analyzed in this phase and business proposal is put forth
with a very general plan for the project and some cost
estimates. During system analysis the feasibility study of
the proposed system is to be carried out. This is to ensure
that the proposed system is not a burden to the company.
For feasibility analysis, some understanding of the major
requirements for the system is essential.
SCREENSHOTS
Fig. 2-D schematic illustration of measuring segmentation errors between
the ground truth G and proposed automatic segmentation C.
LCDG-sub model
we adapt to the LCDGs our EM-based techniques [32] for
identification of a probability density with a continuous
linear combination of Gaussian model. For completeness,
the adapted algorithms are outlined in Appendix A. The
entire segmentation algorithm is as follows with for each
successive MRA slice Xs , s = 1, . . . , S,
Fig: 5.1 Input Image in MRA
ISSN: 2231-5381
http://www.ijettjournal.org
Page 97
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
Fig: 5.2 Cumulative Gaussian Filtering
Fig: 5.6 Iteration of Object in Voxel Computation
Fig: 5.3 Probability Model for Object Analysis
Fig: 5.7 Surf Analysis for Identifying Multi Object
Fig: 5.4 Likelihood Function Analysis
Fig: 5.8 Processing Probability Map Generation
Fig: 5.5 Gray Scale Contiguous Voxel Estimation
Fig: 5.9 Object Thresholding for MRA Voxel Analysis
ISSN: 2231-5381
http://www.ijettjournal.org
Page 98
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
workstation) depending on the volume of the tumor which
ranges between 0.5 and 32 cc. Due to inherent parallelism
of the proposed algorithm, computation time can be
significantly reduced.
REFERENCES
[1]
[2]
[3]
Fig: 5.10 Probability Interpolated Object in Blood Vessel
[4]
[5]
VI. CONCLUSION
[6]
6.1 Conclusion
This paper presents a generalized automated approach for
the extraction of the 3-D cerebrovascular system that is
suitable for any imaging modality, e.g., CTA, TOF-MRA,
and PC-MRA images. Voxel intensity-based models for
classification are employed using a linear combination of
discrete Gaussians (LCDG) to accurately identify the
empirical distribution of the gray level intensity in the
images. A modified EM-based approach has been used to
identify the LCDG models. Accuracy and validity of the
method has been demonstrated on 85 in vivo MRA datasets
along with synthetic phantom data, confirming the high
accuracy and speed of the proposed LCDG-based extraction
method,a segmentation algorithm for the problem of tumor
delineation which exhibit varying tissue characteristics. As
the change in necrotic and enhancing part of the tumor after
radiation therapy becomes important, we also applied the
Tumor-cut segmentation to partition the tumor tissue further
into its necrotic and enhancing parts. We presented
validation studies over a synthetic tumor database and two
real tumor databases: one from Harvard tumor repository
and another from a clinical database of tumors that
underwent radiosurgery planning at Radiation Oncology
Department of ASM. The performance over particularly
datasets of highly heterogeneous tissue content
demonstrated an overlap in the range 80%–90%, however,
with a desired low surface distance error, average median
surface distances of 1.0–1.5 mm, respectively.
6.2 Future Work
Furthermore, performance change over varying initial seeds
were also reported as standard deviations, and shown to be
important in assessing true robustness of the proposed
algorithm in real application scenarios. The user interaction
time is just a few seconds and typical computation times
vary between 1 s to 16 min (on a 3.17 GHz dual processor
ISSN: 2231-5381
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Y. Sato, S. Nakajimaa, N. Shiragaa, H. Atsumia, S. Yoshidab, T.
Kollerc, G. Gerigc, and R. Kikinisa, “Three-dimensional multi-scale
line filter for segmentation and visualization of curvilinear structures
in medical images,” Med. Image Anal., vol. 2, no. 2, pp. 143–168,
1998.
K.Krissian,G.Malandain,N.Ayache, R.Vaillant, andY. Trousset,
“Model based multiscale detection of 3D vessels,” in Proc. IEEE
Conf. Comput. Vision Pattern Recognit., 1998, pp. 722–727.
F. Catt´e, P-L. Lions, J.-M. Morel, and T. Coll, “Image selective
smoothing and edge detection by nonlinear diffusion,” SIAM J.
Numer. Anal., vol. 29, no. 1, pp. 182–193, 1992.
C. Lacoste, G. Finet, and I. E. Magnin, “Coronary tree extraction
from X-ray angiograms using marked point processes,” in Proc.
IEEE Int. Symp. Biomed. Imag., 2006, pp. 157–160.
M. A. G¨uls¨un and H. Tek, “Robust vessel tree modeling,” in Proc.
Int. Conf. Med. Image Comput. Comput—Assist. Intervent., 2008,
pp. 602– 611.
M. P`echaud, R. Keriven, and G. Peyr´e, “Extraction of tubular
structures over an orientation domin,” in Proc. IEEE Conf. Comput.
Vis. Pattern Recognit., 2009, pp. 336–342.
H. Li and A. Yezzi, “Vessels as 4-D curves: Global minimal 4-D
paths to extract 3-D tabular surfaces and centerlines,” IEEE Trans.
Med. Imag., vol. 26, no. 9, pp. 1213–1223, Sep. 2007.
N. Zhu and A. C. Chung, “Minimum average-cost path for real time
3D coronary artery segmentation of CT images,” in Proc. Int. Conf.
Med. Image Comput. Comput.—Assist. Intervent., 2011, pp. 436–
444.
A. C. Jalba, M. H. Wilkinson, and J. B. Roerdink, “CPM: A
deformable model for shape recovery and segmentation based on
charged particles,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 26,
no. 10, pp. 1320–1335, Oct. 2004.
L. M. Lorigo, O. D. Faugeras, W. E. Grimson, R. Keriven, R.
Kikinis, A. Nabavi, and C. F.Westin, “Curves: Curve evolution for
vessel segmentation,” Med. Image Anal., vol. 5, no. 3, pp. 195–206,
2001.
T. Deschamps and L. D. Cohen, “Fast extraction of tubular and tree
3Dsurfaces with front propoagation methods,” in Proc. IEEE Int.
Conf. Pattern Recognit., 2002, pp. 731–734.
M. Holtzman-Gazit, R. Kimmel, N. Peled, and D. Goldsher,
“Segmentation of thin structures in volumetric medial images,” IEEE
Trans. Image. Process., vol. 15, no. 3, pp. 354–363, Feb. 2006.
B. Vijayakumar, Ashish Chaturvedi and K. Muthu Kumar, “Effective
Classification of Anaplastic Neoplasm in Huddling Stain Image by
Fuzzy Clustering Method”, International Journal of Scientific
Research, Issue 3 volume 2, March-April 2013,
[14] N. D. Forkert, D. S¨aring, T. Illies, J. Fiehler, J. Ehrhardt, H.
Handels, and A. Schmidt-Richberg, “Direction-dependent level set
segmentation of cerebrovascular structures,” Proc. SPIE, Image
Process.: Med. Imag., vol. 7962, pp. 1–8, 2011.
[15] M. Kass, A.Witkin, and D.Terzopoulos, “Snakes:Active contour
models,” Int. J. Comput. Vision, vol. 1, pp. 321–331, 1988.
[16] C. Xu and J. L. Prince, “Snakes, shapes, and gradient vector flow,”
IEEE Trans. Image Process., vol. 7, no. 3, pp. 359–369, 1998.
[17] V. Caselles, R. Kimmel, and G. Sapiro, “Geodesic active contours,”
Int. J. Comput. Vision, vol. 22, no. 1, pp. 61–79, 1997.
[18] D. L. Wilson and J. A. Noble, “An adaptive segmentation algorithm
for time-of-flight MRA data,” IEEE Trans. Med. Imag., vol. 18, no.
10, pp. 938–945, Oct. 1999.
[19] A. C. S. Chung and J. A. Noble,“Statistical 3D vessel segmentation
using a Rician distribution,” in Proc. Int. Conf. Med. Image Comput.
Comput.— Assist. Intervent., 1999, pp. 82–89.
[20] A. Webb, Statistical Pattern Recognition. New York: Wiley, 2002.
[21] D. Nain, A. Yezzi, and G. Turk, “Vessels segmentation using a
shape driven flow,” in Proc. Int. Conf. Med. Image Comput.
Comput.—Assist. Intervent., 2004, pp. 51–59.
http://www.ijettjournal.org
Page 99
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
[22] M. W. Law and A. C. Chung, “A deformable surface model for
vascular segmentation,” in Proc. Int. Conf. Med. Image Comput.
Comput.—Assist. Intervent., 2009, pp. 59–67.
[23] R. Gan, A. C. Chung, C. K. Wong, and S. C. Yu, “Vascular
segmentation in three-dimensional rotational angiography based on
maximum intensity projections,” in Proc. IEEE Int. Symp. Biomed.
Imag., 2004, pp. 133–136.
[24] S. Hu and E. A. Hoffman, “Automatic lung segmentation for
accurate quantization of volumetric X-ray CT images,” IEEE Trans.
Med. Imag., vol. 20, no. 6, pp. 490–498, Jun. 2001.
[25] Vijayakumar, B., and Ashish Chaturvedi. "Automatic Brain Tumors
Segmentation of MR Images using Fluid Vector Flow and Support
Vector Machine." Research Journal of Information Technology 4.
[26] Vijayakumar, B., and Ashish Chaturvedi. "Tumor Cut-Segmentation
and Classification of MR Images using Texture Features and Feed
Forward Neural Networks." European Journal of Scientific Research
85.3 (2012): 363-372
[27] Vijayakumar, B., and Ashish Chaturvedi, “Idiosyncrasy Dissection
and Assorting of Brain MR Images Instigating Kernel Corroborated
Support Vector Machine,” Archive Des Science
[28] Vijayakumar, B., and Ashish Chaturvedi,“Abnormality segmentation
and classification of brain MR images using Kernel based Support
vector machine,” Archive Des Science,Volume 66,Issue 4.
[29] B. Vijayakumar, Ashish Chaturvedi and K. Muthu Kumar, “Brain
Tumor in Three Dimenesional Magnetic Resonance Images and
Concavity Analysis”, International Journal of Computer Application,
Issue 3, Volume 1 (February 2013).
ISSN: 2231-5381
http://www.ijettjournal.org
Page 100