International Journal of Scientific Research Engineering & Technology (IJSRET)

Volume 2 Issue 10 pp 642-646 January 2014 www.ijsret.org ISSN 2278 0882

IJSRET @ 2014
An Efficient Brain Tumor Detection and classification using
Neural Network Model
Neetu George, Jayachandran A
PG Scholar Department of Computer Science and engineering, PSN College of Engineering, India
Assistant Professor, Dept of Computer Science and engineering, PSN College of Engineering, India
Abstract - Outlining the tumor contour could be a major
step in planning spatially localized radiotherapy. In this
paper, it propose a fast and efficient tool for
segmentation of solid tumors the proposed uses a
cellular automata (CA) based seeded tumor
segmentation method on contrast enhanced T1 weighted
magnetic resonance (MR) images, that standardizes the
volume of interest (VOI) and seed selection, is projected.
The proposed approach may be a complete tumor
segmentation method on post contrast T1 MR Images
that standardizes the VOI and seed selection, employs
CA transition rules adapted to the problem and evolves a
level set surface on CA states to impose spatial
smoothness. In this system, it establishes the connection
of the CA-based segmentation to the graph-theoretic
methods to show that the iterative CA framework solves
the shortest path problem. Moreover, a sensitivity
parameter is introduced to adapt to the heterogeneous
tumor segmentation problem, and an implicit level set
surface is evolved on a tumor probability map
constructed from CA states to impose spatial
smoothness. Adequate information to initialize the
algorithm is gathered from the user just by a line drawn
on the maximum diameter of the tumor, in order with the
clinical practice.
KeywordsRelay, Selection, Optimization Problem,
Allocation, cooperative, MIMO
1. INTRODUCTION
Segmentation of tumors on medical images isnt
only of high interest in serial treatment monitoring of
disease burden in oncologic imaging, however also
gaining popularity with the advance of image guided
surgical approaches [Zou et al. - 2004]. Outlining the
tumor contour is a major step in planning spatially
localized radiotherapy. On T1 images obtained after
administration of a contrast agent (gadolinium), blood
vessels and also the parts of the tumor, wherever the
contrast may pass the blood-brain barrier are observed as
hyper intense areas.
Region-based active contour models are wide
used in image segmentation. Generally, these region-
based models have several advantages over gradient-
based techniques for segmentation, together with greater
robustness to noise. But, classical snakes had the
problem of beingonly as good as their initialization,
still using level-set snakes in 3D. As the tumor class
doesnt have a strong spatial prior, several small
structures, mainly blood vessels, are classied as tumor
as they additionally enhance with distinction. Ho et.al.
used fuzzy classication of pre and post contrast T1
images to get a tumor probability map to evolve a level-
set snake. Liu et.al. have adapted the fuzzy
connectedness framework for tumor segmentation by
constructing a rectangular volume of interest selected
through identifying the rst and last slice of the tumor
and specifying a set of voxels in the tumor region.
Interactive algorithms have become popular for
image segmentation problem in recent years. Graph
based seeded segmentation framework has been
generalized such that graph cuts (GC), random walker
(RW), shortest paths, and power watersheds [Couprie et
al. - 2009] have been interpreted as special cases of a
general seeded segmentation algorithm, that solves a
minimization problem involving a graphs edge weights
constrained by adjacent vertex variables or probabilities.
In [Sinop and Grady - 2007], the connection between
GC, RW, and shortest paths was shown to depend on
dierent norms: L1 (GC); L2 (RW); L (shortest
paths), in the energy that is optimized. Though it was
reported that the shortest paths and RW produce
relatively more seed-dependent results, it may be argued
that the global minimum of an image segmentation
energy is worth equal to the ability of its energy to
detain underlying statistics of images, and a local
minimum may produce a solution closer to the ground
truth than that of a global minimum.
Therefore, with good prior information provided
as in the case of a seeded image segmentation problem,
International Journal of Scientific Research Engineering & Technology (IJSRET)
Volume 2 Issue 10 pp 642-646 January 2014 www.ijsret.org ISSN 2278 0882
IJSRET @ 2014
eciently nding a good local minima becomes
significant and worthwhile. On the other side, cellular
automata (CA) algorithm motivated biologically from
bacteria growth and competition, is based on a discrete
dynamic system dened on a lattice, and iteratively
propagates the system states via local transition rules. In
this paper, it reexamines the CA algorithm to establish
the connection of the CA-based segmentation to the
graph-theoretic methods to illustrate that the iterative
CA framework converges to the shortest path algorithm.
Next, in the clinical radiotherapy planning, where
manual segmentation of tumors are carried out on CT
fused post contrast T1-MR images by a radio-oncology
expert, it modify the CA segmentation towards the
nature of the tumor properties undergoing radiation
therapy by adapting relevant transition rules. Lastly, a
smoothness constraint using level set active surfaces is
imposed over the resulting CA states.
2. RELATED WORKS
Marcel Prastawa, Elizabeth Bullitt, Sean Ho,
Guido Gerig et al. describes a framework for automatic
brain tumor segmentation from MR images. The
detection of edema is done concurrently with tumor
segmentation, since the knowledge of the extent of
edema is vital for diagnosis, planning, and treatment.
While several other tumor segmentation methods rely on
the intensity enhancement produced by the gadolinium
contrast agent in the T1-weighted image, the technique
proposed here doesnt need contrast enhanced image
channels. The only needed input for the segmentation
procedure is the T2 MR image channel, however it may
make use of any further non-enhanced image channels
for improved tissue segmentation.
Tony F. Chan, Member, IEEE, and Luminita A.
Vese proposed a new model for active contours to
identify objects in a known image, based on techniques
of curve evolution and MumfordShah functional for
segmentation and level sets. Their model may detect
objects whose boundaries arent essentially outlined by
gradient. It reduces an energy that may be seen as a
particular case of the minimal partition problem. Within
the level set formulation, the problem becomes a mean-
curvature flow like evolving the active contour that will
stop on the desired boundary.
Yuri Y. Boykov Marie-Pierre Jolly describes a
new technique for general purpose interactive
segmentation of N-dimensional images. The user marks
certain pixels as object or background to offer hard
constraints for segmentation. Further soft constraints
incorporate both boundary and region information.
Graph cuts are accustomed to find the globally optimal
segmentation of the N-dimensional image. The obtained
solution provides the best balance of boundary and
region properties among all segmentations satisfying the
constraints. The topology of this segmentation is
unrestricted and both object and background
segments may consist of several isolated parts.
3. CELLULAR AUTOMATA APPROACH
A graph consists of a pair , ) E V G , = with vertices
(nodes) V v e and edges V V E e _ e . The weight of
an edge, eij, is indicated by w and is assumed here to be
nonnegative and undirected (i.e., ij wji wij = ). It will
use closed neighbourhood NG[v] where , ) v NG vi e .
The edge weights are similarity measures estimated
using measured data (e.g. voxel intensity) for vertices:
, ) , ) 1 , 0 e = I f wiij and self-similarity 1 = ii w . State
of a vertex , ) s vi s = is specied with a real value
, ) j 1 , 0 e = xi vi x and a label { ; , , FG BG Ij e pair.
Starting with initial states of vertices, in every iteration,
vertices of graph G is updated by the following rule:
j vi vi NG vj for wjixj x if s Si
i i
t
e > =
+1
(1)
Vertex states are initialized by user supplied seeds p
i
P
like:
, ) , ) , ) , ) , ) P v for v s and P v for p l v s
i i i i i
e = e = , 0 , 1
0 0
(2)
However, this will additionally cause an update on vi at
ti ti t > > * , which violates the condition in (3). Then, at
the converged map, there exists a neighbor * vi for
every vertex vi such that:
, ) - - - = li ixi wi si ,
(3)
, ) - - - - - - - = li xi i wi si , and
, ) - - - - - - - - = li xi i wi i wi si ,
(4)
It can follow this path for any vertex until it reaches a
seed that is never updated:
, ) , )
, )
|
|
.
|

\
|
=
[
O vi Pi
i jk i
p l w v s ,
(5)
International Journal of Scientific Research Engineering & Technology (IJSRET)
Volume 2 Issue 10 pp 642-646 January 2014 www.ijsret.org ISSN 2278 0882
IJSRET @ 2014
Thus, this algorithm cuts the graph G to independent
subgraphs for every seed, consisting of spanning trees
with seeds at root nodes.
If it set edge weights depending on similarity of image
, ) R R I 3 :
I e w
jk
B
jk
V =

(6)
where I
jk
V denotes a Euclidean norm on the
dierence between intensities of two adjacent vertices vj
and vk. Maximization the product of wjk s along the
path O becomes equivalent to minimization of the
summation of I
jk
V s along the same path.
, ) v p O I
jk
V is a discrete approximation to a
geodesic or I shortest path between the seed pi to a
voxel vi . Each voxel is then alloted to the foreground
label if there is a shorter path from that voxel to a
foreground seed than to any background seed, wherever
paths are weighted by image content. With this
interpretation, cellular automata algorithm solves the
shortest paths energy form formulated. The major
advantage of using CA algorithm is its capability to get a
multi-label solution in a concurrent iteration. Another
advantage is the local transition rules are simple to
interpret, and its probable to impose prior knowledge,
specic to the problem, into the segmentation algorithm.
Fig.1: (a) The graph is initialized with similarities as
edge weights and vertex values 1 for seeds, 0 elsewhere;
(b-c) intermediate propagation steps for CA; (d) shows
the nal vertex values obtained from CA that may also
be attained as the shortest path from every vertex to a
seed.
SEED SELECTION
As every path, dening the labeling of a vertex
ends at a seed, the eciency of the algorithm may be
increased by choosing the background seeds on a closed
surface around the volume of interest (VOI) as the result
of labeling inside the VOI are equivalent to using the
entire data set.
Robustness to seed selection is a vital property of a
segmentation algorithm, since it is natural to expect
similar results for identical tumor whereas allowing the
user to guide the segmentation process interactively by
imposing constraints in dierent way. In RECIST tumor
response criteria, a common procedure to follow-up
tumor progress is to compute the maximum observable
tumor diameter. The seed selection algorithm uses
identical idea to follow the known clinical routine to that
the clinicians are accustomed to follow the Step:
- Crop the line 15% from every end and thicken to
3 pixels wide to get f
g
seeds.
- Select bounding box of the sphere having 20%
longer of the line as VOI.
4. LEVEL SET ON THE STRENGTH MAP
Smoothing is an important prior in segmentation
of brain tumors from post contrast T1 images, due to
three major reasons: First, an area surrounded by tumor
tissue is considered as a tumor region even the intensity
characteristics probable to be healthy. Secondly, it is
potential to embrace misclassied necrotic regions to
tumor region that are typically surrounded by enhanced
tissue. Lastly, its possible to keep out nearby structures
like arteries that are enhanced by administration of the
contrast agent.
Cellular automata algorithm assigns a label l,
and a likelihood value xi in the interval (0,1] to every
voxel vi . The final indicates how much its probable to
assign one of the labels to the voxel. Remapping values
of the nal map { ; V i xi X e = to the interval (-1,1) for
all voxels in V, it obtain
, )
, ) , )
, )
, ) , )

=
background is l if
X X
X x
foreground is l if
X X
X x
M
i
i
i
i
i
min max
min
min max
min
(7)
with values Mi at a voxel i. Lastly, a level set snake is
evolved on map M with a piecewise constant region
assumption, though by using a local Gaussian kernel to
dene inner and outer regions about the propagating
surface, to get the nal tumor segmentation map.
Steps of the proposed cellular automata based
tumor segmentation algorithm is given in Fig. 2. First,
the user draws a line over the largest visible diameter of
International Journal of Scientific Research Engineering & Technology (IJSRET)
Volume 2 Issue 10 pp 642-646 January 2014 www.ijsret.org ISSN 2278 0882
IJSRET @ 2014
the tumor (a); employing this line, a VOI is chosen with
foreground-background seeds (b); tCA algorithm is run
on the VOI to get a label map and strengths at every
voxel (c); label maps and strengths are combined to get
the signed strength values, i.e. map M, such as contours
have value of zero (d). The map M is accustomed to
evolve a level-set snake. In (d), initial level set contour is
represented in white, and nal evolved contour is given
in black. Comparison to expert segmentation (blue) is
visualized in (e), overlayed with tCA result (red), and
tCA-Level set result (yellow).
Fig. 2. Steps of the proposed tumor segmentation
method: see text for explanations
5. EXPERIMENTAL RESULT
The Defined approach is Processed on various tumor
image and the outputs are shown with better accuracy
and detection.
Fig.4: Learning Model of the tumor type
Fig.5: neural network model training for classification
Fig.6: best training rate with the current system
6. CONCLUSION
It presented a segmentation algorithm for the
problem of tumor delineation that exhibit varying tissue
characteristics. As the change in necrotic and enhancing
part of the tumor after radiation therapy becomes
important, it additionally applied the Tumor-cut
segmentation to partition the tumor tissue more into its
necrotic and enhancing parts.
The projected segmentation algorithm for the
problem of tumor delineation, has solely two main
parameters: + , 1 tumor, { ; bg fg, 1e and mean
curvature term weight in the level set evolution. One
future work embraces optimizing both curvature term
and the tumor sensitivity parameter over a larger
tumor database, though the results over 18 tumors of
varying degrees showed that the algorithm executes with
high overlap ratios even with the xed heuristic values.
One more item is to investigate the issues related to the
VOI and seed selection procedure. The current work
includes assessment of the tumor response to therapy,
that is built on the given segmentation framework.
Strengths of the proposed technique embrace its simple
interaction over a single slice and less sensitivity to the
International Journal of Scientific Research Engineering & Technology (IJSRET)
Volume 2 Issue 10 pp 642-646 January 2014 www.ijsret.org ISSN 2278 0882
IJSRET @ 2014
initialization (demonstrated by lower coefficient of
variation values), its efficiency in terms of computation
time, and robustness with respect to different and
heterogeneous tumor types.
REFERENCES
[1] Zou, K.H., Wareld, S.K., Bharatha, A.,
Tempany, C.M.C., Kaus, M.R., Haker, S.J.,
Wells, W.M., Jolesz, F.A., Kikinis, R.:
Statistical validation of image segmentation
quality based on a spatial overlap index.
Academic Radiology 11(2), 178189 (2004)
[2] Chan, T.F., Vese, L.: Active contours without
edges. IEEE Transactions on Image Processing
10(2), 266277 (2001)
[3] Ho, S., Bullitt, E., Gerig, G.: Level-set evolution
with region competition: Automatic 3-d
segmentation of brain tumors. In: ICPR, vol. 1,
p. 10532 (2002)
[4] Liu, J., Udupa, J.K., Odhner, D., Hackney, D.,
Moonis, G.: A system for brain tumor volume
estimation via mr imaging and fuzzy
connectedness. Computerized Medical Imaging
and Graphics 29, 2134 (2005.
[5] Boykov, Y., Jolly, M.P.: Interactive graph cuts
for optimal boundary and region segmentation
of objects in n-d images. In: ICCV, pp. 105112
(2001)
[6] Grady, L.: Random walks for image
segmentation. PAMI 28(11), 17681783 (2006)
[7] Couprie, C., Grady, L., Najman, L., Talbot, H.:
Power watersheds: A new image segmentation
framework extending graph cuts, random walker
and optimal spanning forest. In: ICCV (2009)
[8] Sinop, A., Grady, L.: A seeded image
segmentation framework unifying graph cuts
and random walker which yields a new
algorithm. In: ICCV (2007)
[9] Szeliski, R., et al.: A comparative study of
energy minimization methods for markov
random elds with smoothness-based priors.
PAMI 30(6) (2008)
[10] Vezhnevets, V., Konouchine, V.: Growcut -
interactive multi-label n-d image segmentation
by cellular automata. In: Graphicon, Novosibirsk
Akademgorodok, Russia (2005) S. Ho, E.
Bullitt, and G. Gerig, Level-set evolution with
region com-petition: Automatic 3-D
segmentation of brain tumors, inProc. ICPR ,
2002, vol. 1, p. 10532.
[11] Y. Boykov and M.-P. Jolly, Interactive graph
cuts for optimal boundary and region
segmentation of objects in n-d images, in Proc.
ICCV, 2001, pp. 105112.
[12] L. Grady, Random walks for image
segmentation, in IEEE Trans. Pattern Anal.
Mach. Intell. , Nov. 2006, vol. 28, no. 11, pp.
17681783.
[13] C. Couprie, L. Grady, L. Najman , and H.
Talbot, Power watersheds:A new image
segmentation framework extending graph cuts,
random walker and optimal spanning forest, in
ICCV, 2009, pp. 731738.
[14] A. Sinop and L. Grady, A seeded image
segmentation framework uni-fying graph cuts
and ran dom walker which yields a new
algorithm, in ICCV, 2007, pp. 18.
[15] A. Criminisi, T. Sharp, and A. Blake, GeoS:
Geodesic image segmen-tation, in Comput.
Vis. ECCV 2008, 2008, vol. 5302, pp. 99112.