International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
Craniopharyngioma Neoplasm
Incarnation Analysis of Functional MR
Medical Image Classification
J.Vignesh#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— There has been increasing interest in pattern
classification methods and neuroimaging studies using
permutation tests to estimate the statistical significance of a
classifier (p-value). Permutation tests usually use the test error
as a dataset statistic to estimate the p-value(s)by measuring the
dissimilarity between two or more number of populations.
Using the test error as a dataset statistics; however, may
camouflage the lowest recognizable classes,and the resulting pvalue will be biased toward better values because of the highly
recognizable classes;thus,lower p-values could sometimes be
the result of under coverage.In this study,we investigate this
problem and propose the implementation of permutation tests
based on a per-class test error as a data set statistic. We also
propose a model that is based on partially scrambling the
testing samples when computing the non-permuted statistic in
order to judge the p-value’s tolerance and to draw conclusions
regarding, which permutation test procedures are more
reliable. For the same purpose, we propose an other model
that is based on chance-level shifting of the permuted statistic.
We tested the set we proposed models on functional magnetic
resonance imaging data that we recollected while human
subjects responded to visual stimulation paradigms, and our
results showed that these models can aid in determining, which
permutation test procedure is superior. We also found that
permutation tests that use a per class test error as a data set
statistic are more reliable in addressing the null hypothesis
that all classes in the problem domain are drawn from the
same distribution.
Keywords—: ROI, Permutation tests, MRI, fMRI,Multi voxel
pattern analysis and brain tumor
I.
I. INTRODUCTION
The Craniopharyngiomas develop in the area of the brain
called the hypothalamus, which is close to the pituitary
gland. It is usually found in children or young adults and
accounts for around 1% of all brain tumours. The mixed
solid and cystic nature of the tumor is clear on MR images.
By MRI examination, the tumor is of variable T1 signal,
often hyperintense. The T1 hyper-intensity is usually
secondary to high protein content in the cyst fluid. On T2weighted sequences, including FLAIR, the solid portion is
ISSN: 2231-5381
again usually het-erogeneous, whereas the cysts are
invariably hyperintense. Following contrast agent, there is
almost invariable enhancement of the solid portion and the
peripheral rim of the cystic portion on MR image. The
enhancement of the solid portion may be either uniform or
heterogeneous.
Fig 1.1 Craniopharyngioma Neoplasm in the MRI
Machine
variable behavior of protons within different tissues leads to
differences in tissue appearance. The amount of signal
produced by specific tissue types is determined by their
number of mobile hydrogen protons, the speed at which
they are moving, and the tissue’s T1 and T2 relaxation
times [Armstrong et al., 2004], [Lee et al., 2004] (Table 1.1
summarizes the terms used to describe MRI techniques). As
T1 and T2 relaxation times are time dependent, the timing
of the RF pulse and the reading of the radiated RF energy
change the appearance of the image. The repetition time
(TR) describes the time between successive applications of
RF pulse sequences. The echo time (TE) describes the delay
http://www.ijettjournal.org
Page 77
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
before the RF energy radiated by the tissue in question is
measured.
The pulse sequence, which is described by the TR and TE
and indicates the technique used to administer the RF
energy, can be chosen to maximize the effect of differences
in T1 or T2. This gives rise to the description of an MRI
image as T1 or T2 weighted [Stark and Bradley, 1999].The
standard MRI pulse sequence for anatomic and pathologic
detail is a spin echo sequence. T1-weighted images (short
TR, short TE) provide better anatomic detail, while T2
weighted images (long TR, long TE), which are more
sensitive to water content, are more sensitive to pathology.
The intermediate or proton density images (long TR, short
TE) provide improved contrast between lesions and
cerebrospinal fluid.Because of its simplicity, flexibility, and
ease of implementa- tion, pattern classification is
increasingly appearing in many scientific fields.
Classification methods, for example, have been used in
modeling diseases from gene expression data [1,2], in
inferring ethnicity from mitochondrial DNA sequences [3],
in robotic navigation problems [4], and in the decoding of
mental states [5–7]. An important issue in pattern
classification, however, is determining whether the
classifier (and, hence, the classifica- tion) is significant.
This task is usually performed by estimating how much the
test error deviates from chance. There has been recent
interest in this issue, especially in neuroimaging applications and genome analysis studies, where researchers
usually work on constrained datasets that suffer from low
sample sizes and high dimensionality. Because class
distributional properties are unknown and distributional
properties of test statistics are complex, permutation tests
can be used to test for classification significance. These
methods are based on the null hypothesis that classes have
identical distributions; in these cases, an attempt is made to
reject the hypothesis and prove otherwise.
To construct the empirical cumulative distribution, we
com pute, using a dataset statistic, the value t 0 ¼ T ðx1 ,y1
,x2 ,y2 ,. . .,xL ,yL Þ for the actual labels; then, we find the
corresponding p-value (p0) under the empirical distribution
P. If p0 rp, we reject the null hypothesis.
We shall denote the methodology that makes use of the
OATE dataset statistic shown in (1)–(4) as the T1procedure and the p-value calculated with it as p0. The
value of p0 states how likely the observed error would be
obtained by chance
Assume that we choose T as an
appropriate dataset statistic. Let a be the acceptable
significance level, and let PL be the set of all of the
ISSN: 2231-5381
permutations of the indices 1,y, L that are derived from S.
The permutation test pseudo code may be given as follows:
For m=1:M do:
Let permuted ¼{t1,t2,y,tM}, where M is the number of
permutation iterations.
II.
LITERATURE SURVEY
Genetic algorithms offer a powerful search method
for a variety of learning tasks, and there are different
approaches in which they have been applied to learning
processes. Structural learning algorithm on vague
environment (SLAVE) is a genetic learning algorithm that
uses the iterative approach to learn fuzzy rules. SLAVE can
select the relevant features of the domain, but when
working with large databases the search space is too large
and the running time can sometimes be excessive. We
propose to improve SLAVE by including a feature selection
model in which the genetic algorithm works with
individuals (representing individual rules) composed of two
structures. In order to achieve an adequate segmentation,
appropriate initializations for each considered tissue class
are an important requirement. Commonly applied methods
include manual placing of example tissue points6 or
registration of an atlas dataset onto a patient dataset.11
However, different anatomies as well as different shapes,
sizes, and locations between atlas and patient dataset
complicate this task, and a precise registration can take
a significant amount of time. Furthermore, the employed
brain atlas may lead to incorrect initializations, e.g. in case
of large deformations due to tumor growth. Especially,
information about pathological tissue such as tumor is
generally missing in current brain atlases. A common
approach for the modelling of tumor tissue is to provide an
intensity model, assuming a mixture of density functions.
Popovic et al. proposed an a priori tumor intensity model
(or believe map) for the segmentation of calvarial tumors
from CT images.10 In their work, a mixture of Gaussians
generated from a number of training datasets is incorporated
into a level set framework.
III. METHODOLOGY
.
3.1 EXISTING METHOD
Fisher [8] was the first to introduce permutation tests, which
he proposed in the 1930s and called exact tests. Recent
permuta- tion test procedures in classification [9–11]
calculate the classi- fication significance by finding the
cumulative distribution after estimating how much the test
error of the classifier (the non- permuted test error) deviates
from the permuted test error values of the same classifier.
http://www.ijettjournal.org
Page 78
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
yields approximately $2.03eþ59 permutations for the 72
timepoints in each run.
3.2 DISADVANTAGES
Unfortunately, this cumulative distribution does not take
into account the fine details that are related to the error in
predicting each class. It is possible in a two-class problem,
for example, that the classification error is very high for
samples that belong to one class while the error of
classifying samples that belong to the other class is low, yet
permutation tests show significance. This issue would have
additional side effects when determining the significance of
classifying more than two classes. We should mention that
[12] made use of permutation tests to perform statistical
analysis of functional neuroimaging data (i.e., without using
classifiers). We should mention that [12] made use of
permutation tests to perform statistical analysis of
functional neuroimaging data (i.e., without using
classifiers). Many other studies were based on [11], using
permutation tests to estimate the significance of a
classification.
3.3 PROPOSED METHOD
The study presented in [13] emphasizes that studying the
classifier performance via permutation tests is effective;
however, the authors provided a technique that is based on
scrambling the features within the classes to study whether
the classifier exploits the dependency between the features
in the classification. In [10], the authors made use of the
wavelet transform [14] to speed up the permutation tests.
Aside from the research that is presented in [13], all of the
other studies used permutation tests that were similar to the
method discussed in [11].
3.4 ALGORITHM
Permutation scheme
The following are a few points that describe the practical
issues that are related to the permutation scheme that we
used in this study:
1. Instead of permuting the samples (timepoints), we
permuted only the labels, which is more feasible and was
used in all of the previous studies.
Thus, permutation tests were used as a tool, and the
resulting p-values, especially when they showed that the
poorly performing classifier was significant, were
sometimes taken as vital findings. To offer an example, it
was shown in [2] that using a regularized least-squares
classifier may significantly classify healthy controls from
Crohn’s disease subjects from profiles of single nucleotide
polymorphisms (the classification accuracy was $ 0.612,
p¼0.011). It is not possible to judge this p-value because of
differences in sensitivity and specificity, and the situation
would be more complicated in a multiclass classification
problem. Thus, we think that permutation tests may better
reflect the classification significance by taking into account
the error for each class instead of the overall test error. In
this study, we investigate how the per-class classification
error would affect the estimation of the classification
significance and how it may be more sensitive to p-value
changes than using the test error. Therefore, we propose a
dataset statistic that is based on the per-class test error; in
addition, we propose a method for adapting the permutation
test to handle this statistic.
3.5 ADVANTAGES
Comparing two permutation test procedures is not an easy
task because of the randomness of the procedures. To
compare the proposed permutation test procedure, we will
need a model to judge, which of the permutation tests
procedures outperform the procedure most commonly used
in the literature. Thus, for the purpose of comparison, we
propose two models. The first model is based on shifting the
chance level by some value and observing how this shift
affects the characteristics of the p-value, and the second
model is based on partial scrambling the testing samples of
the non-permuted statistic. We run several classification
experiments on functional magnetic resonance imaging
(fMRI) data that are collected while subjects respond to
visual stimulation paradigms
IV.
2. Uniformly distributed pseudorandom numbers were
gener ated for all of the permutations samples that we
used. Then,
these numbers were used to scramble the
labels.
3. For each subject in the dataset that we used, there were
12
runs; each run had eight blocks, with one block for
each
stimulus type (class), and each block had nine
timepoints.
4. We scrambled the labels for each run (e.g., within run
permutations). In this case, we needed to use a multiset
permutation, which, when using a multinomial coefficient,
ISSN: 2231-5381
PERMUTATION ANALYSIS
List of Modules
1.
Permutation Test For Classification
2.
Permutation test pseudocode
3.
The multiclass effect on permutation tests
4.1 Permutation Test For Classification
http://www.ijettjournal.org
Page 79
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
To construct the empirical cumulative distribution, we
com pute, using a dataset statistic, the value t 0 ¼ T ðx1 ,y1
,x2 ,y2 ,. . .,
xL ,yL Þ for the actual labels; then, we find the
corresponding p-value (p0) under the empirical distribution
P. If p0 rp, we reject the null hypothesis.
We shall denote the methodology that makes use of the
OATE dataset statistic shown in (1)–(4) as the T1procedure and the p-value calculated with it as p0. The
value of p0 states how likely the observed error would be
obtained by chance
4.2 Permutation test pseudocode
Assume that we choose T as an appropriate dataset
statistic. Let a be the acceptable significance level, and let
PL be the set of all of the permutations of the indices 1,y, L
that are derived from S. The permutation test pseudo code
may be given as follows:
For m=1:M do:
Let permuted ¼{t1,t2,y,tM}, where M is the number of
permutation iterations.
Construct an empirical cumulative distribution.
4.3 The multiclass effect on permutation tests
When the above OATE dataset statistic is used to estimate
the classification significance via permutation tests, it
suffers from the undercoverage problem because it neglects
the per-class test error, and thus, the p-value sensitivity to
the per-class error is not satisfactory. For example, a twoclass classifier that has an error of 40% will occasionally
yield a rather good p-value because chance level is 50%;
however, the 40% might result from the average error
between class-I error (say 10%) and class-II error (say
70%). Thus, despite the fact that the class-II’s classification
error was 70%, the permutation tests may still show that the
classification is highly significant. The same problem might
also occur when classification problems with more than two
classes are under investigation. A better alternative to
ISSN: 2231-5381
OATE that measures the overall test error would be based
on the Per-Class Test Error dataset statistic (PCTE).
For this case, we propose the PCTE dataset statistic, as
follows:
V. EXPERIMENT RESULT
The name “MATLAB” is derived from “matrix laboratory.”
MATLAB is an application that allows you to create
mathematically oriented programs. It is generally
recommended that you can get the most out of it if you have
familiarity with linear algebra. However, as it is, MATLAB
can be used in many ways, an almost any context, ranging
from the most elementary forms of math to projects pursued
at a doctoral level. It is a calculator, a plotter, and a
programming tool. It has many modules that can be added
to it that address different specializations, such as biology,
physics, and electrical engineering, among many others. At
CU, MATLAB is used in Calculus III labs and other upper
division courses, such as Differential Equations and Linear
Algebra. MATLAB was first developed in the 1970’s by
Cleve Moler, who was on the faculty of the University of
New Mexico.The programming language used with
MATLAB is usually referred to as MATLAB script or Mscript. 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
http://www.ijettjournal.org
Page 80
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
SCREENSHOTS
INPUT IMAGE
Fig: 5.2c Segmented Image Based on Permutation
(Elapsed time is 9.434162 seconds.)
Fig: 5.1 Input Image
7.2 PROCESSING IMAGE
FIGURE 1
SECOND ITERATION
Fig: 5.2d Segmented Image Based on Permutation
(Elapsed time is 9.434162 seconds.)
Fig: 5.2 Segmented Image Based on Permutation
(Elapsed time is 19.558152 seconds.)
Fig 5.3 Permutation Result Based on Cluster Analysis
Fig: 5.2 Segmented Image Based on Permutation
(Elapsed time is 38.995707 seconds.)
ISSN: 2231-5381
http://www.ijettjournal.org
Page 81
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
VI.
CONCLUSION
6.1 CONCLUSION
We showed that permutation tests that use the showed that
permutation tests that use the (overall) test error as a dataset
statistic to estimate the significance of a classification might
suffer from an undercoverage problem. Permu- tation tests
that use a dataset statistic that is based on the per test
error(e.g.,the value of the test error for each class) result in
more reliable p-values and would be more sensitive to slight
changes in classification errors than the commonly used
methods in the literature.The partial label scrambling model
that we proposed is very robust when used in comparing
different permutation test procedures.
6.2 FUTURE WORK
An other model that we proposed in this study is based on
shifting the chance level to verify the significance of the pvalue estimation; this model gave results that lead to similar
conclusions to those of partial label scrambling.The
proposed permutation test procedure and its related dataset
statistic have the same computational complexity order of
the method that is commonly used in the literature. The
same conclusion applies to partial label scrambling and to
chance level shifting models.
REFERENCE
1. Bezdek, J.C., Hall, J.L., Clarke, L., 1993. Review
of MR image segmentation using pattern recognition.
Medical Physics 20, 1033–1048.
2. Borgelt, C., Kruse, R., 2003. Speeding up fuzzy
clustering with neural network techniques. In: Proc. 12th
IEEE International Conference on Fuzzy Systems
(FUZZ-IEEE’03, St. Louis, MO, USA) IEEE Press,
Piscataway, NJ, USA.
3. Boudraa, A.E., Champier, J., Cinotti, L., Bordet,
J.C., Lavenne, F., Mallet, J.J., 1996. Delineation and
quantization of brain lesions by fuzzy clustering in
positron emission tomography computerized. Medical
Imaging Graphics 20, 31–41.
4. Boudraa, A.O., Dehak, S.M., Zhu, Y.M., Pachai,
C., Bao, Y.G., Grimaud, J., 2000. Automated
segmentation of multiple sclerosis lesions in
multispectral MR imaging using fuzzy clustering.
Computers in Biology and Medicine 30, 23–40.
ISSN: 2231-5381
5. Chen, S., Zhang, D., 2004. Robust image
segmentation using FCM with spatial constraints based
on new kernel-induced distance measure. IEEE
Transactions on Systems Man and Cybernetics – Part B:
Cybernetics 34 (4), 1907–1916.
6. Chen, W., Maryellen, L.G., Bick, U., 2006. A
fuzzy c-means (FCM)-based approach for computerized
segmentation of breast lesions in dynamic contrastenhanced MR images. Academic Radiology 13 (1), 63–
72.
7. Chiu, S.L., 1994. Fuzzy model identification based
on cluster estimation. Journal of Intelligent and Fuzzy
Systems 2, 267–278.
8. Chuang, K.-S., Tzeng, H.-L., Chen, S., Wu, J.,
Chen, T.-J., 2006. Fuzzy c-means clus-tering with
spatial
information
for
image
segmentation.
Computerized Medical Imaging and Graphics 30, 9–15.
9. Clark, M.C., Hall, L.O., Goldgof, D.B.,
Velthuizen, R., Murtagh, F.R., Silbiger, M.S., 1998.
Automatic tumor-segmentation using knowledge-based
techniques. IEEE Transactions on Medical Imaging 117,
187–201.
10. Dubes, R.C., Jain, A.K., 1980. Clustering
methodology in exploratory data analysis. In: Advances
in Computers. Academic press, New York, pp. 113–225.
11. Fletcher-Heath, L.M., Hall, L.O., Goldgof, D.B.,
Murtagh, F.R., 2001. Automatic seg-mentation of nonenhancing brain tumors in magnetic resonance images.
Artificial Intelligence in Medicine 21, 43–63.
12. Gelbard, R., Goldman, O., Spiegler, I., 2007.
Investigating diversity of clustering meth-ods: an
empirical comparison. Source, Data & Knowledge
Engineering Archive 63 (October (1)), Iris-wine.
13. Gering, D.T., Grimson, W.E.L., Kikinis, R., 2002.
Recognizing deviations from nor-malcy for brain tumor
segmentation. In: Medical Image Computing and
Computer-Assisted Intervention – MICCAI. Springer,
2488, pp. 388–395.
14. Girolami, M., 2002. Mercer kernel-based
clustering in feature space. IEEE Transac-tions on
Neural Networks 13 (3), 780–784.
15. Gustafson, D.E., Kessel, W., 1979. Fuzzy
clustering with a fuzzy covariance matrix. In:
http://www.ijettjournal.org
Page 82
International Journal of Engineering Trends and Technology(IJETT) - Volume 4 Issue 1-Jan 2013
Proceedings of IEEE Conference on Decision Control,
San Diego, CA.
16. Hathaway, R.J., Bezdek, J.C., 2006. Extending
fuzzy and probabilistic cluster-ing to very large data
sets. Computational Statistical and Data Analysis 51,
215–234.
25. 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,
17. Hathaway, R.J., Hu, Y., 2009. Density-weighted
fuzzy c-means clustering. IEEE Trans-actions on Fuzzy
Systems 17 (February (1)).
18. Ichihashi, H., Honda, K., 2005. Fuzzy c-means
classifier for incomplete data sets with outliers and
missing values. Proceedings of the International
Conference on Computational Intelligence for
Modelling Control and Automation, and Inter-national
Conference on Intelligent Agents 2, 457–464.
19. Kannan, S.R., 2008. A new segmentation system
for brain MR images based on fuzzy techniques.
Applied Soft Computing 8 (4), 1599–1606.
20. 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.
21. 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): 363-372Krishnapuram, R., Kim, J., 1999.
A note on the Gustafson–Kessel and adaptive fuzzy
clustering algorithms. IEEE Transactions on Fuzzy
Systems.
22. Vijayakumar, B., and Ashish Chaturvedi,
“Idiosyncrasy Dissection and Assorting of Brain MR
Images Instigating Kernel Corroborated Support Vector
Machine,” Archive Des Science
23. 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
24. 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 83