Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
1Activity
0 of .
Results for:
No results containing your search query
P. 1
A Hybrid Approach for DICOM Image Feature Extraction, Feature Selection Using Fuzzy Rough set and Genetic Algorithm

A Hybrid Approach for DICOM Image Feature Extraction, Feature Selection Using Fuzzy Rough set and Genetic Algorithm

Ratings: (0)|Views: 38 |Likes:
Published by ijcsis
The proposed hybrid approach for feature extraction, feature reduction and feature selection of Medical
images based on Rough set and Genetic Algorithm (GA). A Gray Level Co-occurrence Matrix (GLCM) and Histogram based texture feature set is derived. The optimal texture features are extracted from normal and infected Digital Imaging and Communications in Medicine (DICOM) images by using GLCM and histogram based features. The inputs of these features are taken for the feature selection process. The selected features is solved by using Fuzzy Rough set and GA. These optimal features are used to classify the DICOM images into normal and infected.
The performance of the algorithm is evaluated on a series of DICOM datasets collected from medical laboratories.
The proposed hybrid approach for feature extraction, feature reduction and feature selection of Medical
images based on Rough set and Genetic Algorithm (GA). A Gray Level Co-occurrence Matrix (GLCM) and Histogram based texture feature set is derived. The optimal texture features are extracted from normal and infected Digital Imaging and Communications in Medicine (DICOM) images by using GLCM and histogram based features. The inputs of these features are taken for the feature selection process. The selected features is solved by using Fuzzy Rough set and GA. These optimal features are used to classify the DICOM images into normal and infected.
The performance of the algorithm is evaluated on a series of DICOM datasets collected from medical laboratories.

More info:

Published by: ijcsis on Feb 19, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

02/19/2012

pdf

text

original

 
 
A Hybrid Approach for DICOM Image FeatureExtraction, Feature Selection Using Fuzzy Roughset and Genetic Algorithm
 
J. Umamaheswari
Research Scholar, Department of Computer ScienceDr. G.R.D College of Science,Coimbatore, Tamilnadu, India
Umamugesh@yahoo.com
Dr. G. Radhamani
 
Director, Department of Computer ScienceDr. G.R.D College of Science,Coimbatore, Tamilnadu, India.
radhamanig@gmail.com Abstract 
 
The proposed hybrid approach for featureextraction, feature reduction and feature selection of Medicalimages based on Rough set and Genetic Algorithm (GA). A GrayLevel Co-occurrence Matrix (GLCM) and Histogram basedtexture feature set is derived. The optimal texture features areextracted from normal and infected Digital Imaging andCommunications in Medicine (DICOM) images by using GLCMand histogram based features. The inputs of these features aretaken for the feature selection process. The selected features issolved by using Fuzzy Rough set and GA. These optimal featuresare used to classify the DICOM images into normal and infected.The performance of the algorithm is evaluated on a series of DICOM datasets collected from medical laboratories.
 Keywords- Fuzzy roughest; GLCM; Texture features; Histogram Features and region features.
I.
 
I
NTRODUCTION
Nowadays DICOM image analysis is becoming moreimportant for diagnosis process. This process is not easy wayfor optimal identification and early detection of diseases forimproving the surviving rate. Generally the DICOM image is avaluable and most reliable method in early detection.Different methods of DICOM image featurereduction have been used to solve by statistical methods,texture based methods and feature is extracted by using imageprocessing techniques [3]. Some other methods are based onfuzzy theory [1] and neural networks [2].The lack of systematic research on features extractedand their role to the classification results forces researchers toselect features arbitrarily as input to their systems. Geneticalgorithms have been successful in discovering an optimal ornear-optimal solution amongst a huge number of possiblesolutions (Goldberg 1989). Moreover, a combination of genetic algorithms and fuzzy can prove to be very powerful inclassification problems. Previously genetic algorithms havebeen used either to evolve neural network topology (Stathakisand Kanellopoulos 2006) [4] or to select features ( Kavzoglouand Mather 2002 [5]) but not both at the same time.GLCM, Histogram, level set, Gabor filters, andwavelet transform [6, 7, 8, 9] are the approaches for textureclassification problem. The Gabor filters are poor due to theirlack of orthogonality that results in redundant features, whilewavelet transform is capable of representing textures at themost suitable scale, by varying the spatial resolution and thereis also a wide range of choices for the wavelet function.In medical image analysis, the determination of normal and infected brain is classified by using texture.DICOM and CT image texture proved to be useful todetermine the Normal brain [10] and to detect the braindisease part [11].There is a big problem in selecting the optimalfeatures in medical imaging. The evaluation of possiblefeature subsets is usually a painful task. So the large amount of computational effort is required. Fuzzy roughest and Geneticalgorithm (GA) appear to be a selective approach to choosethe best feature subset while maintaining acceptable featureselection. Siedlecki and Sklansky [12] compared the GA withclassical algorithms and they proposed the GA for featureselection. Fuzzy rough set proved to be the best selectionmethod for optimal classification.A new method for extracting features in DICOMimages with lower computational requirements is proposedand selection percentage is analyzed. The tables provide theuser with all relevant information for taking efficient decision.Thus a synergy of genetic algorithms and fuzzy is used forfeature selection in our proposed method.The remaining paper is organized as follows. Section2 describes the feature extraction process. The featureselection problem is discussed in Section 3, while Section 4contains the experimental results. Finally section 5 presentsconclusion and references.
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 11, November 201185http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
 
II.
 
F
EATURE
E
XTRACTION
 Feature extraction methodologies analyze objects andimages to extract the features that are representative of thevarious classes of objects. In this Work intensity histogramfeatures and Gray Level Co-Occurrence Matrix (GLCM)features are extracted [12].
2.1 Intensity Histogram Features
Intensity Histogram analysis has been extensivelyused. The intensity histogram features are mean, variance,skewness, kurtosis, entropy and energy. These are shown inTable 1.
T
ABLE
1 F
EATURES OF
I
NTENSITY
H
ISTOGRAM
 
The average value of intensity histogram featuresobtained for different type of medical image is given Table 2as follows:
TABLE
 
2
 
I
NTENSITY
H
ISTOGRAM
F
EATURES FOR
M
EDICAL
I
MAGES
 
2.2 GLCM Features
The Gray-Level Co-occurrence Matrix (GLCM) is astatistical method that considers the spatial relationship of pixels, which is also known as the gray-level spatialdependence matrix. The pixel and the adjacent pixel isconsider as the spatial relationship and also another spatialrelationships can be specified between these two pixels.The Following GLCM features were extracted in thispaper : Autocorrelation, Contrast, Correlation, ClusterProminence, Cluster Shade, Dissimilarity Energy, Entropy,Homogeneity, Maximum probability, Sum of squares, Sumaverage, Sum variance, Sum entropy, Difference variance,Difference entropy, Information measure of correlation,information measure of correlation, Inverse differencenormalized.The value obtained for the above features for atypical normal and infected DICOM image is given in thefollowing Table 3,
TABLE
 
3 G
LCM
F
EATURES AND
V
ALUES
E
XTRACTED
F
ROM
N
ORMAL
&
 
I
NFECTED
M
EDICAL
I
MAGES
 
III.
 
F
EATURE
S
ELECTION
 To improve the prediction accuracy and minimize thecomputation time, feature selection is used. Feature selectionoccurs by reducing the feature space. This is achieved byremoving irrelevant, redundant and noisy features whichperforms the dimensionality reduction. Popularly used featureselection algorithms are Sequential forward Selection,Sequential Backward selection, Genetic Algorithm andParticle Swarm Optimization. In this paper a combinedapproach of fuzzy roughest method with Genetic Algorithm isproposed to select the optimal features. The selected optimalfeatures are considered for classification.
3.1 Genetic Algorithm (GA) based Feature selection:
During classification, the number of features can belarge, irrelevant or redundant. So the optimal solution is notoccurred. To solve this problem feature reduction is
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 11, November 201186http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
 
introduced to improve the process by searching for the bestfeatures subset, from the original features.GA is an adaptive method of global-optimizationsearching and simulates the behavior of the evolution processin nature. It is based on Darwin’s fittest principle, which statesthat an initial population of individuals evolves throughnatural selection in such a way that the fittest individuals havea higher chance of survival.The GA maintains a cluster of competing featurematrices. To evaluate each matrix in this cluster, the inputs aremultiplied by the matrix, producing a set of output which arethen sent to a classifier. The classifier typically divides thefeatures into a training set and a testing set, to evaluateclassification accuracy. Generally each feature is encoded intoa vector called a chromosome.fitness = W
A
Accuracy + W
nb
 /N (1)where
 A
is the weight of accuracy and
nb
is the weight of Nfeature participated in classification where N
≠ 0.
 A fitness value will be used to measure the fitness of a chromosome and decides whether a chromosome is good ornot in a given cluster. Initial populations in the genetic processare randomly created. GA uses three operators to produce anext generation from the current generation:
 reproduction
,
 crossover
and
 mutation
.
GA eliminates the chromosomes of low fitness and keeps the ones of high fitness.Thus more chromosomes of high fitness move to thenext generation. This process is repeated until a goodchromosome (individual) is found. The Figure 1 illustrates thefeature selection using the genetic algorithm
.
FIGURE 1 F
EATURE
S
ELECTION USING
GA
The total features extracted are 40. The selectedfeatures using GA method are tabulated as follows:
TABLE
 
4
 
F
EATURE
S
ELECTED BY
G
ENETIC
A
LGORITHM
M
ETHOD
 
The above Table 5 shows the feature selected by GA method.
3.2 Feature selection by Rough Set 
Fuzzy set involves more advanced mathematicalconcepts, real numbers and functions, whereas in classical settheory the notion of a set is used as a fundamental notion of whole mathematics and is used to derive any othermathematical concepts, e.g., numbers and functions [13,14].Rough set theory can be viewed as a specificimplementation of Frege’s idea of vagueness, i.e., imprecisionin this approach is expressed by a boundary region of a set,and not by a partial membership, like in fuzzy set theory.Rough set concept can be defined quite generally by means of topological operations, interior and closure, calledapproximations. The concept of rough set theory is based onthe followings:
3.2.1 Decision Tables
A decision table consists of two different attribute sets.One attribute set is designated to represent Conditions (C) andanother set is to represent Decision (D). Therefore, each rowof a decision table describes a decision rule, which indicates aparticular decision to be taken if its corresponding condition issatisfied. If a set of decision rules has common condition butdifferent decisions then all the decision rules belonging to thisset are inconsistent decisions, otherwise; they are consistent.
3.2.2 Dependency of Attributes
Similar to relational databases, dependencies betweenattributes may be discovered. If all the values of attributesfrom D are uniquely determined by values of attributes from Cthen D depends totally on C or C functionally determines Dwhich is denoted by C
D. If D depends on some of theattributes of C (i.e. not on all) then it is a partial dependency C
D and a degree of dependency (k; 0
≤ k ≤ 1) can becomputed as k = γ(C, D), where γ(C, D) is the consistencyfactor of the decision table. γ(C, D) is defi
ned as the ratio of the number of consistent decision rules to the total number of decision rules in the decision tables.
3.2.3 Reduction of Attributes
Decision tables where feature vectors are the condition(C) and desired values for corresponding classes are thedecisions (D) can also represent classification of featurevectors. Now the dimensionality reduction can simply be
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 11, November 201187http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->