Classification of MR medical images Based Rough-Fuzzy KMeans

IOSR Journal of Mathematics (IOSR-JM)
e-ISSN: 2278-5728, p-ISSN: 2319-765X. Volume 13, Issue 1 Ver. III (Jan. - Feb. 2017), PP 69-77
www.iosrjournals.org
DOI: 10.9790/5728-1301036977 www.iosrjournals.org 69 | Page
Classification of MR medical images Based Rough-Fuzzy K-
Means
Ahmed Mohamed Elmoasry (
*)
Mathematics Department, College of science, Majmaah University, KSA
(
*)
Mathematics Department, Faculty of science, Aswan University, Egypt
Abstract : Image classification is very significant for many vision of computer and it has acquired significant
solicitude from industry and research over last years. We, explore an algorithm via the approximation of Fuzzy
-Rough- K-means (FRKM), to bring to light data reliance, data decreasing, estimated of the classification
(partition) of the set, and induction of rule from databases of the image. Rough theory provide a successful
approach of carrying on precariousness and furthermore applied for image classification feature similarity
dimensionality reduction and style categorization. The suggested algorithm is derived from a k means classifier
using rough theory for segmentation (or processing) of the image which is moreover split into two portions.
Exploratory conclusion output that, suggested method execute well and get better the classification outputs in
the fuzzy areas of the image. The results explain that the FRKM execute well than purely using rough set, it can
get 94.4% accuracy figure of image classification that, is over 88.25% by using only rough set.
Keywords: Rough theory; fuzzy theory; K-means; Image classifications; Uncertain Images; RGB images.
I. Introduction
Image segmentation still challenging tasks in image analysis (processing). And it has more benefits in
pattern recognition field. Image processing search with the extrication of defect information [13, 27, 28]. Image
division is begin to be more important for medicinal diagnosis process. Currently, expansion of an active
computer assisted diagnosis suit which, assist the radiologist thus became very interested. The aim changed
from replacing the radiologist into over a second opinion. [25, 26]. Consequently, efficient research on
lineaments extracted and their function to classification processes make researchers select features arbitrarily as
involvement to their regulations. In that process the image is divided into diverse regions, even though, they
have the same features. Numerous approaches of image segmentation are occurred. Edge-based method, zone-
by mechanisms and threshold-build on mechanisms and so on. Images are separated regarding to their overall
feature division by grouping and image segmentation manners. We could present a segmentation method based
on K-means using rough theory jointly with fuzzy set.
Since the nineties of the previous century, there exist a rapid evolution in size, uncertainty and intricacy
of electronic data being collected and stocked, For model great number of different of medicinal mechanism are
occur, which produce thousands of medical images per a day.The high volume of MR medical images is so
difficult to be read by physicians, the precision rate resort to decrease, and automatic reading of digital MR
medical are turn to very attractive Images. That is why the computer evaluated diagnosis methods are vital to
advocate the medical tools to get best performance and advantages [1]. Based on this increase of huge data, the
task of extracting significate information in the sort of patterns, relationships and applied in implementation as
decision support, appreciation and classification has become hard and essential. Furthermore the necessity to
explore implicit data structures in disordered attribute data calls for effective data analysis with minimal human
interference. We call that, the arithmetic theories are master tools for perform these goals. We consider that, the
master problem in image classification and collaboration with the huge magnitude of information is the high
overlapping between Image classes and uncertainty.Many attempts achieved to investigate into and study this
problem. In [1] a techniques of information is utilized in detection to discover and assort oddity in breast cancer.
They use neural system of networks and series rules as the data mining algorithms. Classification (partition)
results were of 81.25% with neural system and 69.11% with grouping principles.The noticeable advantage on
the neural framework is the high quality of classification, but it require more time for training contrasted to
different methods. Accordingly, it is significant to diminish the neural framework practice period and modulate
the classifying accuracy simultaneously. In [2]) A Bayes theory is utilized to search and distinguish interesting
regions. In [3] a manner for merit definition of disadvantage in breast using a border based classification
algorithm has presented. In [4], authors find an evaluation of different manners that enable to get textile
attributes from zones of attention extracted from mammogram images. In [5], catch sight of ripple
transformation, utilized discover micro-groups of calcifications in medicinal images. In [6] a cooperative
method of neural system, deploy to distinguish medicinal image. This classifier technique is constructed on the
backpropagation neural algorithm jointly rough theory. In addendum, some other techniques were offered using

Classification of MR medical images Based Rough-Fuzzy K-Means
fuzzy theory [7] and Markov models [8].Even though this effort, still not exceedingly used mechanism to
distinguish medicinal images.For the reason that, this medical section requires high classifying accuracy and
low training time. In this project, a new mechanism have made within medicinal images classification using the
backpropagation K-Means jointly the rough reduction theory which we indicate fuzzy Rough K-Means ( ). It is
tested on real datasets MIAS [9], interpret the images of mammographic and integrate 92.4% accuracy of
classifying which is better than 85.25% using back-propagation neural network algorithm purely in [9].
Simultaneously, the time of practice is decreased distinctively.
Image classification [10] and [11] is very importing function within image operating together with a
quite necessary preprocessing stage in problems on the zone of image operating, computer visualization, and
model assessment. Medical image classification [10, 11 and [12] is an intricate and defiance job, in order that,
the essential ease of the imagesThe brain has especially more rough explication and partition is very significant
for detecting fibroid necrosis and edema tissues so as to prescribe appropriate therapy [10] and [13]. Most
papers medicinal image classification relates to its application for MR images, particularly in brain imaging.
Because of its merit to integrate variance from a size of tissue parameters, more different pulse sequences occur
for gaining MR images [14], [15] and [16]. Detecting the optimal pulse sequence for integrating precise
classification (12) is therefore an engrossing problem that needs information of the fundamental problem
features (properties or attributes) of anatomy to be classified [10], and [11]. MR images is a significant
characteristic of a particular species imaging technicality for the sake of early discovering of abnormal variance
in tissues and organs. Many techniques almost done for medicinal image classification [10], [15], [11], [14],
[13], and [17]. Rough Theory offered by [18] is a tool to analyze fuzzy and incertitude steady in integrating
decision [19], and [20]. It does not rely on as sessional knowing out of data set, and it test and detect certain
relation among data just from the point of view of data's disgraceful attribute [14], just using the denotation of
maximal and minimal approaches of a class, as quite as approximation space and forms of sets [18]. [21]
Presented a manner which constructs on this view, but uses a rough classifier only. The rough classifier feels
back probabilities that an assured pixel (element) belongs to a assured class. The likelihood knowledge could be
benefit by uncertainty categorization. By employing the analyses aforementioned, we could concerned with
rough theory in to brain image classification. Experiential results explain the Highness of rough method of brain
classification.
We established a classification manner that let us to assessment unbelief in the data. The classification
method is locate on rough theory. That segmentation bring into play to approach the minimum and maximize
classes as probabilities. With the data, by which, we could envisage the classified images as the uncertainty for
any class. The main conception underlying the rough approximation to data classification is to discover, what
range sub-group of given objects (in our context: subdivision of the pixels) appreciates another class of items of
concern. Objects are equated by considering their characterizations. An object features redacted as a
multidimensional of function values representing the object's features or attributes [22]. This property used
while we apply rough approximation to multichannel imaging data.
II. Preliminaries
2.1 Rough Set
Rough theory is stratified for former time to diminish the original feature sets and get classifications rules. In the
2nd
stage the back-propagation k-mean used beside rough set to obtain classifications of medicinal image.
2.1.1 Rough Theory
Rough theory is belt on the idea of approximation domain and approximation models for sets and
concepts [18], [19], and [20]. Let that the data is tabulated form, called the decision data table. The (row lines)
of a data table represented to the objects (in our situation: pixels) and the vertical line (columns) represented to
the features (in our situation: pixel attributes). For information of data training, we take sub-set as a sample of
objects. We consider that each samples object has a section tag assigned to it to indicate to which class it
belongs. We name the attribute label a decision attribute, the residue of features (attribute) are called conditional
attributes. Suppose U indicate the group of all sample space (objects) and A the group of all properties of that
objects (conditional features).
2.1.2 Information system
An information or data system of a phenomena, is an order pair , where is the group of objects and A is
the group of properties of that objects [23]. Each implies a function where is the group of
values that property a may have.
In applications, we permanently discern between conditional , and decision features , where .
Often, we define decision systems let and .Furthermore, let denote the equivalent
class deﬁned as
}~:{][ yxyx BB  (1)

The relationship ∼B is nominated similarity relation. It constructs the base of rough theory. So, ∼B is a
subdivision of all classes that have matching detailing.
2.1.3 Similarity relation
Each subcategory of features put together a similarity relation (or same)
for every there exist equivalence class in division of
U is generated by ~B.
2.1.4 Rough Approximation
Suggest an informative data system and . Suppose we interpret two processes setting
to each , two sets and , denoted the B- minimize and the B- maximize approximation of X,
respectively, and interpreted within next relations:
UX
XxBxBxB

 })(|)({)(*
 (2)
UX
XxBxBxB

 })(|)({)(* (3)
Hence, the B-upper set approximation is the coalition of all B- equivalent classes that joint the set, whereas the
B-lower set approximation is the combination of every B-equivalent classes that are involved in the set [24].
The set
)()()( *
*
XBXBXBNB  (4)
Will be indicated to as the B-border region of X. If the border region of X has no element, i.e., ,
then is friable (precise) with regarding to B; in the inverse state, i.e., if , then is indicated to as
rough (imprecise) with regarding to B. Thus, the group of objects is rough (imprecise) if we can't decide it by
the denotation of data or information, i.e. it has some objects that could be set to either as member of the sub-set
or its completeness in vision of given information.
2.2 K-means
When applying the rough classification to medical image, we get involves domains of ambiguity. If we
desire to get a decision for elements in the ambiguous area or if we desire to figure a probability for those
elements closeness to an assured class (or segment), we desire to couple our rough classification with k-means
segmentations. Consequently, to do this, we compute the means into every feature in each rule. For those
elements that are in the ambiguous regions, we measure the dimensions
(5)
While xj indicates the property value of object x and kij denotes the intermediate amount of the
property for the class. Then, object x could be set to the nearest suitable rule or code with regarding to the
defined distance. That procedure itself would be bring to bear in case of inconsistencies, i.e., when there exist
moreover matching rule for an object. The defined distances represent probabilities. Every matching rule
contributes probabilities to its conclusion weight. The probabilities would combined class with powerful
magnitude of probabilities is chosen. Quality assessment similar to support, strength, accuracy, and coverage
(20), attached with the decision rules can be bring to bear for reducing subgroup of decision rules.
2.2.1 The Algorithm of Rough theory and K-means (RKM)
The algorithm of rough theory and K-means (RKM) is composed with two steps. The first is the
attributes decrease using rough information gain theory. The second is back- promulgation algorithm. Imagine
which all numeric attributes have gotten discrete.
Algorithm 1: Rule generation
Input: the Information data system and reduct
Output: decision ruling (R)
1: for do
2: for do
3: ;
4: end for
5: ;
6:
7: end for
8: Regain RULES (R)

2.2.2 K-means algorithm
K-Means basic algorithm be created of next proceedings:
• Initialize
• loop till ending stipulation is met:
1. For every element in that image, stat that cell to a class which, the dimension from this cell to center or
class mean, that, is minimized.
2. The center or class mean of every class Recomputed, derive from pixels which, be owned by it.
• End loop;
2.3 Rough classification
Given an informative system, we can apply rough theory to compute minimize and maximize approximations as
fully as possible, negative and positive regions and boundaries. The calculation stage which be in need have
achieved for a rough segmentation is to formulate the rules which do just as an exact classifier. That rules are
utilized as a base of notation of best cuts.
2.3.1. The Best cuts
• Features selection
Let IS = (U, A ∪ {d}) be an informative system with conditional features A and attributes of decision d.
To inform an informative system from an image, first we get samples out of image which we aspire to classify
(to the smallest extent or degree sample for each class, then we add more channels to (red green blue) color
like * Lab colors, mean for rgb within the 8-neighborhood, variance, standard deviation, etc. then we could
compute all probable cuts as following:
Where }{}....{ 21 Ua (x) : xvvv a
n
aa
 and nna 
Subsequently the group of whole possible cuts on A is denoted by:
)}
2
v+v
,a(,),
2
v+v
,a(),
2
v+v
,a{(=C 13221
a
a
n
a
n
aaaa

 (6)
Class respecting to all probable cuts on all features is denoted by: Aa
aA cC


1) 2.3.2. Find Best cuts
Concept for searching best cut is to seek for a cut which recognizes largest amount of pairs of elements.
Let where Xi is partitioning U, thereafter we compute for all the size of pairs of
elements WX
(a, c) which discerns by each cut of C A as following,
(7)
are numbers of elements out of X be suited to the j th
decision class and being on the left and right-handed- side
of the cut (a, c) (correspondingly). Choose cut that, differentiate the major size of pairs of
elements in L. let in C max inside BCA, then eject it out of CA. Eject all pairs of elements out of L recognized by
C max.
2.4 Algorithm for classify MRI medical image
Input: The consistent decision table A.
Output: The semi minimum subset of cuts P consistent with A
Data Construction: D: the semi minimum subset of cuts;
L = PART (D): the partitioning of U nominated by D;
CA: group of all probable cuts on A.
Method: D = f; L =U; CA = group of all probable cuts on A Compute the value WD(a, c) for all cuts from CA
and seek for a cut (a∗, c∗) which maximize the function WD(•), i.e.
(a∗, c∗) = arg (a, c) max WD(a, c) set D = D∪(a∗, c∗); CA =CA−(a∗, c∗) (8)
To whole X ∈ L do;
If X be formed of elements from one decision class therefore eject X out of L;
If (a∗, c∗) divides the set X to X1, & X2
Thereafter eliminate X away from L
Join the two sets X1, & X2 to L
If L is become empty thereafter
Stop.
Else Go to 2.
End.
2.5 Fuzzy-Rough Sets
In real enforcement, knowledge is either rough (fuzzy) or exact-valued, for that reason, traditional
rough theory enter a problem. It wasn't available in the traditional theory to say which two attribute values are

comparable and the how much comparable percent; i.e.; two relative values may only disparity as an outcome
of ambiguity, but RST deems them as distinct as two values of a various volume. So that, there exist a demand
to develop mechanisms that, produce a process for knowledge formation of fuzzy and exact-value attribute data
which use the size to which values are indistinguishable. That could be done over use fuzzy jointly rough
theory. Fuzzy jointly rough sets contain the concerning but distinguished concept of vagueness (for fuzzy
theory) and the similarity (for rough theory), together take place with reason of distrust data. A T-function fuzzy
likeness relationship is utilized for approach fuzzy X low and up approach are:
 )(),,(inf)( yyxIx xRUyXR PP
  (9)
 )(),,(sup)( yyxIx xRUyXR PP
  (10)
Where, I is an ambiguous (fuzzy) and T is a norm. RP is the fuzzy likeness relationship produced by the group of
attributes P:
 ),(),( yxTyx aP RPaR   (11)
is the account to that, elements x & y are likeness for attribute a, and might be determined in
numerous ways. In an alike way to the main ambiguity rough approach, the fuzzy assured region would be
indicated as:
)(sup)( /)( xxPOS XRDUXDR PP
  (12)
A substantial matter in knowledge dissection is the detection of independencies among features. The rough-
fuzzy dependency grade D to subset of attribute P would be appointed as:
U
X
D UX DR
P
P 

)(
)(
)(
 (13)
A fuzzy-rough reduct R would be indicated as a least number of attributes that, save the dependency grade with
whole knowledge
(14)
2.5.1 Algorithm of fuzzy and K-Means
Classification algorithm of Fuzzy & K-means partitions data points into k classes
and classes are associated with representatives (class center) Cj. The connection among
the information point and class representative is fuzzy. That is, a membership is applied to state the
membership grade of information point Xi into class center Cj. Indicate the group of information points as
. The algorithm of Fuzzy & K--mean is belt on decreasing the next deformation:
(15)
With regard to the representative class and memberships , where N is the stage of information points; is
the fuzzy variable; k is the stage of class; and is dimension between information point and class
representative . It is renowned that ui,j would be content with the next restriction:
The main procedure of is transitive a given group of model vectors into an amended one via
partitioning information points.. It starts with a combination of primary class means and repeats this procedure
till ending contented standard, i.e.; every two classes haven't the same representative class; if there occur two
class means coincide, a class mean should concerned to cancel seashell in the repeated procedure. If ,
then and for , where is a teeny positive number. Now, the fuzzy k-means classification
algorithm is given as follows:-
1. Enter a collection of initial class centers and the amount of. Set p=1.
2. Specified collection of class centers , compute for and . Update
memberships using the following equation:
(16)
If , set , where is a teeny positive number.

3. Compute the mean to any class employing the posterior equation below to get a new combination of class
representatives
(17)
If for to , then stop, where is a teeny number greater than zero.
Otherwise set and move to stage 2. The main complication of Fuzzy K-mean is from stage 2 &
stage 3.
Anywise, the complication computational of step 3 is much less than that of step 2. Thence, the complication
computational, in terms of the figure of distance computation, of FKM is where , is the figure of
iterations is.
III. Proposed Technique
Fuzzy & k-means is one of the classical algorithms available for the classification. Even though this
algorithm is brittle as it permits an element to be happen perfectly in one group. To beat the impediments of
brittle classification fuzzy based classification was introduced. The distribution of element is fuzzy based
manners can be refined by rough classification. Upon on the min and max approximates of rough group, the
rough fuzzy k-means classification algorithm upgrade the compilation of membership function become more
reasonable
3.1 Rough Set Based Fuzzy and K-Means Algorithm
Certain procedure of classification algorithm are defined as:
1. Locate the class figure parameter initial matrix of member function, the upper
approximate limit of class, an appropriate number and .
2. We can figure centroids by:




 N
j
m
ij
N
j
j
m
ij
i
U
XU
C
1
1
(18)
3. If the max approximate, then . Otherwise, update as shown below
(19)
4. If .
3.1.1 Membership of features value
First, initial class centers were produced by randomly selecting points from an image
point set. Where , is the pixels number), for each class centers is
defined by numeric attribute . Then every attribute is given by its membership figure
conformable to three degree fuzzy, defined as, low , medium , and high , which characterized
respectively by a - membership function
(20)
Where is the radius of the -membership function with as the central point. To pick out the center and
radius .Thus, we obtain an initial classification set center wherever each class center is explained by a
combination of fuzzy set.
3.1.2 Decision Table for the Initial class Centers Set
Def. 1: figure of likeness among two distinct categories centers is given as:
Maximize the value of the likeness, the closer the two classification center is.

Def. 2: In a same class centers set, if a class center has a same similarity value to another one, subsequently they
are defined as redundant class mean of each other.
Proposition 1: If are redundant class center each other, are redundant class center each other,
then be owned by to a redundant the selfsame of class center, i.e.
Based on what mentioned above, taking initial class centers as objects, taking class centers features , the
middle degree c and the radius as conditional attributes, taking grade of likeness between two different class
centers as decision attribute by computing the -membership value, thus a decision system for the first
category of class centers can be assigned as:
Where , it indicates an elementary class centers set; is a finite category of the
elementary class center of attributes (where is a collection of condition attributes, is a collection of the
decision attributes); (where is a range of the elementary class center of attribute);
is the redundant data, which defines a likeness relation on .
3.1.3 Eliminating redundant class centers from the initial class centers set
Assuming denotes a decision rule, (condition) and (decision) denote the restriction that
respectively, denotes two distinct class centers respectively, and other assumptions are
as what above-mentioned selfsame. Based on what described above, the initial class means could be optimized
by reduction theory correspond to the following steps:
1. Deducing the compatibility of each rule of an elementary class center set decision table
and , then rules of an
elementary class center set decision table are compatible; if
then rules of an
elementary class center set decision table are not compatible.
2. Ascertaining redundant conditional attributes of an elementary class center set decision table; if an initial
class center set decision table are compatible, then when a redundant
attribute is and it can be leaved out, otherwise can’t be leaved out. If an initial class center set decision
table are not compatible, then computing its positive region if when
then can be leaved out, otherwise can’t be leaved out.
3. Eliminating redundant decision items from an initial class center decision table. For each condition attribute
carries out the aforementioned procedure till condition attribute set does not change. As soon as
redundant initial class centers in the initial class set is eliminated, a reduced class center set is used as the
initial input variance for image segmentation. To figure quality of classes, the index was
used:
(21)
A smaller reflects that, the classes have greater separation with each other and are more compact.
Based on what descript above, now the procedure for Rough Sets based segmentation manner can be
summered as follows:
1. Initiate the figure of classes to randomly, where and is amount of picture points.
2. Randomly select pixels from raw data to be class means.
3. Optimize the elementary class means category by Rough.
4. Put variable and a tiny figure greater than zero .
5. Calculate (at ) or update (at ) the matrix of membership using equation (18).
6. Update the class centers by equation (19).
7. Compute the corresponding index using equation (21).
8. Repeat step 5-8 until
9. Return the best and best center positions.
3 .2 Quality Measures
In request to adjudicate the accuracy and fineness of our segmentations, we should first define the quality
measures. When using synthetic data, the land verity is known and we can figure the segmentation accuracy
Where, is the amount of misclassified pixels and is the overall amount of pixels. For actual data, we can
use other measurements for example compactness and isolation.

3.3 Classification Multichannel Medical Image
We applied our semi-automatic system to assort or classify Multichannel image from the medical
section partition. The data set we used are obtained of a human brain of resolution 336 × 411 × 1004 We have
chosen the 2D slice depicted are be obvious in the Figure. No. 1, for assembling the training set. A small sample
from that slide was selected to choose six different classes interactively. The chosen regions are be obvious in
the Figure. No. 1. We initiated the 21- features and used our classification regarding on rough classification. We
deduce also a fineness (quality) term Q (I) for the classification by integrating the precision (accuracy) figures
for the 6 classes. We use rough set and k-means to assort or classify brain medical MRI into 6 classes.
IV. Experimental Results
User-led classification using brushing in case space and using machine learning style to classify all data
from the brushed selection: (a) Specifying the practice set by interactively Specifying image Areas and
respective classes (or clusters). (b) Medical color imaging data segmentation based on RKM classification,
where blue color is used to view an area of imperfect between the two neighboring classes. (Data set courtesy of
Art Toga, university of California, L.A., USA.) Experimental results on real images are described in detail. In
these experiences, the figure of numerous kinds of elements in the image from manual analyses was treated as
the figure of classes to be indicated. They were also used as the parameter for . The index
value has been utilized throughout to evaluate the quality of the classification for all algorithms. All experiments
were implemented on PC with 3.2 GHz i5 processor using MATLAB (version9.0). Proposed algorithm applied
on all images. This RFKM image classification method partitions into different regions exactly. Visually as well
as theoretically our method gives better results other than state of the art methods like, FCM, RFCM. We present
a classification time of experiment for 2 experiments and shows that RFKM performs better than FCM and
RFCM
Table 1: clustering time of experiment for 2 experiments
Average of the
XB index values
Clustering
time (in sec)
FCM 0.033024 12.64
RFCM 0.030578 5.48
RFKM 0.028197 4.92
V. Conclusion
A cooperative fuzzy, rough and-K-means classifying method applied to medical image classification
has presented. This classifier used the algorithm of K-means backpropagation and rough theory. In this work
rough set information gain reduction theory is applied to decrease the redundancy attributes firstly to lessen time
and computational burden. Moreover, we demonstrated how important the image pre-segmenting phase is in
building a classifier. The evaluation of the fuzzy rough –k-means (FRKM) was carried out on MIAS dataset and
empirical outcomes show that the quality of FRKM reaches 94.4% than 88.43% which execute back-
propagation algorithm itself. Also, to assess the clarity of classes, the index was utilized as class
validity index. Experimental results indicate the superiority of the proposed method in image segmentation.
Acknowledgements
The author would like to thank Deanship of Scientific Research at Majmaah University for supporting this work.
References
[1]. ANTONIE, Maria-Luiza; ZAIANE, Osmar R.; COMAN, Alexandru.Application of data mining techniques for medical image
classification.In: Proceedings of the Second International Conference on Multimedia Data Mining.Springer-Verlag, 2001. P. 94-
101.
[2]. ZHANG, Xiao-Ping; DESAI, Mita D. Wavelet based automatic thresholding for image segmentation. In: Image Processing, 1997.
Proceedings., International Conference on. IEEE, 1997. p. 224-227.
[3]. BOTTIGLI, Ubaldo; GOLOSIO, Bruno. Feature extraction from mammographic images using fast marching methods. Nuclear
Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 2002,
487.1: 209-215.
[4]. SHARMA, Mona; SINGH, Sameer. Evaluation of texture methods for image analysis. In: Intelligent Information Systems
Conference, The Seventh Australian and New Zealand 2001. IEEE, 2001. P. 117-121.
[5]. HAHN, Hee Il. Wavelet transforms for detecting microcalcifications in mammography. 1995.

[6]. BRZAKOVIC, D.; NESKOVIC, M. Mammogram screening using multiresolution-based image segmentation. International journal
of pattern recognition and Artificial Intelligence, 1993, 7.06: 1437-1460.
[7]. RAJ, Amitha; JAYASREE, M. Automated Liver Tumor Detection Using Markov Random Field Segmentation. Procedia
Technology, 2016, 24: 1305-1310.
[8]. YUN, Jiang, et al. A better classifier based on rough set and neural network for medical images. In: Data Mining Workshops, 2006.
ICDM Workshops 2006. Sixth IEEE International Conference on. IEEE, 2006. p. 853-857.
[9]. PETERS, James F. Classification of perceptual objects by means of features. International Journal of Information Technology and
Intelligent Computing, 2008, 3.2: 1-35.
[10]. PARAMESHWARAPPA, Vinay; NANDISH, S. A segmented morphological approach to detect tumour in brain images.
International Journal of Advanced Research in Computer Science and Software Engineering, 2014, 4.1: 408-412.
[11]. SENTHILKUMARAN, N.; RAJESH, R. Edge detection techniques for image segmentation–a survey of soft computing
approaches. International journal of recent trends in engineering, 2009, 1.2: 250-254.
[12]. SHEN, Shan, et al. MRI fuzzy segmentation of brain tissue using neighborhood attraction with neural-network optimization. IEEE
transactions on information technology in biomedicine, 2005, 9.3: 459-467..
[13]. PAWLAK, Zdzisław. Rough sets. International Journal of Parallel Programming, 1982, 11.5: 341-356.
[14]. PAWLAK, Zdzisaw; SOWINSKI, Roman. Rough set approach to multi-attribute decision analysis. European Journal of
Operational Research, 1994, 72.3: 443-459.
[15]. KOMOROWSKI, Jan, et al. Rough sets: A tutorial. Rough fuzzy hybridization: A new trend in decision-making, 1999, 3-98.
[16]. KRYSZKIEWICZ, Marzena. Rough set approach to incomplete information systems. Information sciences, 1998, 112.1-4: 39-49.
[17]. MITRA, Sucharita; MITRA, Madhuchhanda; CHAUDHURI, Bidyut Baran. A rough-set-based inference engine for ECG
classification. IEEE Transactions on instrumentation and measurement, 2006, 55.6: 2198-2206.
[18]. PAL, Sankar K.; SHANKAR, B. Uma; MITRA, Pabitra. Granular computing, rough entropy and object extraction. Pattern
Recognition Letters, 2005, 26.16: 2509-2517.
[19]. KUMARAN, N. Senthil; RAJESH, R. Edge detection techniques for image segmentation–A survey. In: Proceedings of the
International Conference on Managing Next Generation Software Applications (MNGSA–08). 2008. p. 749-760.
[20]. SENTHILKUMARAN, N.; RAJESH, R. A study on split and merge for region based image segmentation. In: Proceedings of UGC
Sponsored National Conference Network Security (NCNS-08). 2008. p. 57-61.
[21]. LAKSHMI, S., et al. A study of edge detection techniques for segmentation computing approaches. IJCA Special Issue on
“Computer Aided Soft Computing Techniques for Imaging and Biomedical Applications” CASCT, 2010, 35-40.
[22]. ELMOASRY, Ahmed; MASWADAH, Mohamed Sadek; LINSEN, Lars. Semi-automatic rough classification of multichannel
medical imaging data. In: Visualization in Medicine and Life Sciences II. Springer Berlin Heidelberg, 2012. P. 71-89.
[23]. PAWLAK, Zdzisław. Rough set theory and its applications. Journal of Telecommunications and information technology, 2002, 7-
10.
[24]. DO VAN NGUYEN, Koichi Yamada; UNEHARA, Muneyuki. Rough set approach with imperfect data based on dempster-shafer
theory. Journal of Advanced Computational Intelligence and Intelligent Informatics, 2014, 18.3.
[25]. BORJI, A.; HAMIDI, M. Evolving a fuzzy rule-base for image segmentation. International Journal of Intelligent Systems and
Technologies, 2007, 28: 178-183.
[26]. REDDY, E. Venkateswara; REDDY, E. S. Image segmentation using rough set based fuzzy K-means algorithm. International
Journal of Computer Applications, 2013, 74.14.
[27]. ALMEIDA, R. J.; SOUSA, J. M. C. Comparison of fuzzy clustering algorithms for classification. In: Evolving Fuzzy Systems,
2006 International Symposium on. IEEE, 2006. P. 112-117.
[28]. RAO, VUDA SREENIVASA; VIDYAVATHI, Dr S. Comparative investigations and performance analysis of FCM and MFPCM
algorithms on iris data. Indian Journal of Computer Science and Engineering, 2010, 1.2: 145-151.
[29]. ELMOASRY, Ahmed. Nanogeneralized-closed sets and Slightly NanoSeparation Axioms. Global Journal of Pure and Applied
Mathematics (GJPAM), 2015, 11.Number 1: 1-8.
[30]. ELMOASRY, Ahmed. -Weak Structures. INDIAN JOURNAL OF APPLIED RESEARCH, 2014, 4.1: 351-355.
[31]. ELMOASRY, Ahmed. Measure space on Weak Structure. IOSR Journal of Mathematics (IOSR-JM), 2014, 10.Issue 1 Ver. I.: PP
54-57.
[32]. KUMAR, R. SARAVANA; ARASU, G. THOLKAPPIA. ROUGH SET THEORY AND FUZZY LOGIC BASED
WAREHOUSING OF HETEROGENEOUS CLINICAL DATABASES.”.
[33]. LINGRAS, Pawan; CHEN, Min; MIAO, Duoqian. Precision of rough set clustering. In: International Conference on Rough Sets
and Current Trends in Computing. Springer Berlin Heidelberg, 2008. P. 369-378.
[34]. ELMOASRY, Ahmed. Topological view for uncertain probability. 2010.
[35]. HALDER, Amiya; DASGUPTA, Avijit. Color image segmentation using rough set based K-means algorithm. International Journal
of Computer Applications, 2012, 57.12.
[36]. ELMOASRY, Ahmed. Bayesian inference on the type II extreme value distribution based on type II progressively censored sample.
Computers and Mathematics with Applications, 2003, 44.11.
[37]. GONG, Zengtai; SUN, Bingzhen; CHEN, Degang. Rough set theory for the interval-valued fuzzy information systems. Information
Sciences, 2008, 178.8: 1968-1985.
[38]. DUBOIS, Didier; PRADE, Henri. Rough fuzzy sets and fuzzy rough sets. International Journal of General System, 1990, 17.2-3:
191-209.

Classification of MR medical images Based Rough-Fuzzy KMeans

More Related Content

What's hot (17)

Similar to Classification of MR medical images Based Rough-Fuzzy KMeans (20)

More from IOSRJM (20)

Recently uploaded (20)

Classification of MR medical images Based Rough-Fuzzy KMeans