+ All Categories
Home > Documents > A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection...

A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection...

Date post: 27-Sep-2020
Category:
Upload: others
View: 0 times
Download: 0 times
Share this document with a friend
16
A Dynamic Approach for Exudates Detection in Diabetic Retinopathy Images Using Clustering 1 P.V. Rama Raju, 2 I.H.S. Mani Sree, 3 P. Krishna Kanth Varma and 4 G. Nagaraju 1 S.R.K.R Engineering College, Bhimavaram, India. [email protected] 2 S.R.K.R Engineering College, Bhimavaram, India. [email protected] 3 S.R.K.R Engineering College, Bhimavaram, India. [email protected] 4 S.R.K.R Engineering College, Bhimavaram, India. [email protected] Abstract Diabetic retinopathy (DR) is a kind of disease that attacks retina of human eye occurs due to diabetes because of this there is elaboration of sugar levels in body. Patient loss his vision due to DR; earlier exposure can diminish the complication of visual detoriation. Existence of micro- aneurysms, cotton-woolspots, hemorrhages and exudates are indication of mild DR .Exudates are foremost signs of DR and can be blocked with a recent diagnosis. Digital fund us image collected from fund us camera helps in analyzing the exudates in prior way. This paper proposes a method which has two essential steps they are coarse segmentation executed by k-means clustering and fine segmentation executed by morphological image processing for disclosure of exudates on retinal images with very low contrast. Firstly contrast limited adaptive histogram equalization technique is used for preprocessing of retinal images. Later segmentation of the processed images is done through K-Means clustering. In order to specify these segmented regions into Non-Exudates and Exudates a special set of features which are based on color and texture are International Journal of Pure and Applied Mathematics Volume 119 No. 18 2018, 751-765 ISSN: 1314-3395 (on-line version) url: http://www.acadpubl.eu/hub/ Special Issue http://www.acadpubl.eu/hub/ 751
Transcript
Page 1: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

A Dynamic Approach for Exudates Detection in

Diabetic Retinopathy Images Using Clustering 1P.V. Rama Raju,

2I.H.S. Mani Sree,

3P. Krishna Kanth Varma and

4G. Nagaraju

1S.R.K.R Engineering College,

Bhimavaram, India.

[email protected] 2S.R.K.R Engineering College,

Bhimavaram, India.

[email protected] 3S.R.K.R Engineering College,

Bhimavaram, India.

[email protected] 4S.R.K.R Engineering College,

Bhimavaram, India.

[email protected]

Abstract Diabetic retinopathy (DR) is a kind of disease that attacks retina of

human eye occurs due to diabetes because of this there is elaboration of

sugar levels in body. Patient loss his vision due to DR; earlier exposure can

diminish the complication of visual detoriation. Existence of micro-

aneurysms, cotton-woolspots, hemorrhages and exudates are indication of

mild DR .Exudates are foremost signs of DR and can be blocked with a

recent diagnosis. Digital fund us image collected from fund us camera

helps in analyzing the exudates in prior way. This paper proposes a

method which has two essential steps they are coarse segmentation

executed by k-means clustering and fine segmentation executed by

morphological image processing for disclosure of exudates on retinal

images with very low contrast. Firstly contrast limited adaptive histogram

equalization technique is used for preprocessing of retinal images. Later

segmentation of the processed images is done through K-Means clustering.

In order to specify these segmented regions into Non-Exudates and

Exudates a special set of features which are based on color and texture are

International Journal of Pure and Applied MathematicsVolume 119 No. 18 2018, 751-765ISSN: 1314-3395 (on-line version)url: http://www.acadpubl.eu/hub/Special Issue http://www.acadpubl.eu/hub/

751

Page 2: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

derived. Morphological operations are done to get the perfect classification.

As disclosure of exudates existing in limited areas can also be identified

using this technique hence this technique appears encouraging.

Key Words:Diabetic retinopathy, exudates, micro-aneurysms,

hemorrhages, k-means clustering, fund us image, morphological image

processing.

International Journal of Pure and Applied Mathematics Special Issue

752

Page 3: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

1. Introduction

Diabetes arises when the body is inadequate to generate or accurately use

insulin. Insulin is the hormone responsible for transforming food into energy for

daily life[1]. In order to reduce the vision loss regular screening is necessary.

Automatic optic disc (OD) detection from retinal fundus images is a very

prominent objective for detection of different types of eye diseases. And using

this step other retinal parts like blood vessels and macula are also identified.

Exudates and hemorrhages are next step identification of abnormalities present

in the retina so that disease severity can be explained for that explanation the

exudates and hemorrhages should be located correctly the OD allows to

construct the coordinate system of retina in effective way hence the correct

position of exudates and hemorrhages are obtained. Optic disc is the major part

and the opening point for the most of the blood vessels that supplies blood to

the retina. In a normal human eye the optic nerve head carries count of 1 to 1.2

million neurons from the eye towards the brain [1]. Due to increase in blood

glucose level their exist some changes in retinal blood vessels which act as a

major cause of Diabetic Retinopathy(DR).Introductory signs of Diabetic

Retinopathy(DR) is identification of exudates. Exudates are yellow-white

lesions with comparatively distinct margins. As the blood vessels get damaged

within the retina which allows the leaking and depositing of the Exudates and

these Exudates are nothing but lipids and proteins. Ophthalmologists detect the

exudates commenced in retina but this exposure is very tough and time taking.

In addition to this, detection can be done manually but it requires chemical

dilation which has very bad impact on patients and again time consuming

process. Henceautomatic screening techniques for exudates is outstanding

process to reduce the time, cost and labor.

Fig. 1(a): Normalretina 1(b): Diabetic Retina

Above figures explain the basic difference between normal and diabetic eye where

fig 1(a) shows the normal retina and fig 1(b) identifies the diabetic retina where

there is clear identification of exudates, hemorrhages, micro-aneurysms. Many

techniques have been proposed to identify the exudates in a given retinal images.

Some of them are discussed below. In [4] proposed a method where

morphological reconstruction techniques are used directly to obtain the exudates

automatically. Another method [5] expresses the detection of exudates, micro-

aneurysms and optic-disc, macula(anatomical structures of retina) are obtained

using a retinal image analysis scheme. But main problem with this method is it

haven’t explained the concept of exact exudates detection hence they considered

some private database of fundus images.[6]This method is very similar to the

International Journal of Pure and Applied Mathematics Special Issue

753

Page 4: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

method explained previously [4] where initially optic disc is segmented later

using morphological image processing techniques main feature exudates are

extracted. In. [7,8] detection of exudates are done by three main steps. Initial

step includes the color normalization which is applied to input image later

contrast enhancement is done then using a clustering approach of Fuzzy c-

means clustering segmentation is exhibited. Finally, the third step includes the

detection of exudates using the neural networks[9] .This method proves more

flexible than other techniques for feature extraction of exudates as it considered

the neural networks. Again [10] which classifies the patch in to exudates and

non-exudates using another technique of Support vector machine(SVM).Fuzzy

c-means clustering followed by morphological techniques are used to detect

exudates[11] .

2. Methodology

The basic work span is pre-processing, Optic disc elimination, exudates

extraction and classification.

Fig. 2: Flow-Chart for Detection of Exudates

In the above Fig.2 clearly shows the stages in proposed method.

(i) Image pre-processing

Because of variations in luminosity, contrast and brightness inside retinal

images it wi make it more complicated to extract the retinal features and other

bright features from exudates in images [12]. Finally to lower this issues Image

Pre-Processing is more convenient to make the image suitable for further

process and it will eliminate noise present in the image and the increases the

illumination combination with retinal images . The image preprocessing is

briefly illustrated below.

International Journal of Pure and Applied Mathematics Special Issue

754

Page 5: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

HSI Color Space Conversion

Red, green and blue components is a RGB image contains M×N×3 array of

color pixels.RGB image is having initial stage of preprocessing and it is

translated to HSI colorspace. The main reason for adoption of HSI color space

is intensity of matrix can be differentiated from other components for important

data to be needed for exudates diagnosis, it contains intensity matrix the

distance will be calculated between optic disk then pixel value and exudates and

non-exudates pixels can be extracted[12]. Hue is a color attribute that represents

a pure color, Saturation gives a measure of the degree it is the part of white

light mixed with the hue .

Median Filter

The Easiest method to overcome extracted noise without blurring sharp edges

is done by using median filter. It is an effective choice for the removal of

noises especially salt and pepper noise and horizontal scanning images. In

image preprocessing, salt and pepper noise is added to intensity band and

median filter of 3X3 size.

Adaptive Histogram Equalization

The retinal fundus image has illumination in non-uniform way i.e.., they have

variations in brightness[12]. In an image comparison with sides the centre of

the image has more brightness and brightness diminishes as the distance

increases from centre. Hence adaptive histogram equalization(AHE)is analyzed

to provide Uniform illumination for entire image. Using the AHE darkest part

of the input image becomes brighter and bright part which high illumination is

remains constant or reduced to provide even illumination hence the resultant

image is uniform illuminated image[12]. The results are shown below in

fig(3).In this figure (a) shows the original retinal image and (b)shows the

original I band And finally (c) clearly identifies the I band after pre-processing

in which there is abundant increase in the intensity levels.

(a) (b) (c)

Fig. 3: Pre-Processing Result

Optic Disc Detection

The Optic Disc (OD)is the bright appearance on the retina and it is circular.

While processing the OD might misdiagnose as exudates because of its high

contrast similar to the exudates. To diminish this issue firstly theoptic disc

should be removed from the image. OD is identified by its high intensity value

International Journal of Pure and Applied Mathematics Special Issue

755

Page 6: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

hence contrast stretching is the best option. It is a linear transform in which

considers Gmin and Gmax where the Gmin assigns the value of 0 to the image

and Gmax assigns the 255 and the remaining gray levels are placed between 0

to 255, enhancing the contrast is mainly originated by covering the entire range

of gray levels. The contrast enhancement in an image is mainly declared using

specific equations taken from paper [13].

Ic1(x,y)=

255

Gmax −Gmin Ip x, y − Gmin (1)

Ip---- preprocessing output image. TheIc1contrast stretched image is shown in

fig 4(a).Now the binarization of image is mainly done using the α1 which is a

simple threshold value where α1=0.9 is considered and the threshold result used

as mask. In order to exclude candidate regions inversion of mask image and

superimposing on the original image is done. As the dilation is important and

required for the morphological reconstruction, and now place the R on the

existing superimposed image is done and given in an equation form as follows,

Ic2=RfI

(Ic1) (2)

The important part of performance is obtained by fitting the contour of marker

image under the image to be masked, for this many dilations of marker image

Ic2under image to be masked fI are continuously imitated[13].Here the

subtraction of original image from the reconstructed image is calculated and the

obtained difference is nothing but the grey level α2thresholding.

Ic3=Tα2

(fI-Ic2) (3)

Automatic detection of threshold is mainly measured using ostu algorithm. By

doing this all the high intensity pixels are retained and rest are removed by fig

4(b).Implementing the closing morphological function ϕ on the resulting image

.A structuring element S1 isused of disc shaped the range is not fixed it can be

varied in this we mainly considered the disc shape of radius six.

Ic4=∅S1(Ic3

) (4)

In general, largest circular area is the brightest part in retinal image and it is the

optic disc. In some situations exudates areas is very huge than the optic disc. In

such cases finding the specific area among all other regions whose shape is

exact circular. Circularity is obtained by

M=4π∗A

p2 (5)

Astatistics of pixels in the elected regions and p sum of pixels perimeter.

Area whose compactness is close to binary one is pure circular one.

Morphological dilation δ on Preferred area where Ic5 ensures the inclusion of all

pixels in OD area.S2 is another structuring element of a flat disc shaped with

radius six.

OD= 𝛿𝑆2 (𝐼𝑐5)(6)

International Journal of Pure and Applied Mathematics Special Issue

756

Page 7: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

(a) (b)

(c) (d)

Fig. 4: Optic Disc Detection

(ii) Image Segmentation

Image segmentation is done by considering the k-means clustering concept. The

retinal image has to be classified in terms of clusters to identify the presence of

exudates.

K-Means Clustering

The technique of clustering has been extensively used in computer vision such

as image retrieval and image segmentation . The objective of clustering is to

group together image samples whose features are similar to each other, as well

as separate the dissimilar images. They only have access to the feature vectors

of the images, no information is given on where to place a particular sample

within a partition or what cluster should it be assigned. K-means is one of the

most widely used and simplest of the clustering algorithms and the most suited

method for clustering in pattern recognition. In this paper, we use k-means to

group mammograms into clusters as a preprocessing step of image retrieval. Let

A = (𝑎1,𝑎2….𝑎𝑛 ) be a set of d-dimensional real vectors to be partitioned into a

set of k( ≤ n) clusters, C = (𝑐1, 𝑐2….𝑐𝑘)[14]. Every data point comes under a

single cluster. The task of k-means is to find a partition that minimizes the sum

of distance functions of the points belonging to the cluster from the empirical

mean of that cluster. Let µ𝑖be the empirical mean of the cluster 𝑐𝑖 . The sum of

distance functions of the points of cluster 𝑐𝑖 is given by:

J(𝑐𝑖)= ||𝑎 − 𝜇𝑖||2

𝑎∈𝑐𝑖(7)

is also called the squared-error of the cluster. The objective of k-means is to

minimize the sum of these values for each cluster 𝑐𝑖 belonging to C,

J(k) = ||𝑎 − 𝜇𝑖||2

𝑎∈𝑐𝑖𝑘𝑖=1 (8)

International Journal of Pure and Applied Mathematics Special Issue

757

Page 8: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

This is the algorithm for K-means clustering taken from this[14]paper. The

steps involved in this technique are illustrated as follows:

Randomly arrange k points into the space defining data points to be

clustered. These k points become the initial centers of the k clusters.

Assign to each data point the cluster that has the closest centroid.

Recalculate the position of the k centroids. The centroid of the cluster is

counted by taking the empirical mean of all data points related to that

particular cluster.

Repeat steps 2 and 3 until the positions of all the centroids do not

change.

K-means clustering is considered in this rather than other techniques like fuzzy

c-means, modified k-means clustering in all those methods k-means is very

efficient for our application of identifying the exudates as this k-means provides

the interpretation of results in easy way and practical working can be done even

some assumptions will be broken hence it is considered as best for making into

clusters.

(iii) Feature Selection and Extraction

Feature is the most important factor plays a prominent role in the area of image

processing. Before appropriating features, various image preprocessing

techniques like thresholding, resizing, binarization, normalization etc. are tested

on the sampled image. Therefore, extraction of features techniques are practiced

to obtain features which will very much cooperative in recognition and

classification of images. Feature selection plays a key role in many pattern

recognition problems such as image classification Features are nothing but the

diseased areas like exudates and non-exudates if in a retinal image exudates has

to be extracted then exudates becomes the feature and remaining areas are

considered as non- features. More features does not always lead to a better

classification performance, thus feature selection is usually performed to select

a compact and relevant feature subset in order to reduce the dimensionality of

feature space, which will eventually improves the classification accuracy and

reduce time consumption feature selection methods can be classified to two

categories: filter models and wrapper models. Models in filters usually

considers the characteristics of feature data which and are computationally

efficient..Hence now depending upon the color, intensity, texture exudates are

mainly extracted from the retinal images. Therefore exudates are mainly

extracted and differentiated using this features. Mainly two features are selected

as input to k-means clustering is as follows:

1. In the retinal image, exudates are separated from other pixels because of

their high intensity..Filtering operation is mainly applied only on green

part which posses high amount of information rather than red and blue

and reduces the denoising and performs the smoothing[13]. Exudates are

identified by application of median filter on green component of RGB

image as the green component posses more information than the

remaining two. The resultant image is subtracted from the original gray

International Journal of Pure and Applied Mathematics Special Issue

758

Page 9: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

scale image. Hence this subtracted image becomes one of the input to

the k-means clustering.

2. Another feature which can be applied as second input to the k- means is

obtained as follows: The retinal images each and every pixel has

different contrast and gray scale values. Exudates contained in the image

are mainly identified by their high grey level and high contrast. For

contrast enhancement CLAHE is applied on green component. There is

a small confusion between exudates and blood vessels as they both are

acquiring the same contrast hence in order to reduce these problem

blood vessels should be eliminated. Hence closing operation ∅is done

for this:

Ig1=∅S3(Ig)(9)

Here Ig is Enhanced green component

For closing operation a structuring element 𝑆3 whose width should be greater

than the blood vessels are considered.In order to differentiate the exudates area

from others. Local variation is selected because the distribution pixel size will

be the direction in separating the exudates from that particular area with the

others since it shows the traits of the closely distributed exudates clusters.

Local Variation is given as

Ig2=

1

N−1 (Ig1

i − μIg1

(x))2i∈Y(x) (10)

N is a number of pixels in Y(x),μIg1

is the mean of Ig1 i and i∈Y(x),x isa set of

pixels in sub-window[13].A window size 9× 9 𝑖𝑠used because good results can

be obtained if this window size is considered there is a chance of losing

exudates if window size is more

Fig. 5: Feature Selection

Hence from the above two features exudates are mainly obtained by

segmentation using k-means clustering algorithm .Scattering of the observations

throughout the image is obtained by a factor called Variance. Small and high

variance explained as: When the observation data is near to the mean and also to

each other the it is small variance in similar to this observation data points are

International Journal of Pure and Applied Mathematics Special Issue

759

Page 10: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

distributed far from each other and also to the mean are considered as high

variance. Execution results have shown that the cluster with least variance with

draw the exudates pixels in huge count. Using morphological reconstruction the

one with lowest variance value is opted as an input for fine segmentation of

exudates. The resulting clusters are as illustrated in Fig 6.

(a) (b)

(c)

Fig. 6: K- Means Clustering Output

Morphological Reconstruction

Morphology is a extensive set of image processing operations that process

images based on shapes. Morphological operations consists of structuring

elements and these are applied to an input image, which creates an output

image of the same size the most essential morphological operations are dilation

and erosion.

As the morphological function identifies the useful way of expansion and

compression of the features depending upon the requirement. Hence

consideration of opening and closing is easy way of making the image useful

for the further proceedings. Imv be the minimum cluster variance. A

morphological dilation operator δis applied on selected clusterof disk shape

structuring element S2to involve all the exudates regions present at boarders

they are excluded as they have the low contrast[13].

Idv =δS2 (Imv )(11)

International Journal of Pure and Applied Mathematics Special Issue

760

Page 11: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

On the marker image Ig1Morphological reconstruction is implemented. This

reconstruction is the successive geodesic dilation under the mask. In this

reconstruction of non-candidate regions which are non-exudates are done and

removing of exudates candidate regions are done simultaneously. The mask

image is retrieved by placingIdv to zero gray level in the original image. The

process explained above is done to identify the exudates contours and to

segregate them from remaining contrasted regions. The output image Idv 1 is as

shown in Fig 7(a).

Idv 1=

0 𝑖𝑓𝐼𝑑𝑣 ≠ 0

𝐼𝑔1𝑖𝑓𝐼𝑑𝑣 = 0 (12)

Idv 1 under Ig1

using morphological reconstruction is then calculated as follows:

Ig3=RIg1

(Idv 1)(13)

The difference of the original image Ig1and the reconstructed image Ig3

is

applied with a simple thresholding operation produces the output image.

Ig4=Tα2

(Ig1-Ig3

)(14)

To obtain the exudates the optic disk OD should be removed whereas the

above explained process detects and also removes the OD from the above image

which gives the final outcome. The exudates thus obtained are as shown in Fig

7(b).

7(a):Diabetic Retinopathy Image

7(b):Exudates

Fig. 7: Exudates Detection

International Journal of Pure and Applied Mathematics Special Issue

761

Page 12: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

3. Results

In table-1 briefly discuss about the retina condition of humans with diabetes and

their stages are shown below. Different retina images are considered in the

below table and depending upon their exudates area diabetic patients disease

severity can be known whether it is mild or severe case.

Below table clearly shows the original image and exudates obtained through it

are compared and depending upon their extreme considerations severity of the

disease is identified.

Table 1: Performance Measure for Exudates Area Calculation and Severity

Identification

4. Conclusion and Future Scope

Initial stage of diabetic retinopathy is identified by exudates. CLAHE is used

for identifying for low contrast images. The second stage image segmentation is

used for enhancement for color image by using K-means clustering that is the

unsupervised clustering algorithm-means clustering is used for more color

information for the result of improvement of classification..Segmentation of

image into exudates and non-Exudates are classified by a special set of

technique called morphological reconstruction. By using this method

extraction of color and texture are obtained by taking some set of features in to

count. Using this way, the exudates are observed and the success rate will be

Retinal

image

Exudates Exud

at-es

area

value

Results Retinal

condition

stages

1310

278

1666

2212

1154

International Journal of Pure and Applied Mathematics Special Issue

762

Page 13: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

high. Micro-aneurysm is detected which is one of the earliest symptoms of

Diabetic Retinopathy can be predicted and its performance can be compared in

the future work.

References

[1] G.S.Annie Grace vimala, S.Kaja Mohideen,”Automatic detection of optic disc and exudates from retinal images using clustering approach”, IOSR Journal of Computer Engineering (IOSRJCE) ISSN: 2278-0661 Volume 2, Issue 5 (July-Aug. 2012), PP 43-48

[2] Dr. P.V. Rama Raju, M.Harikrishnam Raju,M.Chanti:Image identification classification detection and obtaining 3D shape using removal technique and segmentation in international journal of control theory and applications(IJCTA) 5(3),45-52

[3] Dr. P.V. Rama Raju ,T.Madhuri,T.Lakshmi Kanth,V.Usha sri:Human identification by segmentation and enhancement of Sclera using MATLAB in international journal of computational research(IJCER)6(4),2275-2280.

[4] Walter T, Klein J-C, Massin P, Erginay A. A contribution of image processing to the diagnosis of diabetic retinopathy – detection of exudates in color fundus images of the human retina. Transactions on Medical Imaging 2002;21(10):1236–43.

[5] Lalonde M, Laliberte F, Gagnon L. RetsoftPlus: a tool for retinal image analysis. In: Proceedings of the 17th IEEE symposium on computer-based medical systems (CBMS’04). IEEE; 2004. p.542–7.

[6] Sopharak A, Uyyanonvara B, Barmanb S, Williamson TH. Automatic detection of diabetic retinopathy exudates from non-dilated retinal images using mathematical morphology methods. Computerized Medical Imaging and Graphics 2008;32:720–7.

[7] Osareh A, Mirmehdi M, Thomas B, Markham R. Automated identification of diabetic retinal exudates in digital colour images. British Journal of Ophthalmology 2003;87:1220–3.

[8] Osareh A, Mirmehdi M, Thomas B, Markham R. Automatic recognition of exudative maculopathy using fuzzy c-means clustering and neural networks. In: Claridge JBE, editor. Medical image understanding and analysis; 2001. p. 49–52.

[9] Osareh A, Mirmehdi M, Thomas B, Markham R. Classification and localization of diabetic-related eye disease. In: Heyden MNPJA, Sparr G, editors. 7th European conference on computer vision. 2002. p. 502–16.

[10] Osareh A, Mirmehdi M, Thomas B, Markham R. Comparative exudates classification using support vector machines and neural

International Journal of Pure and Applied Mathematics Special Issue

763

Page 14: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

networks. In: Dohi RKT, editor. 5th international conference on medical image computing and computerassisted intervention. 2002. p. 413–20.

[11] Sopharak A, Uyyanonvara B, Barman S. Automatic exudates detection from non-dilated diabetic retinopathy retinal images using fuzzy c-means clustering. Sensors 2009;9:2148–61.

[12] Asha Gowda Karegowda,Asfiya Nasiha,M.A.Jayaram,” Exudates Detection in Retinal Images using Back Propagation Neural Network”, International Journal of Computer Applications (0975 – 8887) Volume 25– No.3, July 2011.

[13] R. S. Biyani and B. M. Patre, "A clustering approach for exudates detection in screening of diabetic retinopathy," 2016 International Conference on Signal and Information Processing (IConSIP), Vishnupuri, 2016, pp. 1-5.

[14] Devang Kulshreshtha, Vibhav Prakash Singh, Ayush Shrivastava, Arpit Chaudhary, Rajeev Srivastava, “Content-Based Mammogram Retrieval Using k-means Clustering and Local Binary Pattern”, 978-1-5090-6238-6/17/$31.00 ©20 17 IEEE.

[15] R.C.Gonzalez and R.E.Woods, Digital Image Processing (3rdedition). Singapore; Pearson Education,2008.

International Journal of Pure and Applied Mathematics Special Issue

764

Page 15: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

765

Page 16: A Dynamic Approach for Exudates Detection in Diabetic ...A Dynamic Approach for Exudates Detection in Diabetic Retinopathy I mages Using Clustering 1P.V. Rama Raju, ... Digital fund

766


Recommended