International Journal Of Emerging
Technology & Research
Volume 1, Issue 1, Nov-Dec, 2013 (www.ijetr.org)
Diabetic Retinopathy Detection using Image Processing: A
Survey
Muhammad Waseem Khan
Department of Computer Science, COMSATS Institute of Information Technology
Wah Cantt, Pakistan
Abstract
Diabetic retinopathy (DR) is a diabetes related eye disease which
occurs when blood vessels in the retina become swelled and
leaks fluid which ultimately leads to vision loss. Several image
processing techniques including Image Enhancement,
Segmentation, Image Fusion, Morphology, Classification, and
registration has been developed for the early detection of DR on
the basis of features such as blood vessels, exudes, hemorrhages,
and microaneurysms. This paper presents a review of latest work
on the use of image processing techniques for DR feature
detection. Image Processing techniques are evaluated on the
basis of their results.
Keywords: Diabetic Retinopathy, Fundus Image, Blood
Vessels, Exudes, Hemorrhages, Microaneurysms.
1. Introduction
Image Processing is widely used to diagnose the eye
diseases in an easy and efficient manner. It also supports
Ophthalmologists to screen their patients and to do clinical
study as well[1]. Major eye related diseases that cause
blindness worldwide are Diabetic Retinopathy, Glaucoma,
and age-related macular degeneration. It is found that in
America, almost 950,000 people became blind in 2002 and
2.5 million people have visionary problems due to these
diseases[2].
The structure of blood vessels in eye retina gives
information about changes followed by these retina related
eye diseases [3]. Some of eye features, i.e., vascular blood
vessels, fovea, and optic disc (OD) are used to identify
diabetic retinopathy (DR) and other eye related diseases.
Many screening tools are available to diagnose DR [4],[5].
Digital fundus cameras are used to take the retinal vessel
images; therefore, unnecessary brightness, environment,
and the process of acquisition of fundus image degrade the
image quality to some extent. Hence image enhancement is
always required to improve the quality of desired image.
© Copyright reserved by IJETR
ISSN (E): 2347-5900 ISSN (P): 2347-6079
Certain methods are proposed by researchers to enhance
the quality of retinal images[6],[7, 8].
Some of the image processing techniques used by
researchers in order to diagnose eye diseases including
Image Enhancement[9], Registration, Fusion,
Segmentation, Feature Extraction[10], Morphology, and
Classification [11],[12]. Image registration is used in order
to detect changes in medical images. Different images
captured from different angles are aligned in a single
coordinate system in order to get registered
successfully[13]. Image fusion is used to integrate different
type of information from different images into a single
image. Segmentation[14] is used to divide an image into
multiple regions on the basis of color, intensity, and
objects. Image Classification is used to label the group of
pixels on the basis of grey values or other parameters.
Image analysis is used to make the content of image easy
to understand[15],[16]. Diabetic Retinopathy is one of the
common eye diseases and a reason of blindness in the
world [17].
People in developed countries are mostly become the
victim of DR due to the lack of treatment and required
resources. People with Diabetes are 25% more likely to
become the victim of DR[18]. Features of retinal eye
which are used for the detection of DR are blood vessels,
exudes, hemorrhages, Microaneurysms, and textures[19].
This feature grows when glucose level reaches to
maximum limit and destroys capillaries which then leak
blood in retina[20].
In this paper, several image processing techniques used for
the detection of Diabetic Retinopathy (DR) will be
16
International Journal Of Emerging Technology & Research
Volume 1, Issue 1, Nov-Dec, 2013 (www.ijetr.org) ISSN (E): 2347-5900 ISSN (P): 2347-6079
discussed. Different Lesion detection techniques used for
DR are also presented with appropriate results. Image
processing methods are evaluated on the basis of these
results. In Section II, some of the latest work on the use of
image processing techniques in detection of DR is
presented; it will provide an overview of these techniques.
Section III will finally conclude the paper.
2. Current Challenges, Trends, and Issues
Diabetic retinopathy (DR) is one of the major causes of
blindness in the world. It occurs when diabetes affects the
circulatory blood system of eye retina and damages the
blood vessels in the retina which leads to partial or
complete blindness[21]. The effect of blood leakage from
these vessels creates certain lesions in eye retina, e.g.,
Microaneurysms, Hemorrhages, Neovascularisatio, Hard
exudates, Soft exudates, Cotton wool spots, and venous
loops[22]. Nonproliferative DR (NPDR) and Proliferative
DR (PDR) are two types of DR. Stages of DR can be
classified as Mild NPDR[23], Moderate NPDR, Severe
NPDR, and PDR[24].
Bayesian detection algorithm [25] is used to classify the
changes in retinal fundus image in order to diagnose the
diabetic retinopathy. This method can detect brightness
variation, fundus image artifacts, outliers, and
segmentation errors. Segmentation of optical disk, blood
vessels, and fovea is performed in order to detect
variations in the fundus image. The algorithm can
successfully detect lesions, e.g., Microancurysm, Exudates,
and Cotton wool spots. The algorithm is failed to analyze
vascular changes in the fundus image.
A new hybrid vessel segmentation algorithm [26] having
morphological edge detector with TopHat segmentation
technique is used to diagnose Diabetic Retinopathy. The
algorithm is tested on DRIVE image database having 20
color images. Their algorithm achieves True Positive
Fraction (TPF) of 0.8214 and False Positive Fraction
(FPF) of 0.0941. The algorithm can’t detect small image
region vessels and Neovascularisation.
Color retinal images [27] are segmented by removing
background and noise from the image. It is found that
noisy regions in image consume more processing time.
Therefore, RGB retinal image is transformed to Hue,
Saturation, Intensity (HSI) to perform noise segmentation.
After it, Morphological operations are used to remove
single pixel noise. Diaretdb0 and Diaretdb1 with 219
retinal images are used as retinal image databases to test
the proposed algorithm. 92% images are accurately
segmented.
2-D Gabor wavelet based vessel segmentation [28]
technique is used to enhance the vascular pattern in retinal
© Copyright reserved by IJETR
fundus image. After image enhancement, unsharp filter is
used to remove the blurred effect from the enhanced image
and to sharpen the vascular edges. Then, Morphological
Dilation operator with Canny edge detector is used to get
the final Vessel segmentation mask. Technique is tested on
DRIVE database, average accuracy is 0.9469 with
Standard Deviation of .0053.
K-means clustering [29] and color space features are used
to segment exudates, a lesion of diabetic retinopathy. Pixel
color list data structure is used to enable faster clustering.
In first step, unnecessary fundus mask is removed by
applying thresholding, after it color transformation
constructed two feature spaces, i.e., f1 and f2. The results
of clustering confirm that exudates are bright lesions, and
there color is yellow. Method is evaluated using
DIARETDB1 dataset and sensitivity of 71.96% with PPV
of 87% is achieved.
Bright lesions i.e., Exudates and dark lesions, i.e.,
Hemorrhages with Micro aneurysms (HAM) are detected
on the basis of their intensity values [30]. Firstly, 130
images acquired from DIARETDB0 dataset with their
masks. After it blood vessels are extracted using
morphological techniques[31]. Then, Morphological
Closed and Open operations are used to detect exudates
and optic disk. Finally, green channel is used to detect
HAM. Sensitivity of 93.1% and Specificity of 80.7% is
achieved.
In order to segment the optic disc (OD) from retina image
[32], Morphological and Circular Hough Transform
techniques are used to proposed a new template based
method. MESSIDOR database is used to evaluate the
results. When algorithm is applied on 1200 images of
database, it is found that overlapping is 0.92 as compare to
elliptical and deformable models. The main drawback of
their technique is that it can’t generate performance results
better than Elliptical approach.
Knowledge based rough segmentation algorithm with
unsupervised algorithm [33] is used to detect the retinal
vessels in eye fundus image. Ensemble learning based
segmentation is used to segment large vessels, after it
curve fitting technique is used to detect thin vessels.
DRIVE database is used to evaluate the results, sensitivity
of 0.8854 and specificity of 0.9363 is found.
Adaptive thresholding based on pure splitting technique
[34] is used in order to detect exudates in retinal images.
Coarse segmentation is used to find the local variations in
image pixels, i.e., clear border of image candidates.
Morphological operation will then refine the results of
Coarse segmentation. DIARETDB1 database is used to
evaluate results. Sensitivity of 91.2 with specificity of 99.3
is found.
17
International Journal Of Emerging Technology & Research
Volume 1, Issue 1, Nov-Dec, 2013 (www.ijetr.org) ISSN (E): 2347-5900 ISSN (P): 2347-6079
Gradient vector flow (GVF) based segmentation method
[35] is used to segment the optic disc in retinal fundus
image. Mean shift term is used to improve the accuracy of
GVF algorithm. DRIVE database is used to evaluate the
results. Experimental results show that proposed
methodology outperforms classical GVF snakes and level
set segmentation[36] in term of segmentation accuracy.
Lin Lin [37] used threshold based image segmentation,
inscribed parallelogram, and ellipse equation to estimate
pupil center and radius to identify the diabetic retinopathy.
Results shows that algorithm accurately measure pupil size
under eye corners and eyelid occlusion.
Flux mechanism and Graph Cut method [38] are used to
segment the blood vessels in the retina fundus image.
Adaptive Histogram Equalization and Pruning is used to
enhance the blood vessels in first step. After it, Graph Cut
method is used to perform segmentation. DRIVE and
STARE are used as databases to evaluate the technique.
True Positive Rate (TPR) of 89.51% on STARE and
77.6% on DRIVE is found.
2-d Gradient Filter with Intensity Adjustment enhancement
technique [39] is used to detect and segment Exudates in
order to diagnose Retinoblastoma (Rb) and Diabetic
Retinopathy (DR). After statistical analysis, Mean and Std
Dev of 0.0388 and 0.1877 for Retinoblastoma with Mean
and Std Dev of 0.0388 and 0.1790 for Exudative
Maculopathy are found.
Rule based classifier with morphological operations [40] is
used to detect red lesions from retina fundus images.
Median filter is used to increase the visual contrast in pre-
processing step[41]. After it, Morphological technique is
used for Red Lesion detection. Then, Morphological
Closed operation is used to detect blood vessels. Finally,
Rule based Classifier[42] will classify red lesion
candidates. DIARETDB1 database gives sensitivity of
98% and specificity of 86%. Their method fails to detect
red lesions inside fovea during refining process.
Blood vessel and hemorrhage detection algorithm is
proposed in [43]. It classifies the stages of DR into three
different classes, i.e., normal, moderate, and non-
proliferative diabetic retinopathy (NPDR). Blood vessels
are detected using Guassian filter; whereas, Hemorrhage
candidates are detected using density analysis, and
classification is perform at the end. Normal cases are found
to be 90% accurate and NPDR were 87.5% accurate.
K.Narasimhan [44] proposed two algorithms consist of
filtering operations, i.e. , morphological transformation,
and region growing methods for the detection of lesions,
i.e., MicroAneurysms, Haemorhages, and Exudates.
Accuracy of 95% is achieved for DIARETDB1 database
and 92% for a hospital database. Bayessian Neywork with
Support Vector Machine (SVM) is used to classify images.
© Copyright reserved by IJETR
Classification rate of 95% for SVM and 90% for Bayessian
is achieved.
M. Usman Akram [45] used Two dimensional Gabor
Wavelet[46] with multilayered thresholding technique to
proposed a methodology for vessel segmentation to detect
neovascularization, a sign of Proliferative diabetic
retinopathy (PDR). Gabor Wavelet is firstly used for
Vessel enhancement, after it multilayered thresholding is
used to create binary mask for vessel segmentation.
DRIVE and STARE databases are used to evaluate the
proposed technique. Average accuracy of 95% and Std
deviation of 0.03 is found.
Parisut Jitpakdee [47] presents a survey on Hemorrhage
detection to diagnose diabetic retinopathy from retinal
images. Their work review the latest work on common
methods used for detection of hemorrhages, e.g.,
Morphological processing, Neural Network[48],
Classification, Region Growing, and Inverse method.
Comparison of available methods is conduct on three
basis, i.e., Image based, Lesion based, and Pixel based. It
is observed Image based and Lesion based have high
sensitivity but low specificity.
Extreme Learning Machine (ELM) [49] approach is used
to proposed a new methodology for blood vessel detection
in retinal images. The output of pixel classification is given
to ELM. It calculates grey level and fixed moment features
to represent pixels. DRIVE and STARE databases are used
to evaluate the results of proposed method. Accuracy of
90% is achieved.
Feature Extraction [50] method is also used to diagnose
Diabetic Retinopathy. Adaptive histogram approach is
used to extract features. Binary thresholding follows by
morphological operation is used to remove small and
irrelavent objects from the image. After it, boundary
tracing technique is used to detect the optical disk (OD)
boundary. DRIVE and DIARETDB1 databases are used to
evaluate the results on the basis of area, and centroid of
OD.
3. Conclusion
In diabetic retinopathy (DR), retina blood vessels are
damaged due to fluid leakage from these vessels. Different
lesions, i.e., Exudes, hemorrhages, microaneurysms, and
textures are used to detect the stage of DR. In this paper,
use of several image processing techniques for DR lesion
detection are discussed and evaluated. It is found that early
diagnosis of DR can reduce the chance of vision loss upto
50%. Image processing techniques discussed in this paper
can detect the DR accurately. Hybrid methodology should
be used in order to get better result in terms of accuracy
and efficiency for DR detection.
18
International Journal Of Emerging Technology & Research
Volume 1, Issue 1, Nov-Dec, 2013 (www.ijetr.org) ISSN (E): 2347-5900 ISSN (P): 2347-6079
Acknowledgments
I acknowledged COMSATS Institute of Information
Technology, Pakistan to support us in this work.
References
[1] R. ManjulaSri and D. K. Rao, "Review of image
processing and tele ophthalmology research projects
world over," Technology Spectrum Journal, pp. 42-48,
2010.
[2] N. G. Congdon, et al., "Important causes of visual
impairment in the world today," JAMA: the journal of
the American Medical Association, vol. 290, pp. 2057-
2060, 2003.
[3] T. Teng, et al., "Progress towards automated diabetic
ocular screening: a review of image analysis and
intelligent systems for diabetic retinopathy," Medical
and Biological Engineering and Computing, vol. 40,
pp. 2-13, 2002.
[4] U. T. Nguyen, et al., "An effective retinal blood vessel
segmentation method using multi-scale line detection,"
Pattern recognition, 2012.
[5] J. Jan, et al., "Retinal image analysis aimed at blood
vessel tree segmentation and early detection of neural-
layer deterioration," Computerized Medical Imaging
and Graphics, vol. 36, pp. 431-441, 2012.
[6] C. E. Marwan D. Saleh, "An automated decision-
support system for non-proliferative diabetic
retinopathy disease based on MAs and HAs detection,"
Computer Methods and Programs in Biomedicine, vol.
108, pp. 186-196, 2012.
[7] P. Feng, et al., "Enhancing retinal image by the
Contourlet transform," Pattern Recognition Letters,
vol. 28, pp. 516-522, 2007.
[8] E. Grisan, et al., "Model-based illumination correction
in retinal images," in Biomedical Imaging: Nano to
Macro, 2006. 3rd IEEE International Symposium on,
2006, pp. 984-987.
[9] M. Yasmin, et al., "Brain Image Enhancement-A
Survey," World Applied Sciences Journal, vol. 17, pp.
1192-1204, 2012.
[10] M. Sharif, et al., "Microscopic Feature Extraction
Method," Int. J. Advanced Networking and
Applications, vol. 4, pp. 1700-1703, 2013.
[11] R. M. Sri and K. Rao, "Novel image processing
techniques to detect lesions using lab view," in India
Conference (INDICON), 2011 Annual IEEE, 2011, pp.
1-4.
[12] V. Bulsara, et al., "Low cost medical image processing
system for rural/semi urban healthcare," in Recent
Advances in Intelligent Computational Systems
(RAICS), 2011 IEEE, 2011, pp. 724-728.
[13] T. Klinger, Image processing with LabVIEW and
IMAQ Vision: Prentice Hall Professional, 2003.
[14] W. Haider, M.Sharif, Mudassar Raza , "Achieving
Accuracy in Early Stage Tumor Identification Systems
based on Image Segmentation and 3D Structure
© Copyright reserved by IJETR
Analysis," Computer Engineering and Intelligent
Systems, vol. 2, pp. 96-102, 2011.
[15] A. C. Bovik, The essential guide to image processing:
Academic Press, 2009.
[16] S. M. Yasmin Mussarat, Muhammad Sharif,Mudassar
Raza,Saleha Masood, "Brain Image Analysis: A
Survey," World Applied Sciences Journal vol. 19, pp.
1484-1494, 2012.
[17] G. L. Ong, et al., "Screening for sight-threatening
diabetic retinopathy: comparison of fundus
photography with automated color contrast threshold
test," American journal of ophthalmology, vol. 137,
pp. 445-452, 2004.
[18] P. J. Watkins, "ABC of diabetes: retinopathy," BMJ:
British Medical Journal, vol. 326, p. 924, 2003.
[19] R. Acharya, et al., "Automated identification of
diabetic type 2 subjects with and without neuropathy
using wavelet transform on pedobarograph," Journal of
medical systems, vol. 32, pp. 21-29, 2008.
[20] R. N. Frank, "The galactosemic dog: a valid model for
both early and late stages of diabetic retinopathy,"
Archives of ophthalmology, vol. 113, p. 275, 1995.
[21] C. Sinthanayothin, et al., "Automated detection of
diabetic retinopathy on digital fundus images,"
Diabetic Medicine, vol. 19, pp. 105-112, 2002.
[22] K. Mahesh, "A Survey of Automated Techniques for
Retinal Disease Identification in Diabetic Retinopathy
" International Journal of Advancements in Research
& Technology, vol. 2, pp. 2278-7763, 2013.
[23] J. Nayak, et al., "Automated identification of diabetic
retinopathy stages using digital fundus images,"
Journal of medical systems, vol. 32, pp. 107-115,
2008.
[24] O. Faust, et al., "Algorithms for the automated
detection of diabetic retinopathy using digital fundus
images: a review," Journal of medical systems, vol. 36,
pp. 145-157, 2012.
[25] H. Narasimha-Iyer, et al., "Robust detection and
classification of longitudinal changes in color retinal
fundus images for monitoring diabetic retinopathy,"
Biomedical Engineering, IEEE Transactions on, vol.
53, pp. 1084-1098, 2006.
[26] I. Abdurrazaq, et al., "Morphology approach for
features extraction in retinal images for diabetic
retionopathy diagnosis," in Computer and
Communication Engineering, 2008. ICCCE 2008.
International Conference on, 2008, pp. 1373-1377.
[27] M. U. Akram, et al., "Retinal images: Noise
segmentation," in Multitopic Conference, 2008. INMIC
2008. IEEE International, 2008, pp. 116-119.
[28] M. Usman Akram, et al., "Gabor wavelet based vessel
segmentation in retinal images," in Computational
Intelligence for Image Processing, 2009. CIIP'09.
IEEE Symposium on, 2009, pp. 116-119.
[29] K. Ram and J. Sivaswamy, "Multi-space clustering for
segmentation of exudates in retinal color photographs,"
in Engineering in Medicine and Biology Society, 2009.
19
International Journal Of Emerging Technology & Research
Volume 1, Issue 1, Nov-Dec, 2013 (www.ijetr.org) ISSN (E): 2347-5900 ISSN (P): 2347-6079
EMBC 2009. Annual International Conference of the
IEEE, 2009, pp. 1437-1440.
[30] R. S. Himanshu Saxena, "Lesion detection in eye due
to Diabetic Retinopathy " International Journal of
Electronics and Computer Science Engineering, vol. 2,
pp. 963-971, 2010.
[31] I. Irum, M. Raza, and M. Sharif., "Morphological
Techniques for Medical Images: A Review," Research
Journal of Applied Sciences, vol. 4, 2012.
[32] A. Aquino, et al., "Detecting the optic disc boundary in
digital fundus images using morphological, edge
detection, and feature extraction techniques," Medical
Imaging, IEEE Transactions on, vol. 29, pp. 1860-
1869, 2010.
[33] E. Oost, et al., "Vessel segmentation in eye fundus
images using ensemble learning and curve fitting," in
Biomedical Imaging: From Nano to Macro, 2010
IEEE International Symposium on, 2010, pp. 676-679.
[34] H. F. Jaafar, et al., "Detection of exudates in retinal
images using a pure splitting technique," in
Engineering in Medicine and Biology Society (EMBC),
2010 Annual International Conference of the IEEE,
2010, pp. 6745-6748.
[35] H. Zhou, et al., "An improved gradient vector flow
algorithm for optic disc segmentation," in Engineering
in Medicine and Biology Society (EMBC), 2010
Annual International Conference of the IEEE, 2010,
pp. 4793-4796.
[36] M. S. Amir Shahzad, Mudassar Raza,Khalid Hussain,
"Enhanced Watershed Image Processing
Segmentation," Journal of Information &
Communication Technology, vol. 2, pp. 1-9, 2008.
[37] L. Lin, et al., "A robust and accurate detection of pupil
images," in Biomedical Engineering and Informatics
(BMEI), 2010 3rd International Conference on, 2010,
pp. 70-74.
[38] A. G. Salazar-Gonzalez, et al., "Retinal blood vessel
segmentation via graph cut," in Control Automation
Robotics & Vision (ICARCV), 2010 11th International
Conference on, 2010, pp. 225-230.
[39] R. Ravindraiah, et al., "Qualitative and quantitative
analysis of segmentation of human retinal images," in
Computer, Communication and Electrical Technology
(ICCCET), 2011 International Conference on, 2011,
pp. 75-79.
[40] H. F. Jaafar, et al., "Automated detection of red lesions
from digital colour fundus photographs," in
Engineering in Medicine and Biology Society, EMBC,
2011 Annual International Conference of the IEEE,
2011, pp. 6232-6235.
[41] M. Sharif, et al., "Illumination Normalization
Preprocessing for face recognition," in Environmental
Science and Information Application Technology
(ESIAT), 2010 International Conference on, 2010, pp.
44-47.
[42] M. Hameed, M.Sharif, Mudassar Raza, Waqas Haider,
M.Iqbal., "Framework for the Comparison of
Classifiers for Medical Image Segmentation with
© Copyright reserved by IJETR
Transform and Moment based features," Research
Journal of Recent Sciences ,ISSN, vol. 2277, p. 2502.
[43] K. Verma, et al., "Detection and classification of
diabetic retinopathy using retinal images," in India
Conference (INDICON), 2011 Annual IEEE, 2011, pp.
1-6.
[44] K. Narasimhan, et al., "An efficient automated system
for detection of diabetic retinopathy from fundus
images using support vector machine and bayesian
classifiers," in Computing, Electronics and Electrical
Technologies (ICCEET), 2012 International
Conference on, 2012, pp. 964-969.
[45] M. U. Akram, et al., "Automated segmentation of
blood vessels for detection of proliferative diabetic
retinopathy," in Biomedical and Health Informatics
(BHI), 2012 IEEE-EMBS International Conference on,
2012, pp. 232-235.
[46] M. J. J. Muhammad Sharif, Muhammad Younas
Javed,Mudassar Raza, "Face Recognition for Disguised
Variations Using Gabor Feature Extraction,"
Australian Journal of Basic and Applied Sciences, vol.
5, pp. 1648-1656, 2011.
[47] P. Jitpakdee, et al., "A survey on hemorrhage detection
in diabetic retinopathy retinal images," in Electrical
Engineering/Electronics, Computer,
Telecommunications and Information Technology
(ECTI-CON), 2012 9th International Conference on,
2012, pp. 1-4.
[48] M. Yasmin, et al., "Neural Networks in Medical
Imaging Applications: A Survey," World Applied
Sciences Journal, vol. 22, pp. 85-96, 2013.
[49] V. Shanmugam and R. D. Wahida Banu, "Retinal
blood vessel segmentation using an Extreme Learning
Machine approach," in Point-of-Care Healthcare
Technologies (PHT), 2013 IEEE, 2013, pp. 318-321.
[50] B. Sumathy and S. Poornachandra, "Feature extraction
in retinal fundus images," in Information
Communication and Embedded Systems (ICICES),
2013 International Conference on, 2013, pp. 798-802.
Muhammad Waseem Khan is a student of
MS (Computer Science) at COMSATS
Institute of Information Technology, Wah
Campus, Pakistan. He has completed his
Bachelor degree from COMSATS Institute
of Information Technology, Lahore in 2010.
His areas of interest are Image Processing,
Information Security and Mobile Computing.
20