Professional Documents
Culture Documents
Abstract—Diabetic Macular Edema affects vision and capture the brightness profile ofmaculain [6] to acieve results
eventually may lead to blindness. Early detection is vital to with a hig degree of accuracy.
prevent the ramifications of the disease requiring the need for
effective computer-aided diagnostic techniques. A robust method B. Ramasubramaniam et. al. [7] propose a MATLAB
is proposedin the following paper to segment exudates from applicationfor K-means segmentation of macula and optic disc.
fundus images using a support vector machine for classification. The symmetry of the macular region is utilized to detect the
This is followed by detection of macula using morphological disease severity post classification using global characteristics
features and comparing the segmented exudates against specially in [8]. Macula detection is reported in [9] by taking the
created regions around the macula. The algorithm further grades displacement from the centre of the optic disc followed by
the image according to how severe the disease is. The proposed thresholding. [10] assesses the performance of novel
algorithm has been tested on 89 images of DIARETDB1 database algorithms in optic disc and macula detection. T. Garduno-
and gives an accuracy of 92.11% in detecting severe case of DME Alvarado et. al. [11] utilizes template matching to detect
and 90%accuracy for the same when tested on 100 images of macula while [12] has been a useful source to assess the
MESSIDOR database.The proposed method can be used to accuracy of different methods for macula detection.Issac et al
directly give an idea about the severity of Diabetic Macular [13] [14] has used adaptive intensity based threshold for optic
Edema in the image ad give a reliable and cost-effective diagnosis disc and cup segmentation from fundus images. Yadav et al
in real time.
[15] has used textural features to differentiate normal from
Keywords—Medical Imaging; Diabetic Macular Edema;
affected fundus images. Singh et al [16] has detected glaucoma
Fundus Image; Macula; Exudates from fundus images by analysing them in wavelet domain.
Dutta et al [17] has proposed a method for segmentation of
super pixels followed by Hough Transform which are
I. INTRODUCTION ultimately used for detection of Glaucoma from fundus images.
Diabetic Macular Edema (DME)is a serious disease of the Singh et al [18] has proposed a region growing method for
eyeprevalent in people having Diabetes Mellitus, which can optic disc segmentation from fundus images. S. Ganguly et al
cause rapid vision loss. Early symptoms of DMEoccur when [19] has proposed an adative intensity based gray level slicing
lipids and proteins from the blood vesselsleak in the retina method for segmentation of red lesions from digital fundus
giving rise to yellowish-white exudates. When these coalesce images for detection of Diabetic Retinopathy. Although a lot of
to form hard exudates, and extend into the macular region, they work has been done in this area, still tere exists a need for a
cause Diabetic Macular Edema leading to rapid vision loss. robust algorithm which is able to accurately segment the
Hence, regular check-ups are a must for people susceptible to macula from the fundus images in order to accuratelygrade the
diabetes. Early detection of the symptoms of DME can greatly severity of Diabetic Macular Edema.
reduce the chances of progression of vision loss with timely The mostsignificant contribution of the proposed method is
treatment. Hence, the need for automatic algorithms that can in the use of an adaptive threshold for segmentation of macula
aid in detecting this disease for a timely diagnosis and in turn, from the fundus images.The mean and standard deviation of a
treatment. The treatment options can be temporary in the form particular image has been computed and strategically combined
of Vitreal injections, or permanent, like laser photocoagulation. to calculate the threshold for accurate segmentation of macula.
Since correction can only be done at the early stage of the The use of statistical features for threshold determination
disease, hence, timely diagnosis of the disease is of paramount makes the algorithm image independent. The proposed
importance. algorithm is tested on images of two different databases and is
Work has been continuing to detect macula by academia.A able to correctly segment the macula from the images.
region based method was used to detect Diabetic Macular Another significant contribution lies in the use of
Edema by N. Sengar et. al. [1]. A Naïve Bayes classifier has morphological, geometrical and orientation features to
been used by Z. Y. Zaidi et. al. [2]. Motion pattern generation correctly identify the exudates and macula and create a grading
has been used as the main method in [3] by A. Johny et. al. to algorithm for the computer aided diagnosis of diabetic macular
detect the macula and grade the disease-severity. The reference edema (DME). The results obtained are convincing and the
position of the optic disc to detect the macula along with proposed work can be used in parts in some realtime screening
morpological features has been reported in [4].An interesting systems.
investigative method using Discrete Wavelet Transformation to
localise the blood vessels, and in turn, the macula as been The remaining paper is structured as follows: Section II
reported in [5] while an ingenuous line operator is used to talksabout the image processing and classification techniques
483
The image processing techniques used in the proposed structuring element should be kept approximately equal to the
work for segmentation of exudates are summarized in Fig. 3. It vessel width. This will inpaint only the blood vessels while the
shows the use of normalization, anisotropic diffusion, intensity macula, being of larger dimension, is still retained. After the
based adaptive threshold for exudates segmentation and feature application of morphological operations like closing, the
based classification for false negatives removal to correct process of adaptive thresholding is immediately followed. The
identify the exudate pixels in the image.The features used to image is inverted and the macula is successfully localized in
remove the false positives are statistical, geometrical and the image.
orientation features and are defined as follows:
Input RGB Image
S.No. Feature Approach
1 Mean Statistical Parameter
2 Standard Deviation Statistical Parameter Grayscale Image
3 Variance Statistical Parameter
4 Perimeter Object Boundary Adaptive Histogram Equalisation
5 Area No of white pixels
6 Eccentricity A measure of how elliptical an object Morphological Closing Operation
is
7 Solidity A measure of how solid an object is Intensity Based Thresholding
8 Aspect Ratio Ratio of Major Axis to Minor Axis
Length Background Removal by Mask Multiplication
9 Distance from OD Using coordinates of optic disc
10 Slope with OD Using coordinates of optic disc Macula Localized
Fig. 5. Flowchart for Macula Detection
Fig. 4. (a) Anisotropic diffusion (b) Thresholded image (c) After removal
of optic disc (d) Post classification and removal of noise
484
image, the fundus can be labelled as one with the possibility of The primary purpose of the global database linked to the
having a mild case of DME. An image that has been found MESSIDOR project is to compare and evaluate segmentation
absent of any exudates can be safely assumed to be normal. algorithms developed for the detection of lesions in fundus
Fig. 7 shows the regions which are been used to grade a images. MESSIDOR is a research program funded by the
fundus image for DME. All these regions are centered at French Ministry of Research and Defense. There is a total of
macula and the severity of the disease is more if the distance 1200 images available that have been graded according to the
from macula is less. severity of diabetic retinopathy and macular edema. The
proposed algorithm has been tested on 100 images of the
database available in TIFF format.
The grading algorithm can be summarised in the
following flowchart. Fig. 9 shows the experimental results using the proposed
NO algorithm for exudates segmentation. Fig. 9(a) displays the
Exudates Present Fundus is Normal fundus image input which is in the RGB format. Fig. 9(b)
YES shows the segmented exudates from the input image. This
YES Severe Case of image in binary format. The results shown in the figure
Exudates Present in Region I
DME detected comprises of samples from both the databases, i.e.
NO DIARETDB1 and MESSIDOR. It can be observed that the
YES Moderate Case of segmentation algorithm and proposed set of features used are
Exudates Present in Region II
DME detected capable enough to reject the false pixels and finally segment
NO the exudates correctly.
YES Mild Case of
Exudates Present in Region III
DME detected
Fig. 8. Grading algorithm according to the severity of DME
485
DME which has been used to determine the performance [8] K. S. Deepak and J. Sivaswamy, "Automatic assessment of macular
parameters. The accuracy pertaining to normal images was edema from color retinal images," in IEEE Transactions on Medical
Imaging, vol. 31, no. 3, pp. 766-776, March 2012.
92.72%, for severe case of DME, the accuracy was 90% and doi: 10.1109/TMI.2011.2178856
for moderate DME, the accuracy came out to be 88.89%. The [9] M. Mubbashar, A. Usman and M. U. Akram, "Automated system for
results generated after testing on both the databases have been macula detection in digital retinal images," 2011 International
summarized in Table I. Conference on Information and Communication Technologies, Karachi,
2011, pp. 1-5.
IV. CONCLUSION [10] E. Trucco et al., "Novel VAMPIRE algorithms for quantitative analysis
of the retinal vasculature," 2013 ISSNIP Biosignals and Biorobotics
The algorithmwas tested on 89 images of DIARETDB1 Conference: Biosignals and Robotics for Better and Safer Living (BRC),
database and 100 images of the MESSIDOR database and Rio de Janerio, 2013, pp. 1-4.
successfully provides an accurate idea of the severity of the [11] T. Garduno-Alvarado, M. E. Martinez-Perez, M. A. Martinez-
Castellanos, L. Rodriguez-Quinones and S. M. Salinas-Longoria, "Optic
patients showing symptoms of Diabetic Macular Edema disc and macula detection in fundus images by means of template
(DME). The disease is more prone to vision loss if the location matching," 2014 36th Annual International Conference of the IEEE
of abnormality is near macula. Also, the disease can be Engineering in Medicine and Biology Society, Chicago, IL, 2014, pp.
considered severe if the distance of abnormality from macula 134-137.
[12] R. Veras, F. Medeiros, R. Silva and D. Ushizima, "Assessing the
is less than one-third of optic disc diameter. The proposed accuracy of macula detection methods in retinal images," 2013 18th
method gives a high degree of accuracy and generates a very International Conference on Digital Signal Processing (DSP), Fira,
low number of false positives which is a desirable quality for 2013, pp. 1-6.
use in real time diagnostic applications. The threshold is doi: 10.1109/ICDSP.2013.6622734.
[13] A. Issac, M. ParthaSarathi & M. K. Dutta, “An adaptive threshold based
adaptive making the process of diagnosis entirely automatic. image processing technique for improved glaucoma detection and
The images obtained from the fundus camera can be analyzed classification”, Computer Methods and Programs in Biomedicine ,
to detect signs of disease which can help in early treatment. Volume 122 , Issue 2 , pp. 229 – 244.
[14] A. Issac, M. Parthasarthi and M. K. Dutta, "An adaptive threshold based
algorithm for optic disc and cup segmentation in fundus images," 2015
ACKNOWLEDGEMENT 2nd International Conference on Signal Processing and Integrated
This work has been supported in part by the Grants from Networks (SPIN), Noida, 2015, pp. 143-147.
Department of Science and Technology, No. [15] D. Yadav, M. P. Sarathi and M. K. Dutta, "Classification of glaucoma
based on texture features using neural networks," 2014 Seventh
DST/TSG/ICT/2013/37”. Also, the authors express their International Conference on Contemporary Computing (IC3), Noida,
gratitude to Dr. S.C. Gupta, Medical Director of Venu Eye 2014, pp. 109-112.
research centre for his invaluable contribution in classification [16] Anushikha Singh, Malay Kishore Dutta, M. ParthaSarathi, Vaclav Uher
of the images. & Radim Burget, “Image Processing Based Automatic Diagnosis of
Glaucoma using Wavelet Features of Segmented Optic Disc from
Fundus Image” Computer Methods and Programs in Biomedicine,
REFERENCES Volume 124 , pp. 108 – 120.
[1] N. Sengar, M. K. Dutta, R. Burget and L. Povoda, "Detection of diabetic [17] M. K. Dutta, A. K. Mourya, A. Singh, M. Parthasarathi, R. Burget and
macular edema in retinal images using a region based method," 2015 K. Riha, "Glaucoma detection by segmenting the super pixels from
38th International Conference on Telecommunications and Signal fundus colour retinal images," 2014 International Conference on
Processing (TSP), Prague, 2015, pp. 412-415. Medical Imaging, m-Health and Emerging Communication Systems
[2] Z. Y. Zaidi, M. U. Akram and A. Tariq, "Retinal image analysis for (MedCom), Greater Noida, 2014, pp. 86-90.
diagnosis of macular edema using digital fundus images," 2013 IEEE [18] A. Singh, M. K. Dutta, M. Parthasarathi, R. Burget and K. Riha, "An
Jordan Conference on Applied Electrical Engineering and Computing efficient automatic method of Optic disc segmentation using region
Technologies (AEECT), Amman, 2013, pp. 1-5. growing technique in retinal images," 2014 International Conference on
[3] A. Johny and A. Thomas, "A novel approach for detection of diabetic Contemporary Computing and Informatics (IC3I), Mysore, 2014, pp.
macular edema," 2016 International Conference on Emerging Trends in 480-484.
Engineering, Technology and Science (ICETETS), Pudukkottai, 2016, [19] S. Ganguly et al., "An adaptive threshold based algorithm for detection
pp. 1-4. of red lesions of diabetic retinopathy in a fundus image," 2014
[4] A. Bansal, A. Vats, A. Jain, M. K. Dutta, R. Burget and J. Prinosil, "An International Conference on Medical Imaging, m-Health and Emerging
efficient automatic intensity based method for detection of macula in Communication Systems (MedCom), Greater Noida, 2014, pp. 91-94.
retinal images," 2016 39th International Conference on [20] P. N. Sharath Kumar, R. R. Kumar, A. Sathar and V. Sahasranamam,
Telecommunications and Signal Processing (TSP), Vienna, 2016, pp. "Automatic detection of exudates in retinal images using histogram
507-510. analysis," 2013 IEEE Recent Advances in Intelligent Computational
[5] D. Deka, J. P. Medhi and S. R. Nirmala, "Detection of macula and fovea Systems (RAICS), Trivandrum, 2013, pp. 277-281.
for disease analysis in color fundus images," 2015 IEEE 2nd [21] A. Narang, G. Narang and S. Singh, "Detection of hard exudates in
International Conference on Recent Trends in Information Systems colored retinal fundus images using the Support Vector Machine
(ReTIS), Kolkata, 2015, pp. 231-236. classifier," 2013 6th International Congress on Image and Signal
[6] S. Lu and J. H. Lim, "Automatic macula detection from retinal images Processing (CISP), Hangzhou, 2013, pp. 964-968.
by a line operator," 2010 IEEE International Conference on Image [22] Lili Xu and Shuqian Luo, "Support vector machine based method for
Processing, Hong Kong, 2010, pp. 4073-4076. identifying hard exudates in retinal images," 2009 IEEE Youth
doi: 10.1109/ICIP.2010.5649080 Conference on Information, Computing and Telecommunication,
[7] B. Ramasubramanian and S. Selvaperumal, "A stand-alone MATLAB Beijing, 2009, pp. 138-141.
application for the detection of Optic Disc and macula," 2016 [23] M. K. Dutta et al., "An efficient grading algorithm for non-proliferative
International Conference on Advanced Communication Control and diabetic retinopathy using region based detection," 2015 38th
Computing Technologies (ICACCCT), Ramanathapuram, 2016, pp. 241- International Conference on Telecommunications and Signal Processing
245. (TSP), Prague, 2015, pp. 743-747.
486