Fractals for complexity analysis of diabetic retinopathy in retinal vasculature images

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Issue: 03 | Mar-2014, Available @ http://www.ijret.org 680
FRACTALS FOR COMPLEXITY ANALYSIS OF DIABETIC
RETINOPATHY IN RETINAL VASCULATURE IMAGES
Nazneen Akhter1
, Yogesh Rajput2
, Sumegh Tharewal3
, K. V. Kale4
, Ramesh Manza5
1, 2, 3
Research Fellow, 4
Professor & Head, 5
Asst. Professor, Dept. of CS & IT, Dr. B. A. M. University, Maharashtra, India
Abstract
Arterial pattern and morphology of distribution is damaged because of diabetes resulting in retinal vasculature deformation. This
aspect is studied in terms of quantification of the degree of complexity associated with the distribution of blood vessels in eye for
healthy and diabetic humans. Retina images of fifteen healthy subjects are compared with those of diabetic subjects. It is found that
the increased complexity of structure and texture of the diabetic subjects results in a higher fractal dimension as compared to those of
healthy subjects. Also the blood vessel patterns, both for healthy and diabetic subjects show self-similarity and scale invariance and
hence the patterns are fractals. For the purpose of characterization of the irregular patterns of blood vessels in retina, box counting
technique is used for the estimation of fractal dimension. A GUI based program is developed in Matlab for implementation of box
counting technique and determination of fractal dimensions. Fractal dimension of retina images for diabetic subjects show higher
fractal dimensions indicating higher degree of structural complexity associated with the image whereas images of healthy subjects
show a lower value of fractal dimension indicating limited complexity of structure. It is shown that fractal dimension can be used to
distinguish diabetic subjects from healthy subjects and hence this technique could be used in diagnosis of diabetes using images of
retina. It is interesting that during other diagnostic procedures related to retina images, this information can be generated as
additional information adding value to the diagnostic procedures. Details of implementation of the technique are presented.
Keywords: Diabetic Retinopathy, Fractal Dimension, Box Counting, Segmentation, Image Processing
--------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
Diabetic retinopathy is damage to the retina caused by
complications of diabetes, which can eventually lead to
blindness. which affects up to 80 percent of all patients who
have had diabetes for 10 years or more.[1]
Despite these
intimidating statistics, research indicates that at least 90% of
these new cases could be reduced if there was proper and
vigilant treatment and monitoring of the eyes.[2]
The longer a
person has diabetes, the higher his or her chances of
developing diabetic retinopathy.[3]
Diabetic retinopathy is one
of the foremost frequent causes of blindness world-wide. In
India, it was the 17th cause of blindness 20 years ago but has
now ascended to the 6th position. Diabetic retinopathy has
been identified as one of the significant causes of blindness or
vision impairment in India. Though cataract is still the leading
cause of blindness, the intense work under the National
Programme for Control of Blindness (NPCB) with the support
of international non-governmental organisations has brought
down its contribution to blindness. There are 41 million of
diabetics in India at present and every diabetic is a potential
candidate for loss of vision due to diabetic retinopathy. This
number is poised to increase significantly. Thus it is an
appropriate time now to concentrate on diabetic retinopathy
and bring the problem under control[4].
However, we have to
recognize that the issues in managing vision impairment due
to diabetic retinopathy are different from cataract. While
cataract blindness is curable by a simple one time surgical
intervention, diabetic retinopathy encompasses a multitude of
problems and can be prevented if detected early and treated. In
the first stage which is called non-proliferative diabetic
retinopathy (NPDR) there are no symptoms. The only way to
detect NPDR is by fundus photography (Retinal fundus
images, using a digital fundus camera).
The retina is the only location where blood vessels can be
directly visualized non-invasively in vivo [5].
The blood vessels
in the retina branch a number of times each time forming a
vessel tree that is similar in characteristics to the branch it
came from. Systems that have self-similarity at multiple scales
are known as fractals. There are some reports of measuring
fractal properties of blood vessel patterns of the retina; most
have involved manual segmentation of the blood vessel
patterns [6-8].
Now with reliable automated vessel
segmentation, attention is turning to analyzing the retinal
vasculature as a fractal.[9]
Fractal objects are self-similar structures that retain a similar
level of complexity across all scales. For example, blood
vessels repeatedly subdivide downstream into smaller blood
vessels with similar network patterns. Fractal dimension (Df)
quantifies the degree of complexity into a single value and is
particularly useful for quantifying non-Euclidean geometric
shapes such as vascular networks. The retinal circulation is a
fractal object [10-12],
and fractal analysis has been used to study

__________________________________________________________________________________________
the embryological development of the retinal vasculature [11]
and vascular changes associated with diabetic retinopathy [12–
16].
Fractal analysis of vasculature formed in human retinal
image can be used as a non-invasive technique for detection of
early retinal vascular diseases. It is a more useful descriptor
that offers a new language for examining the complex patterns
found in ophthalmic practice.
2. METHODS & MATERIAL
For our experiments we made use of the retinal images from
DRIVE database [17].
From set of 40-40 images of healthy and
diabetic subject images, we randomly selected 15 images of
healthy subjects and 15 images of diabetic subjects. For our
proposed system we designed a GUI in MATALB 2012 as
shown in Fig. 1.
Fig-1: GUI showing the steps involved in fractal dimension
calculation of a retina image
Fig 1 shows the sequentially performed steps to obtain the
fractal dimension of retinal vasculature and analyze the
complexity of blood vessels pattern. The high resolution
colour fundus images obtained from DRIVE database were
preprocessed by applying two filters. Firstly the green channel
was extracted from the RGB image because green channel
shows high intensity as compare to red and blue channels.
Then in the next step 2D median filter was applied to remove
noise from the images. And finally the preprocessed image
was converted to grey scale. Further histogram equalization
operation was performed to increase the contrast in the image.
Once the image was enhanced threshold operation was
performed to extract the blood vessels. These extracted blood
vessels were used to calculate the fractal dimension.
For calculating fractal dimension, image segmentation was
done using box counting technique. Box counting is one of the
standard techniques employed in fractal studies. Given shape
or pattern under study, is covered by boxes of different size (r)
and the number of boxes required to cover the complete shape
is recorded as N. If the pattern has self-similarity and scale
invariance, the plot of log(N) versus log(r) is a straight line
indicating that the power law is obeyed this confirms fractal
character. Fractal dimension is estimated from slope of this
best fitting straight line and fractal dimension is equal to slope.
3. RESULTS & DISCUSSIONS
Images of Retina of human subjects were studied to explore
the degree of complexity associated with the patterns of blood
vessel network in human eye to compare the texture for
diabetic patients with those of healthy individuals selecting
fifteen subjects for each. The shapes of patterns of blood
vessel distribution are highly complex and very much differ
from individual and thus it is very difficult to compare the
structure with some standard in order to arrive at discernable
difference. Fractal geometry is known to be effective in
characterization of irregular shapes of this type thus the
concept of fractals and fractal dimensions is employed for
characterization of these patterns. In general higher the fractal
dimension of an irregular shape, higher is the degree of
complexity associate with the shape.
For the purpose of characterization of the irregular shapes
from images of the Retina, high resolution images were used
with identical resolution for both the healthy and diabetic
subjects. Fractal dimension was obtained and used to
characterize the patterns in terms of complexity of structure
and the fractal dimension was estimated using box counting
technique that best suits such applications.
Box counting technique makes use of square boxes of
different sizes (r) and the number of boxes (N) of each size
required to completely cover the image are determined. A
graph is plotted taking log(r) at the axis of x and log(N) on the
y-axis. If the image being analyzed is a fractal then there is
scale invariance and self-similarity and the resulting plot is a
straight line with a slope equal to the fractal dimension (with a
change of sign) of that pattern or shape.
As the images used are colored it is convenient to process the
image to improve the quality for the extraction of the pattern
and convert it to black and white two color images in the form
of bit map for further use. The image processing and counting
of number of boxes of different size required to completely
cover the image was implemented in Matlab and necessary
software with GUI was developed for this purpose. Two
typical images each for healthy subjects are shown in Fig. 2
along with the final images converted to two color bitmaps.

__________________________________________________________________________________________
a b
c d
Fig -2: Typical images of Retina for healthy subjects (a, b)
and their two color bitmaps (c, d)
Two typical images of retina for diabetic subjects and the
corresponding two color bitmaps are shown in Fig. 3. It can be
seen from the comparison of Fig. 1 and 2 that there is more of
structure and texture associated with the images for diabetic
subjects adding to the complexity of shape. This complexity of
shape and additional texture gives rise to a relatively higher
fractal dimension and this can be used in distinguishing
between the two.
a b
c d
Fig -3: Typical images of Retina for diabetic subjects (a, b)
and their two color bitmaps (c, d)
Analysis of these images yield interesting results showing that
the complexity of shape associated with the images for
diabetic subjects is more than that of healthy subjects. The
results of analysis of Box counting for two typical cases for
healthy subjects are presented in Table – 1 and 2.
Table – 1: Results of Box counting of image for healthy
subject (Case-A)
S.No log(r ) log(N)
1 0.0000 4.9193
2 0.3010 4.3925
3 0.6021 3.9113
4 0.9031 3.4735
5 1.2041 3.0781
6 1.5051 2.6875
7 1.8062 2.2765
8 2.1072 1.8129
9 2.4082 1.3010
10 2.7093 0.7782
11 3.0103 0.3010
12 3.3113 0.0000
Table – 2: Results of Box counting of image for healthy
subject (Case B)
S.No log(r ) log(N)
1 0.0000 4.9334
2 0.3010 4.4080
3 0.6021 3.9296
4 0.9031 3.4955
5 1.2041 3.1004
6 1.5051 2.7126
7 1.8062 2.2945
8 2.1072 1.8451
9 2.4082 1.3424
10 2.7093 0.7782
11 3.0103 0.3010
12 3.3113 0.0000
The results from box counting shown in Table I and II are
plotted as log(N) versus log(r) plot in Fig. 4 and 5
respectively. The points plotted as small circles are actual data
and the bold ling joining the points is the best fitting line (least
square fit) for the data points and the equation shown in the
inset is the equation of this straight line. It is seen that all the
points lie along a straight line which is also evident from the
value of R2
that lies close to unity. This confirms the existence
of self-similarity and scale invariance showing that the
patterns do exhibit fractal character.

__________________________________________________________________________________________
y = -1.4860x + 4.8713
R2
= 0.9985
0
1
2
3
4
5
6
0 0.5 1 1.5 2 2.5 3 3.5
log(r)
log(N)
Data
Fit to Data
Fig-4: Log (N) versus log (r) plot for Table I showing a fractal
dimension of 1.486
y = -1.4895x + 4.8945
R2
= 0.9982
0
1
2
3
4
5
6
0 0.5 1 1.5 2 2.5 3 3.5
log(r)
log(N)
Data
Fit to Data
Fig-5: Log (N) versus log (r) plot for Table II showing a
fractal dimension of 1.4895
The results of analysis of Box counting for two typical cases
for diabetic subjects are presented in Table 3 and 4.
Table – 3: Results of Box counting of image for diabetic
subject (Case A)
S.No log(r ) Log(N)
1 0.0000 5.0485
2 0.3010 4.5373
3 0.6021 4.0739
4 0.9031 3.6545
5 1.2041 3.2574
6 1.5051 2.8549
7 1.8062 2.3945
8 2.1072 1.8692
9 2.4082 1.3617
10 2.7093 0.7782
11 3.0103 0.3010
12 3.3113 0.0000
Table – 4: Results of Box counting of image for diabetic
subject (Case B)
S.No log(r ) log(N)
1 0.0000 5.0840
2 0.3010 4.5599
3 0.6021 4.0818
4 0.9031 3.6461
5 1.2041 3.2355
6 1.5051 2.8254
7 1.8062 2.3483
8 2.1072 1.8261
9 2.4082 1.2788
10 2.7093 0.7782
11 3.0103 0.3010
12 3.3113 0.0000
The data from box counting from Table III and IV are plotted
as log(N) versus log(r) plot in Fig. 6 and 7 respectively. The
points plotted are actual data from box counting and the ling
joining the points is the best fitting line for the data and the
equation in the inset is the equation of this best fitting straight
line. It is seen that all the points lie along a straight line which
is also evident from the value of R2
that lies close to unity.
This confirms the existence of self-similarity and scale
invariance showing that the patterns do exhibit fractal
character.
It is clearly seen from Fig. 4 to 7 that the patterns from retina
images do possess self-similarity and scale invariance that is a
characteristic of fractal shapes. The complexity of shape
associated with these patterns is given by the fractal
dimensions. It is also seen that the fractal dimensions of the
patterns of retina images for diabetic subjects are on higher
side as compared to those of healthy subjects. To demonstrate
the fact that the complexity of shape associated with diabetic
subjects retina images is more than those of healthy subjects
the fractal dimensions obtained for fifteen healthy subjects are
compared with those of diabetic subjects in Fig. 8.
y = -1.5427x + 5.0651
R2
= 0.9978
0
1
2
3
4
5
6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
log(r)
log(N)
Data
Fit to Data
Fig-6: Log(N) versus log (r) plot for Table III showing a

__________________________________________________________________________________________
y = -1.5555x + 5.0725
R
2
= 0.9985
0
1
2
3
4
5
6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
log(r)
log(N)
Data
Fit to Data
Fig-7: Log(N) versus log (r) plot for Table IV showing a
1.42
1.44
1.46
1.48
1.50
1.52
1.54
1.56
1.58
1.60
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Subject No
FractalDimension
Healthy
Diabetic
Fig-8: Comparison of fractal dimension of retina images of
fifteen healthy and diabetic subjects.
Comparison of fractal dimension clearly reveals that the
fractal dimension of retina images of diabetic subjects are
higher compared to those of healthy subjects.
4. CONCLUSIONS
Diabetes causes gradual damage to retina resulting in serious
deformation in arteries and veins in retina, this aspect is
studied in terms of quantification of the degree of complexity
associated with the distribution of blood vessels in eye. Retina
images of fifteen healthy subjects are compared with those of
fifteen diabetic subjects. It is fund that the increased
complexity of structure and texture of the diabetic subjects
results in a higher fractal dimension as compared to those of
healthy subjects. Also the blood vessel patterns, both for
healthy and diabetic subjects show self-similarity and scale
invariance and hence the patterns are fractals. It is seen from
Fig. 7 that fractal dimension can be used to distinguish
diabetic subjects from healthy subjects and hence this
technique could be used in diagnosis of diabetes using images
of eye/retina. It is interesting that during other diagnostic
procedures related to retina images, this information can be
generated as additional information adding value to the
diagnostic procedures.
ACKNOWLEDGEMENTS
The authors are thankful to UGC for providing financial
support for the major research project entitled “Development
of Color Image Segmentation and Filtering Techniques for
Early Detection of Diabetic Retinopathy” F. No.: 41 –
651/2012 (SR). And are also thankful to DST for providing
financial support for the major research project entitled
“Development of multi resolution analysis techniques for
early detection of non-proliferative diabetic retinopathy
without using angiography” F.No. SERB/F/2294/2013-14.
The authors also extend their gratefulness to Dr. Manoj
Saswade, Director “Saswade Eye Clinic” Aurangabad and Dr.
Neha Deshpande, Director “Guruprasad Netra Rungnalaya
pvt. Ltd”, Samarth Nagar, Aurangabad for providing the
Database and accessing the Result.
REFERENCES
[1]. "Diabetic retinopathy". Mayo Clinic.
http://www.mayoclinic.org/diseases-
conditions/diabeticretinopathy/basics/definition/con-20023311
[2]. Kertes PJ, Johnson TM, ed. (2007). Evidence Based Eye
Care. Philadelphia, PA: Lippincott Williams & Wilkins.
ISBN 0-7817-6964-7
[3]. Tapp RJ, Shaw JE, Harper CA et al. (June 2003). "The
prevalence of and factors associated with diabetic retinopathy
in the Australian population". Diabetes Care 26 (6): 1731–7.
doi:10.2337/diacare.26.6.1731. PMID 12766102.
[4]. Guidelines for the Comprehensive Management of
Diabetic Retinopathy in India, A VISION 2020: The Right to
Sight INDIA Publication, National Programme for Control of
Blindness (NPCB), Directorate General of Health Services,
Ministry of Health & Family Welfare, Government of India.
[5]. Niall Patton at. el., Retinal image analysis: Concepts,
applications and potential, Progress in Retinal and Eye
Research 25 (2006) 99–127, Elsevier,
doi:10.1016/j.preteyeres.2005.07.001
[6]. Cheng, S.C., Huang, Y.M.: A novel approach to diagnose
diabetes based on the fractal characteristics of retinal images.
IEEE Trans. Inf. Technol. Biomed. 7(3), 163–170 (2003)
[7]. Cheung, N., Donaghue, K.C., Liew, G., Rogers, S.L.,
Wang, J.J., Lim, S.W., Jenkins, A.J., Hsu, W., Lee, M.L.,
Wong, T.Y.: Quantitative assessment of early diabetic
retinopathy using
[8]. fractal analysis. Diabetes Care 32, 106–110 (2009),
Masters, B.R.: Fractal analysis of the vascular tree in the
human retina. Annu. Rev. Biomed. Eng. 6, 427–452 (2004)
[9]. G. Dougherty (ed.), Medical Image Processing:
Techniques and Applications, Biological and Medical Physics,
Biomedical Engineering, DOI 10.1007/978-1-4419-9779-1 11,
© Springer Science+Business Media, LLC 2011

__________________________________________________________________________________________
[10]. Macgillivray TJ, Patton N, Doubal FN, Graham C,
Wardlaw JM. Fractal analysis 2082 DIABETES CARE,
VOLUME 32, NUMBER 11, NOVEMBER 2009
care.diabetesjournals.org of the retinal vascular network in
fundus images. Conf Proc IEEE Eng Med Biol Soc
2007;2007:6456–6459
[11]. Mainster MA. The fractal properties of retinal vessels:
embryological and clinical implications. Eye 1990;4:235–241
[12]. Masters BR. Fractal analysis of the vascular tree in the
human retina. Annu Rev Biomed Eng 2004;6:427–452
[13]. Avakian A, Kalina RE, Sage EH, Rambhia AH, Elliott
KE, Chuang EL, Clark JI, Hwang JN, Parsons-Wingerter P.
Fractal analysis of region-based vascular change in the normal
and non-proliferative diabetic retina. Curr Eye Res
2002;24:274–280
[14]. Daxer A. Characterisation of the neovascularization
process in diabetic retinopathy by means of fractal geometry:
diagnostic implications. Graefes Arch Clin Exp Ophthalmol
1993;231:681–686
[15]. Daxer A. The fractal geometry of proliferative diabetic
retinopathy: implications for the diagnosis and the process of
retinal vasculogenesis. Curr Eye Res 1993;12: 1103–1109
[16]. Daxer A. Mechanisms in retinal vasculogenesis: an
analysis of the spatial branching site correlation. Curr Eye Res
1995;14:251–254
[17]. J.J. Staal, M.D. Abramoff, M. Niemeijer, M.A.
Viergever, B. van Ginneken, "Ridge based vessel
segmentation in color images of the retina", IEEE
Transactions on Medical Imaging, 2004, vol. 23, pp. 501-509.
BIOGRAPHIES
Nazneen Akhter, M.Phil. in Computer
Science, is currently pursuing her doctoral
programme in CS & IT Dept. of Dr. B. A. M.
University. She is at present having BSR
Research Fellowship, previously she has
served for 12 years as Asst. professor and
even held the responsibility of Head of Dept., In her whole
tenure she has served various reputed higher educational
institutions and has also lend her services to one of the
esteemed university in Saudi Arabia. She has 03 books with
ISBN number to her credit. Her research interest includes
biomedical signal processing, Embedded System design,
Biomedical Engineering, Pattern Recognition and
Biometrics. She has presented more than 16 papers in national
and international conferences and she has more than 10 papers
published in reputed journals.
Yogesh M. Rajput has completed M.Sc.
Computer Science in 2012 and presently
working as a Project Fellow in Department of
Computer Science and Information
Technology, Dr. Babasaheb Ambedkar
Marathwada University, Aurangabad. His
area of specialization is Bio-Medical Image Processing and
Diabetic Retinopathy, He has published around 10 papers in
reputed Journals and Conferences.
Sumegh shrikant Tharewal M.Sc.
(computer science) BSR Research fellow at
Dr. Babsaheb Ambedkar Marathwada
University Aurangabad, Ex Assistant
professor at Vivekanand College
Aurangabad.He has published two research
papers in state level and national level conference. His
research area of interests is Digital image processing,
Biometrics, cloud computing, Embedded system.
Dr. K. V. Kale, M.Sc., MCA Ph.D. FIETE,
presently working as a Professor and head,
CS & IT Dept., Dr. B. A. M. University,
Aurangabad, MS-India. He is a fellow of
IETE, Life member of CSI, IAPR, ISCA,
IEEE and elevated to senior member of
IEEE. He is a member & Faculty of Board of Studies of
various universities in India & has designed and developed
new courses in computer science at UG & PG level. He is
recipient of VIJAY SHREE Award. He is actively engage in
research and development and has around 235 research papers
in reputed national/international journals and conferences and
05 books to his credit. He has produced 21 Ph. D. students and
04 M.Phil students under his able guidance. He is reviewer
and editor of several journals of India and aboard. His research
interest includes Image Processing, Pattern Recognition,
Biometrics, Bioinformatics, S/W Egg., Artificial Intelligence,
Computer vision, Neural Networks. He has completed three
major Research projects & currently working on multimodal
Biometrics & Remote Sensing.
Dr. Ramesh R. Manza is NET, SET qualified
and is Ph.D in Computer Science and is
currently working in the capacity of Asst. Prof.
in CS & IT Dept. of Dr. B. A. M. University,
Aurangabad, India. His area of specialization
includes Bio-Medical Image Processing,
Computer Vision, Nano-robotics, MEMS and Biometrics. He
has presented around 105 papers at national and international
conferences and around 42 papers in reputed Journals. He is
leading Two Major research projects sanctioned by UGC and
DST on Diabetic retinopathy. Total 16 students have been
awarded M. Phil. under his guidance and 8 students are
leading their Doctoral program. He is author of 05 Books and
is on editorial board of 03 International Journals. He has been
coordinator, treasurer and organizing secretary of 05
International Conferences and 19 workshops and trainings.

Fractals for complexity analysis of diabetic retinopathy in retinal vasculature images

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Fractals for complexity analysis of diabetic retinopathy in retinal vasculature images (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Fractals for complexity analysis of diabetic retinopathy in retinal vasculature images