232 Int. J. Biomedical Engineering and Technology, Vol. 6, No.

3, 2011

Analysis of Foveal Avascular Zone for grading

of Diabetic Retinopathy

M.H. Ahmad Fadzil and Lila Iznita Izhar

Department of Electrical and Electronic Engineering,
Universiti Teknologi PETRONAS,
Bandar Seri Iskandar, Tronoh 31750, Perak, Malaysia
E-mail: fadzmo@petronas.com.my
E-mail: lilaiznita@petronas.com.my
E-mail: maf4546@yahoo.com

Hanung Adi Nugroho*

Department of Electrical Engineering,
Universitas Gadjah Mada,
Jl. Grafika 2, Kampus UGM, Jogjakarta 55281, Indonesia
E-mail: hanungadin@gmail.com
E-mail: hanung@te.ugm.ac.id
*Corresponding author

Abstract: At present, it is difficult to determine Foveal Avascular Zone (FAZ)

enlargement based on colour fundus images. Fundus image analysis presents
several challenges such as high image variability, improper illumination and
artifacts. A new approach for grading Diabetic Retinopathy (DR) by analysing
FAZ enlargement in colour fundus image has been developed. Investigations
show that FAZ area ranges can be used to indicate progression of the disease.
The mean accuracy and standard deviation of ranges obtained are 92.2% and
3.22, respectively. This new approach is reliable, accurate and fast compared to
the current method based on DR pathologies.

Keywords: biomedical image processing; colour fundus image; diabetic

retinopathy grading; FAZ; foveal avascular zone.

Reference to this paper should be made as follows: Ahmad Fadzil, M.H.,

Iznita, L.I. and Nugroho, H.A. (2011) ‘Analysis of Foveal Avascular Zone for
grading of Diabetic Retinopathy’, Int. J. Biomedical Engineering and
Technology, Vol. 6, No. 3, pp.232–250.

Biographical notes: M.H. Ahmad Fadzil is currently a Professor in the

Department of Electrical and Electronic Engineering, Universiti Teknologi
PETRONAS, Malaysia. He received BSc (1st Class) in Electronics
Engineering, MSc in Telematics and PhD in Image Processing from University
of Essex, UK in 1983, 1984 and 1991, respectively. His research interests
include medical imaging and telemedicine, vision inspection systems, image
compression and coding, and signal processing of seismic data for hydrocarbon
exploration.

Lila Iznita Izhar is currently a Senior Lecture in the Department of Electrical

and Electronic Engineering, Universiti Teknologi PETRONAS, Malaysia.
She received BEng in Electrical and Electronics Engineering from University

Copyright © 2011 Inderscience Enterprises Ltd.

 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 233

of the Ryukyus, Japan in 2002 and MSc in Electrical and Electronics

Engineering from Universiti Teknologi PETRONAS, Malaysia in 2006.
Her areas of research are in medical imaging and computer vision.

Hanung Adi Nugroho is currently a PhD student in the Department of Electrical

and Electronic Engineering, Universiti Teknologi PETRONAS, Malaysia. He is
also a Senior Lecturer in the Electrical Engineering Department at Universitas
Gadjah Mada, Indonesia. He received BEng (Hons) in Electrical Engineering
from Universitas Gadjah Mada, Indonesia in 2001 and Master of Engineering
in Biomedical Engineering from The University of Queensland, Australia in
2005. His research interests include medical imaging and signal processing,
computer vision, and medical instrumentation.

1 Introduction

The finest blood vessels linking arteries to veins that are called retinal capillaries
(thin/micro vessels), tend to be damaged by changes of chemical due to diabetes.
This progressive damage is called Diabetic Retinopathy (DR) and occurs due to a
combination of micro-vascular leakage and micro-vascular occlusion (Bresnick et al.,
1984).
Diabetic Retinopathy can be classified into two main forms based on presence
of pathologies; Non-Proliferative Diabetic Retinopathy (NPDR) and Proliferative
Diabetic Retinopathy (PDR). NPDR can further be divided into three level of severity
namely mild, moderate and severe NPDR. In Malaysia, the diabetic population has
increased over four-fold from 300,000 in 1996 (Anonymous, 1996) to 1,383,675
in 2005 (Anonymous, 2004). According to National Eye Database 2007, among 10,856
Malaysian populations with diabetes, 36.8% has any form of DR, of which 7.1% has
PDR (Goh, 2008). Diabetic Eye Registry also reported that among 7797 diabetic patients
in Malaysia who were seen for the first time at Ophthalmology clinics from January
to September 2007, more than half (54.4%) had DR (Goh et al., 2008). Among these DR
patients, 70% of the eyes have mild or moderate NPDR, while 30% has severe NPDR,
PDR or advanced diabetic eye disease (Goh et al., 2008). In 2008, Singapore Malay
Eye Study on 3261 persons with diabetes found 35.3% prevalence of any retinopathy,
of which 6.8% had PDR (Wong et al., 2008).
NPDR is the earliest form of DR caused by micro-vascular leakage away from the
macula (Osareh, 2004). The occurrence of NPDR are indicated by the presence of
sacculations from the capillary walls (microaneurysms), blood (retinal haemorrhages),
lipid exudates (hard exudates) and retinal edema. The rupture of these microaneurysms
results in haemorrhages. At this stage, there are often no obvious warning signs and
patients suffering from the disease are unaware until it advanced into more severe levels.
Treatment of the disease at this stage may prevent future complications and towards
blindness.
Lesion features namely cotton wool spots, venous beading and Intraretinal
Microvascular Abnormalities (IRMA) can be found in moderate to severe NPDR. These
changes are found in areas of retinal capillary non-perfusion and indicate the severity of
the non-proliferative changes. The Early Treatment Diabetic Retinopathy Study (ETDRS)
identified multiple retinal haemorrhages, venous caliber (width) changes and IRMA as
234 M.H. Ahmad Fadzil et al.

main indicator of risk of advancement to PDR (Report No. 10, 1991, Report No. 12,
1991). Sometimes, there are extensive areas of microvascular occlusion (blockage of
capillary network) throughout the retina. When vessel occlusions occur, the retinal tissues
that lack of nutrition and oxygen release a vasoproliferative factor that stimulates the
growth of new abnormal blood vessels where the normal capillaries are occluded.
This form of DR is called PDR. It may cause bleeding into the cavity of the eye and
produce scars (scarring) with loss of vision (Report No. 12, 1991).
The severity of DR and quantification of diabetic changes are very important to be
classified to assess the treatments and the risk factors for frequent complication
of diabetes. Currently, the International Clinical DR Disease Severity Scale shown in
Table 1 is used in grading of DR (Anonymous, 2003). An example of colour fundus
image taken from a patient with mild NPDR is shown in Figure 1. Using the International
Clinical DR Disease Severity Scale, an ophthalmologist or grader needs to observe
and find DR-related abnormalities present in colour fundus image. This method is found
to be very time consuming and tedious that requires significant training and exercise
and is susceptible to human error.

Figure 1 Colour fundus image with mild NPDR

Source: Ahmad Fadzil et al. (2009)

Table 1 International clinical Diabetic Retinopathy disease severity scale

Proposed disease severity level Findings observable upon dilated ophthalmoscopy

No apparent retinopathy
Mild Non-Proliferative Diabetic
Retinopathy
Moderate Non-proliferative
Diabetic Retinopathy
Severe Non-Proliferative Diabetic
Retinopathy
Proliferative Diabetic Retinopathy
No abnormalities
Microaneurysms only
More than just microaneurysms but less than
Severe NPDR
Any of the following:
1 >20 intraretinal hemorrhages in each of 4 quadrants
2 Definite venous beading in 2+ quadrants
3 Prominent intraretinal microvascular abnormalities
in 1+ quadrant
4 No signs of proliferative retinopathy
One or more of the following:
1 Neovascularisation
2 Vitreous/ pre-retinal haemorrhage
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 235

Several methods have been studied to automatically detect the presence of DR

pathologies such as microaneurysms (Cree et al., 1996; Sinthanayothin et al., 2002),
haemorrhages (Osareh et al., 2003; Sinthanayothin et al., 2002), exudates (Walter et al.,
2002; Sopharak et al., 2008; Hunter et al., 2000) and vessel abnormalities (Nguyen et al.,
1997). Other develop methods have been focused on detection of retinal anatomy such as
optic disc (Balasubramanian et al., 2007; Li and Chutatape, 2000, 2001), fovea (Fleming
et al., 2007; Ibanez and Simo, 1999) and retinal blood vessels (Conforth et al., 2004;
Sinthanayothin et al., 1999; Hoover et al., 2000) as an important requirement for
automatic detection of DR. An extensive review of image analysis and intelligent systems
for automated DR ocular screening has been well summarised (Teng et al., 2002; Patton
et al., 2006). Teng et al. (2002) has been identified several image preprocessing
techniques to mimic some of the processes that an ophthalmologist performs when
examining the retina; for example, segmentation of the retinal vasculature and
identification common pathological features, such as lesion, small aneurysms and
exudates. Cree et al. (1996) developed a digital image processing system to quantify and
monitor the presence of microaneurysms and achieved 82% sensitivity for detecting
microaneurysms with a specificity of 84%. However, this system needed invasive retinal
fluorescein angiograms as the input to get this achievement.
Neural Network (NN) has been exploited to classify the retinal abnormalities in a few
studies. A study by Osareh et al. used retinal images which are broken down into small
squares and then were presented to a back propagation NN to identify the blood vessels,
exudates and haemorrhages (Osareh et al., 2003). The NN was trained for five days and
the sensitivity of the exudate detection method achieved 93.1%. Another NN based
exudate detection research was conducted in Hunter et al. (2000). For lesions detection,
the final NN architecture had 11 input variables (feature vector) and achieved sensitivity
and specificity of 91% using a relatively small number of images. The reported
performance was based on whether each 16 × 16 pixel patches contains exudates or not.
No pixel-based validation was reported for this work. Nguyen et al. (1997) developed a
multilayer feed-forward network for the classification of nine lesions associated with DR:
haemorrhages; microaneurysms, hard exudates, soft exudates (cotton wool spot), venous
abnormalities, arterial abnormalities, arterious venous nicking, retinal elevation, vitreous
haemorrhages, and macular edema. However, the accuracy of the feature extractions and
grading of DR lesions are not of great importance in this work. Sinthanayothin et al. has
developed an automatic computerised screening system based on image processing
techniques and neural network to determine the main components of the retina
(Sinthanayothin et al., 1999), important features of background DR and classify
the normal, abnormal and unknown retinal image (Sinthanayothin et al., 2003).
The computerised screening program achieved 80.21% sensitivity and 70.66% specificity
in classifying the normal and abnormalities of retinal images. An NN classifier can also
be used to improve classifying the abnormalities (Poonguzhali and Ravindran, 2007).
Nevertheless, the above mentioned efforts focus on the detection of retinal anatomy and
retinal abnormalities to determine whether a patient is suffering DR rather than to
monitor and grade DR severity level.
In this research work, the correlation of Foveal Avascular Zone (FAZ) enlargement
with DR progression is investigated. It is known that the FAZ enlarges in DR resulting
from a loss of capillaries in the perifoveal capillary network (Bresnick et al., 1984;
Conrath et al., 2004). This is often observed in early DR such as NPDR and also in PDR
(Khurana, 2003). Previous studies have conducted to measure FAZ in fundus fluorescein
236 M.H. Ahmad Fadzil et al.

angiography (Jun Lee and Jun Koh, 2008; Conrath et al., 2006; Haddouche et al., 2005).
Currently, it is difficult to discern FAZ and to measure FAZ enlargement based on colour
fundus images. This research work aims to develop a new DR grading system based on
FAZ enlargement and analysis. Early detection of FAZ enlargement at NPDR stage may
prevent the progress of the disease to PDR stage and towards visual loss.

2 Approach

Foveal Avascular Zone (FAZ) is the fovea devoid of capillaries in the macula (Richard
et al., 1998). FAZ can be represented as dark circle zone without vessels at the centre of
macula as shown in Figure 2 (Richard et al., 1998). FAZ varies in size for healthy
subjects but usually has a diameter around 500 µm (Saini et al., 2006; Zeffren et al.,
1990; John et al., 2006) and size of about 0.4 mm2 (Bresnick et al., 1984; Mansour, 1990;
Sander et al., 1994; Conrath et al., 2004; Parodi et al., 1995; Bradley et al., 1992). FAZ is
the most accurate vision zone on the retina. The detection of the FAZ using eye fundus
angiographies is presented in Ibanez and Simo (1999). An enlargement of the zone is
usually found in eyes with DR resulting from a loss of capillaries in the perifoveal
capillary network. This condition appears early in the development of the disease. It is
one of the earliest retinal abnormalities caused by diabetes besides microaneurysms and
micro-vascular leakage.

Figure 2 The perifoveal capillary network in enlarged FAZ

Source: Richard et al. (1998)

The small capillaries surrounding the FAZ may possibly be blocked or damaged as a
result of higher hemodynamic stress such as increase in heart rate and blood flow.
The defect of the perifoveal area can lead to a rapid loss in visual acuity (Richard et al.,
1998). Without any measurable loss of visual acuity, diabetic patients may exhibit an
abnormally enlarged FAZ compared to healthy subjects.
In order to analyse FAZ, the binary retinal vasculature map of a fundus image is
firstly obtained. Previous work has been conducted to detect and reconstruct certain
retinal pathologies in addition to the retinal vasculature map (Ahmad Fadzil et al., 2007).
To increase the accuracy on determination of FAZ, the detection and reconstruction of
retinal vasculature is mainly focused on the thin vessels/ capillaries and pathologies
surrounding FAZ.
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 237

The pathologies are microaneurysms, dot and blot haemorrhages and exudates at the
perifoveal network. These pathologies are also reconstructed as they may originate
from the vessels, mainly at the vessel ends, and thus contributing in reconstruction
of vessel ends as well. Even though they are mostly originated from thin vessels,
these thin vessel ends are usually damaged by the formation of the pathologies and
are very low in contrast. Therefore, besides vessel ends surrounding FAZ, the pathologies
are reconstructed as well to increase the accuracy on FAZ determination for potential
grading of DR. A flowchart of FAZ analysis is shown in Figure 3.

Figure 3 A flowchart of FAZ analysis

2.1 Detection and reconstruction of vessels and pathologies

In this work, the detection and reconstruction algorithm reported by Ahmad Fadzil et al.
(2007) is applied to fundus images to detect and reconstruct certain retinal pathologies in
addition to the retinal vasculature map. In this algorithm, for vessel detection, the fundus
image is firstly enhanced using a mean filter followed by Contrast Limited Adaptive
Histogram Equalisation (CLAHE) and bottom-hat morphological transformation to
extract retinal vasculature (blood vessels). A sum of bottom-hat with 12 linear structuring
elements of size 15 and at 0°–180° (incremental at 15°) is performed to ensure all
blood vessels are extracted. Background noise removal is then carried out to reduce
unwanted linear features at the background being enhanced during bottom-hat.
Further enhancement of vessels is then carried out using contrast stretching. For vessel
reconstruction, a gradient-based region growing based on first-order Gaussian derivative
(GRG) is performed on the extracted retinal vasculature. GRG incorporates both gradient
magnitude change and average intensity as the homogeneity criteria that enable the
process to adapt to intensity changes and intensity spread over the vasculature region.
For FAZ determination, the same algorithm is used to detect and reconstruct certain
retinal pathologies in addition to the retinal vasculature map as they may originate
from the vessels (mainly at the vessel ends) and thus contributing in reconstruction of
vessel ends as well. Here, during GRG, seed pixels are also assigned on the pathologies
(Figure 4(a)) so that certain pathologies could be grown in the retinal vasculature map
(Figure 4(b)). It should be pointed out that the detection and reconstruction of vessels
and pathologies is focused on the FAZ surroundings (perifoveal capillary network).
238 M.H. Ahmad Fadzil et al.

Therefore, the seed pixels are placed mainly on the vessel ends and on pathologies at the
perifoveal capillary network.

Figure 4 (a) Extracted retinal vasculature using sum of bottom-hat and (b) detected and
reconstructed retinal vasculature and certain pathologies at the perifoveal capillary
network

(a) (b)

2.2 Detection and selection of vessel and pathology end points

The binary reconstructed retinal vasculature map including pathologies is used for the
detection and analysis of FAZ. Pixels representing the vessel and pathology ends will be
derived from the image. This is done as FAZ is the fovea zone devoid of capillaries.
The boundary of FAZ is formed based on the vessel and pathology ends on the perifoveal
capillary network. The resultant images for each process involved are shown in Figure 5
with vessels inside the rectangle representing the detected perifoveal capillary network
and pathologies. As shown in Figure 5, morphological operations are performed on the
binary image with reconstructed vessel lines and pathology regions. First, the image
undergoes thinning process to obtain an image, Ithin, where the vessels and pathologies
are shrunk to one pixel width lines as shown in Figure 5(b). Then image Ithin, undergoes
an end point removal process where the pixel at the end for each thinned vessel and
pathology lines are removed without removing small objects completely to produce
image, Ino_endpt. as shown in Figure 5(c). The image Ino_endpt is subtracted from the image,
Ithin to produce an image with pixels at only the vessel and pathology ends, Iendpt. as
follows.
Iendpt (x, y) = Ithin (x, y) – Ino_endpt(x, y). (1)
Finally, the derived pixel end points are dilated to assist in the final selection of points
for FAZ determination (Figure 5(d)). As shown in Figure 5(d), the points inside the circle
represent the vessel and pathology end points on the perifoveal capillary network that
will be used to determine FAZ. The vessel and pathology end points obtained are then
superimposed on the CLAHE applied image so that the end points on the perifoveal
capillary network can be distinguished and selected as shown in Figure 6. Here, the
vessel and pathology end points are selected manually by marking the points. The marked
points are mainly the vessel and pathology end points that lie at the periphery of the
fovea. Less than 10 points are usually marked for FAZ determination.
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 239

Figure 5 End points detection process using image with mild NPDR: (a) reconstructed retinal
vasculature; (b) image after thinning; (c) image after removal of end points and
(d) end points after dilation (see online version for colours)

(a) (b)

(c) (d)

Figure 6 Process of FAZ determinations: (a) end points superimposed on CLAHE applied image
and (b) selection of points to determine FAZ

(a) (b)

2.3 Determination of Foveal Avascular Zone

The selected points are then connected to each other for FAZ analysis. The connected
points are used to construct perimeter of FAZ area. The constructed FAZ area is then
measured in pixel as shown in Figure 7. The measured area is used in grading of DR that
will be discussed in the next section. The area of the determined FAZ region is computed
as below.
240 M.H. Ahmad Fadzil et al.

∑ ∑
i = imax j = jmax
A( S ) = I ( xi , y j ). (2)
i =0 j =0

I(x, y) is 1 if the pixel is within the shape, (x, y) ∈ S, and 0 otherwise.

Figure 7 Process of FAZ determinations: (a) end points superimposed on CLAHE applied image
and (b) selection of points to determine FAZ

(a) (b)

2.4 Grading of Diabetic Retinopathy

In grading of DR, the FAZ area is measured for several known DR related fundus images
to obtain FAZ area ranges corresponding to the severity or stages of DR. The FAZ area
ranges that overlap show progression of the disease from a DR stage to the next.
The categorisation of each range used in this work is as follows:
a Range 1 – No DR stage
b Range 2 – Progression range from No DR to mild NPDR
c Range 3 – Non-proliferative DR (NPDR) stage
d Range 4 – Progression range from NPDR to severe NPDR/PDR
e Range 5 – Severe NPDR/PDR stage.
During screening, these ranges can assist ophthalmologist in two ways namely,
i to detect the onset of retinopathy and create awareness of the patient to the fact that
changes have begun to occur in their eyes, prompting measures to improve glycemic
and blood pressure control
ii to allow sight-threatening retinopathy to be detected and treated in an attempt
to promote resolution and prevent progression of the disease.

3 Result

In this work, 26 fundus images of patients are used as dataset: 8 images of normal retina
(No DR), 11 images of NPDR and 7 images with PDR. These images are obtained from
an online database provided by School of Medicine of the University of Birmingham
(2001) and from General Hospital of Kuala Lumpur (Karunakar, 2004–2006). The size of
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 241

these images is 640×480 pixels, 8 bits per colour channel in JPEG format. For the
purpose of DR grading, the FAZ area for each image in the dataset is firstly determined.
The determined FAZ area in pixels for the reported diagnoses for each image in the
dataset is tabulated in Table 2.

Table 2 FAZ areas obtained for normal fundus and eyes with DR by the developed algorithm

Code of patient Reported diagnosis FAZ area (pixels)

Norm1 No DR 1670
Norm2 No DR 2815
Norm3 No DR 2877
Norm4 No DR 2340
Norm5 No DR 2819
Norm6 No DR 2731
Norm7 No DR 2670
Norm8 No DR 3407
Aneu1 Mild NPDR (Aneurysms) 3455
Aneu2 Mild NPDR (Aneurysms) 3380
NPDR1 Mild NPDR (Aneurysms around left macula) 3327
NPDR2 Minimal changes, few MAs 3345
NPDR3 One CWS, few DBHs 4164
NPDR4 Some early Exs temporal and above macula, DBHs 4586
NPDR5 DBHs, few HEs 4345
NPDR6 Some Exs at the right side of macula (not DM) 6565
NPDR7 MAs around fovea 4302
NPDR8 Exs near macula 5677
NPDR9 HM around macula 6032
PDR1 VT, CWS, circinate HE, DBHM, ME 6678
PDR1 Pre-PDR 6451
PDR2 Severe ishaemic changes, DM plus ischaemic pre-PDR 7873
PDR3 Severe PDR 7992
PDR4 Fine NV 7248
MAs = Microaneurysms; Exs = Exudates; HExs = Hard Exudates; HM = Haemorrhage;
BHM = Blot Haemorrhages; DBHs = Dot and Blot Haemorrhages; CWS = Cotton Wool
Spot; NVD = Neovascularisation in optic disc; NV = Neovascularisation; ME = Macular
Edema; MEx = Macular Exudates; DM = Diabetic Maculopathy.

The distribution of the FAZ areas is then plotted as shown in Figure 8 to form the above
mentioned ranges. Overlapping of FAZ areas between two stages (based on the reported
diagnosis for each image) indicate the progression of the disease. Non-overlapping areas
indicate DR stages. This is further illustrated in Figure 9.
The FAZ areas for Range 1 (Normal or No DR stage), Range 2 (progression from
normal to mild NPDR), Range 3 (NPDR stage), Range 4 (progression from NPDR to
severe NPDR/PDR) and Range 5 (severe NPDR/PDR stage) are shown in Table 3.
242 M.H. Ahmad Fadzil et al.

Note that the values of the upper and lower bounds for each range are obtained by
rounding the identified FAZ area to the nearest hundredth.

Figure 8 Distribution of FAZ area for dataset showing DR ranges using the developed algorithm
(see online version for colours)

Figure 9 FAZ area ranges for dataset showing DR stages (Ranges 1, 3 and 5) and progression
ranges of DR (Ranges 2 and 4) using the developed algorithm (see online version
for colours)

Table 3 FAZ area ranges for DR stages and progression for given dataset

Name DR stages and progressions Area range (pixels)

Range 1 Normal stage <3300
Range 2 Progression from Normal to Mild NPDR 3300–3400
Range 3 NPDR stage 3400–6500
Range 4 Progression from NPDR to severe NPDR/PDR 6500–6600
Range 5 Severe NPDR/PDR stage >6600

From Figures 7 and 8, it is observed that the FAZ area for patient Norm 8 which is the
maximum FAZ area for normal eyes exceeds the minimum FAZ area of eyes with mild
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 243

NPDR (patient NPDR1). This shows that overlapping of FAZ area between two DR
stages has occurred. Therefore, the FAZ area of both cases (3327 pixels for patient Norm
8 and 3407 pixels for patient NPDR 1 (refer to Table 2) are rounded to the nearest
hundredth to be the range of progression from normal to mild NPDR which will be
3300–3400 pixels (as tabulated in Table 3). This range (Range 2) would be an early
indicator for progression of normal to mild NPDR. The detection of Range 2 cases create
awareness to both ophthalmologists and DR patients so that suitable treatment can be
taken to prevent progression to mild NPDR stage, which is the earliest stage of DR.
Range 1, which is the range for normal FAZ area, would therefore be below the lower
bound of Range 2 that is <3300 pixels while the upper bound of Range 2 becomes the
lower bound of Range 3 that indicates the next stage of DR that is the mild NPDR stage.
Range of FAZ area for the mild NPDR stage is called Range 3.
To identify the upper bound of Range 3, the range showing progression from mild
NPDR to severe NPDR/PDR is firstly investigated. The maximum FAZ area of eyes with
mild NPDR (patient NPDR 6) has found to overlap the minimum FAZ area of eyes with
severe NPDR/PDR (patient Pre-PDR 1) as shown in Figure 9. Thus, these two FAZ areas
are used to obtain Range 4 which is the range showing progression from mild to severe
NPDR/PDR. Taking the round of the FAZ areas to the nearest hundredth, Range 4 is
found to be from 6500 pixels to 6600 pixels. Ophthalmologist can be more attentive
in findings other retinal changes associated in the fundus image that falls in this range
to be more assured of the stages. The lower bound of Range 4 is then taken as the
upper bound of the previous range which is Range 3 (range for mild NPDR). Therefore,
Range 3 is from 3400 pixels to 6500 pixels.
The FAZ areas above the upper bound of Range 4 (>6600 pixels) indicate the severe
NPDR/PDR stage. Usually severe NPDR is diagnosed when few microaneurysms (MAs),
dot and blot haemorrhages (DBHs), and exudates (Exs) can be found in all four quadrants
of a fundus image and at least one of the following signs; cotton wool spot, venous
beading in two or more quadrants and IRMA in one or more quadrants (Khurana, 2008).
The above findings can also be found in very severe NPDR except that instead of one
at least 2 or 3 signs of severe NPDR can be present. For PDR stage, the occurrence of
neovascularisation or proliferation of new, abnormal vessels from the capillaries over the
changes of very severe NPDR can be the main indicator of the stage.

4 Analysis

As pointed out by ophthalmologists (Bresnick et al., 1984; Ballerini, 1998), the

enlargement of FAZ can be an indicator of DR progression. If the FAZ enlargement is
associated with existence of capillary free zone (region devoid of capillary) in any other
parts of the fundus image, this could be a strong indicator of PDR.
Apart from the size of FAZ, the qualitative changes such as irregular contour with
notching and indentations can quantify FAZ alterations in diabetic patients. It is found
that the FAZ dimensions were strongly positively correlated with the severity of capillary
non-perfusion (blockage of capillary) and the presence of PDR (Bresnick et al., 1984).
Ophthalmologists rely on a qualitative description of FAZ such as the degree of its
enlargement (area, diameter, circumference and etc.) and its shape change to detect the
presence and assess the severity of DR (Ballerini, 1998).
244 M.H. Ahmad Fadzil et al.

Currently, this study focused on the analysis of FAZ based on the area size. Using the
dataset, the enlargement of FAZ in DR is seen to be correlated with the severity of DR.
The distribution of FAZ area plotted also shows the potential of grading of DR based
on FAZ area. This is because the ranges showing progression of the disease can be
distinguished when overlaps occur between FAZ areas of different stages. The overlaps
can be identified based on the reported diagnosis shown in Table 2. The accuracies are
found based on the ranges obtained using distribution of FAZ areas by Manual CLAHE
which is the gold standard as shown in Figure 10.
In order to obtain values for each range (normal stage, progression to mild NPDR
range, NPDR stage, progression to severe NPDR/PDR range, and severe NPDR/PDR
stage), the overlapping of FAZ areas between two stages are identified. The values for
each range are then tabulated in Table 4. Note that the values are all rounded to the
nearest hundredth except for values of Ranges 4 and 5 which are rounded to the nearest
tenth due to the small size of Range 4. Another graph is plotted to show the area of
overlapping of FAZ between stages which are shaded in Figure 11. Comparison is made
between these ranges with the ranges obtained from the developed algorithm shown in
Table 5.

Figure 10 Distribution of FAZ area for dataset showing DR ranges using manual segmentation
after CLAHE process (see online version for colours)

Figure 11 FAZ area ranges for dataset showing DR stages (Ranges 1, 3 and 5) and progression
ranges of DR (Ranges 2 and 4) using manual segmentation on images after CLAHE
(see online version for colours)
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 245

Table 4 FAZ area ranges using Manual-CLAHE for DR

Name DR stages and progressions Area range (pixels)

Range 1 Normal stage <3000
Range 2 Progression from Normal to Mild NPDR 3000–3300
Range 3 NPDR stage 3300–6000
Range 4 Progression from NPDR to severe NPDR/PDR 6000–6010
Range 5 Severe NPDR / PDR stage >6010

Table 5 The lower and upper bounds of each range for the results obtained by the proposed
method and Manual-CLAHE with the accuracies for DR

Lower bound (pixels) Upper bound (pixels)

Accuracy Accuracy
Algorithm Manual-CLAHE (%) Algorithm Manual-CLAHE (%)
Range 1 – – – 3300 3000 90.0
Range 2 3300 3000 90.0 3400 3300 96.9
Range 3 3400 3300 96.9 6500 6000 91.7
Range 4 6500 6000 91.7 6600 6010 90.2
Range 5 6600 6010 90.2 – – –

The lower and upper bounds of each range obtained using the developed algorithm and
the Manual-CLAHE are compared as tabulated in Table 5. The accuracy of each bound
obtained by the developed algorithm is calculated as follows:
  Manual-CLAHE − Algorithm  
Accuracy for Algorithm-Ranges =  1 −   × 100  %. (3)

  Manual-CLAHE  
From Table 5, it can be observed that the obtained accuracy for the lower bounds and the
upper bounds of the ranges for DR grading using the developed algorithm is between
90.2–96.9%. The mean accuracy is about 92.2% and the standard deviation is about 3.22.
The developed method for determination of FAZ is a semi-automated approach
where human intervention is necessary during determination of FAZ. A manual FAZ
determination (manual segmentation) using enhanced fundus image by CLAHE is used
to obtain the gold standard. This type of manual segmentation is chosen as the user
can select points interactively without any limitations (without using detected end points).
In a CLAHE enhanced fundus image, the fine vessels in the macula region are
clearly visible and thus make FAZ determination more accurate as long as large DR
pathologies are not affecting the region. The determined FAZ by this method is used
as gold standard for comparison with the determined FAZ by the developed algorithm
and by Manual-Colour. One of the drawbacks of this manual segmentation is the time
spent to select 20 points or more to determine FAZ. The selection of the points is also
repetitive as a single mistake will result in starting the process from the beginning
again and again. Therefore this approach is time consuming and not suitable for medical
practice. The determined FAZ by this method is used as gold standard for comparison
with the determined FAZ by the developed algorithm and by Manual-Colour.
Normally, to analyse FAZ area, DR patients have to undergo fluorescein angiogram
procedure. If the original colour fundus images are analysed, ophthalmologists will have
246 M.H. Ahmad Fadzil et al.

some difficulties to view and analyse the FAZ. The accuracy achieved by this method is
in the range of 16.6–98.9%. An enhanced fundus image could assist ophthalmologist in
viewing and analysing FAZ more effectively.
The range of accuracy achieved by the developed algorithm in determination of FAZ
is from 69.1% to 99.8% with the mean accuracy of 90.9% and standard deviation of 7.53.
The range of accuracy obtained by Manual-Colour (ophthalmologist standard) is found to
be from 16.6% to 98.9% with a lower mean accuracy of 78.3%. The standard deviation
of the accuracy obtained by Manual-Colour is as high as 20.7 that show the inconsistency
based on manual (visual) method. Therefore, based on the mean accuracy, the developed
method is able to improve the ophthalmologist standard by 16.1%. The standard
deviation of obtained FAZ area by the developed method is smaller by a factor of 2.8.
This improvement is calculated based on equation (4) stated below.
|value1 − value 2 |
Improvement = . (4)
value 2

This shows that the developed algorithm is more consistent in determining FAZ region
and thus more reliable than using colour fundus image in the analysis of FAZ that is
usually practiced by the ophthalmologist for mostly early DR cases.
The accuracy of determined FAZ region can be increased if the accuracy of vessels
detection and reconstruction can be improved or if the under segmentation at the
capillaries of perifoveal network could be reduced. This is because the extraction of end
points along vessels for determination of FAZ boundary is based on the detected and
reconstructed vessels in the earlier process. It can be said that the accuracy of FAZ area
is affected by the segmentation of the retinal vasculature.
For the purpose of DR grading, FAZ area of normal eyes (8 fundus images) and
eyes with DR (non-proliferative (NPDR) – 11 images, proliferative (PDR) – 7 images)
are computed and plotted. From the FAZ area distribution, it is found that there is
overlapping of FAZ areas between two stages. The overlapping ranges are highlighted to
indicate progression of the disease to another stage. The FAZ area that overlaps the later
stage is used as the upper bound of the ranges while the minimum FAZ area of the later
stage is used as the lower bound of the ranges. The area ranges that show progressions of
the disease stage by stage are analysed and highlighted in the same graph. The lower
bounds may indicate the maximum FAZ area for the previous stage and thus will be the
indicator of progression to the next stage while the upper bounds may indicate the
maximum FAZ area for progression to the next stage. Area within the progression range
shows that the disease is in high risk of progressing to the next stage. Using the
developed method, five ranges of FAZ area for DR grading are obtained as shown in
Table 6.

Table 6 FAZ area ranges using developed method

Name DR stages and progressions Area range (pixels)

Range 1 Normal stage <3300
Range 2 Progression from normal to mild NPDR 3300–3400
Range 3 NPDR stage 3400–6500
Range 4 Progression from NPDR to severe NPDR/PDR 6500–6600
Range 5 Severe NPDR/PDR stage >6600
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 247

The mean accuracy obtained for the lower bounds and the upper bounds of the ranges
using the developed algorithm is 92.2% based on the ranges obtained by Manual-CLAHE
(gold standard). Currently, the accuracies on the ranges obtained for DR grading using
the developed method are obtained based on ranges obtained by the manual
segmentation. In grading of DR, the mean accuracy and standard deviation of ranges
obtained using the developed method is 92.2% and 3.22 respectively for both lower and
upper bounds. From the overall results obtained, based on the correlation between FAZ
area and severity of DR, it is believed that the size of FAZ can be used for grading of
normal to severe NPDR/PDR.
The limitation of the developed algorithm is on under segmentation of the retinal
blood vessels. Under segmentation mainly occurs at fine vessels or capillaries in
detection and reconstruction of retinal vasculature. Therefore, the accuracy of the vessel
end points obtained for FAZ determination is also influenced by under segmentation
(false negative) on the retinal vasculature. This will result in enlarging of the FAZ region.
This explains the slightly larger computed values of FAZ areas by the developed
method compared to manually computed FAZ areas using CLAHE applied image
(gold standard).

5 Conclusion

A new approach for analysing of FAZ for grading the severity of DR has been developed
(Ahmad Fadzil and Izhar, 2008). It is based on the binary map of retinal vasculature
obtained from the extraction and reconstruction of retinal vasculature where the vessel
ends and pathologies surrounding FAZ are derived for accurate determination of the
FAZ. The performance of the FAZ determination is based on manual segmentation
using enhanced image by CLAHE which is set as the gold standard. Another manual
segmentation is performed using original colour fundus images that represent the
current way of FAZ determination by ophthalmologists. Based on the gold standard,
the developed method is found to improve the accuracy of current method by
ophthalmologists by 16.1%. The standard deviation achieved by the developed method
is also smaller by a factor of 2.8. This shows that the developed algorithm is more
consistent in determining FAZ region and thus more reliable than using colour fundus
image in the analysis of FAZ that is usually practiced by the ophthalmologist for mostly
early DR cases.
The paper has shown that FAZ area has a potential to be used for grading of DR
(no DR to severe NPDR/PDR) with good accuracy based on its correlation with the
severity of DR. The most important objective of the DR grading based on FAZ area is to
obtain FAZ area ranges indicating the stage and progression of the disease. In grading
of DR, the mean accuracy and standard deviation of ranges obtained using the developed
method is 92.2% and 3.22 respectively for both lower and upper bounds. If an FAZ area
lies within the progression ranges obtained, actions could be taken to prevent the
progression of the disease and towards blindness. As enlargement of FAZ may also be
due to hypertension (Khurana, 2008). If a non-diabetic patient is diagnosed to have FAZ
enlargement, the patient may have a possibility of suffering of high blood pressure
or hypertension. This new approach is found reliable to determine and analyse FAZ
with good and better accuracy compared to the current method of making diagnosis
248 M.H. Ahmad Fadzil et al.

by ophthalmologists. By using the developed approach, early onset of DR can be detected

as enlargement of FAZ is usually observed in early DR stages.

Acknowledgement

The authors would like to acknowledge the contributions of Dr. Karunakar, Department
of Ophthalmology, Hospital Kuala Lumpur and Professor P.A. Venkatachalam for their
contributions to the early part of the research work.

References
Ahmad Fadzil, M.H. and Izhar, L.I. (2008) A Non-Invasive Method for Analysing the Retina
for Ocular Manifested Diseases, Malaysia.
Ahmad Fadzil, M.H., Izhar, L.I., Venkatachalam, P.A. and Karunakar, T.V.N. (2007) ‘Extraction
and reconstruction of retinal vasculature’, Journal of Medical Engineering Technology,
Vol. 31, pp.435–442.
Ahmad Fadzil, M.H., Nugroho, H.A., Nugroho, H. and Izhar, L.I. (2009) Fundus Image Database
for Non Invasive Diabetic Retinopathy Monitoring and Grading System (Finders), Universiti
Teknologi PETRONAS, Perak, Malaysia.
Anonymous (1996) Screening for Diabetic Retinopathy: National Malaysian Eye Survey, Ministry
of Health, Malaysia.
Anonymous (2003) American Academy of Ophthalmology. Preferred Practice Pattern: Diabetic
Retinopathy 2003, American Academy of Ophthalmology, San Francisco, California.
Anonymous (2004) Disease Center: Wrong Diagnosis, Statistics by country for type 2 diabetes.
Balasubramanian, S., Khanna, S. and Chandrasekaran, V. (2007) ‘Localization of optic disk using
independent component analysis and modified structural similarity measure’, MVA2007 IAPR
Conference on Machine Vision Applications, Tokyo, Japan.
Ballerini, L. (1998) ‘Detection and quantification of foveal avascular zone alterations in Diabetic
Retinopathy (DR)’, 5th Internet World Congress for Biomedical Sciences – INABIS 1998.
McMaster University, Hamilton, Ontario, Canada.
Bradley, A., Applegate, R.A., Zeffren, B.S. and Heuven, W.A.J. (1992) ‘Psychophysical
measurement of the size and shape of the human foveal avascular zone’, Ophthalmic and
Physiological Optics, Vol. 12, pp.18–23.
Bresnick, G.H., Condit, R., Syrjala, S., Palta, M., Groo, A. and Korth, K. (1984) ‘Abnormalities
of the foveal avascular zone in diabetic retinopathy’, Arch Ophthalmol., Vol. 102,
pp.1286–1293.
Conforth, D.J., Jelinek, H.J., Leandro, J.J.G., Soares, J.V.B., Cesar Jr., R.M., Cree, M.J.,
Mitchell, P. and Bossomaier, T. (2004) ‘Development of retinal blood vessel segmentation
methodology using wavelet transforms for assessment of diabetic retinopathy’, Eighth Asia
Pacific Symposium on Intelligent and Evolutionary Systems, Cairns, Australia.
Conrath, J., Giorgi, R., Raccah, D. and Ridings, B. (2004) ‘Foveal avascular zone in diabetic
retinopathy: quantitative vs qualitative assessment’, Eye, Vol. 19, pp.322–326.
Conrath, J., Valat, O., Giorgi, R., Adel, M., Raccah, D., Meyer, F. and Ridings, B. (2006)
‘Semi-automated detection of the foveal avascular zone in fluorescein angiograms in diabetes
mellitus’, Clinical and Experimental Ophthalmology, Vol. 34, pp.119–123.
Cree, M.J., Olson, J.A., Mchardy, K.C., Forrester, J.V. and Sharp, P.F. (1996) ‘Automated
microaneurysm detection’, International Conference on Image Processing, 1996.
Proceedings, Lausanne, Switzerland, Vol. 3, pp.699–702.
 Analysis of Foveal Avascular Zone for grading of Diabetic Retinopathy 249

Fleming, A.D., Goatman, K.A., Philip, S., Olson, J.A. and Sharp, P.F. (2007) ‘Automatic detection
of retinal anatomy to assist diabetic retinopathy screening’, Physics in Medicine and Biology,
Vol. 52, pp.331–345.
Goh, P.P. (2008) ‘Status of diabetic retinopathy among diabetics registered to the diabetic eye
registry, national eye database, 2007’, The Medical Journal of Malaysia, Vol. 63, pp.24–28.
Goh, P.P., Elias, H., Fariza, N.N. and Mariam, I. (2008) ‘National eye database – a web based
surveillance system’, The Medical Journal of Malaysia, Vol. 63, pp.20–23.
Haddouche, A., Rasigni, M., Adel, M. and Bourennane, S. (2005) ‘Statistical method for foveal
detection in retinal angiograms’, Electronics Letters, Vol. 41, pp.632–633.
Hoover, A.D., Kouznetsova, V. and Goldbaum, M. (2000) ‘Locating blood vessels in retinal
images by piecewise threshold probing of a matched filter response’, Medical Imaging, IEEE
Transactions on, Vol. 19, pp.203–210.
Hunter, A., Lowell, J., Owens, J., Kennedy, L. and Steele, D. (2000) ‘Quantification of diabetic
retinopathy using neural networks and sensitivity analysis’, in Malmgren, H., Borga, M.
and Niklasson, L.F. (Eds.): Artificial Neural Networks in Medicine and Biology, Springer,
Goteborg, Sweden.
Ibanez, M.V. and Simo, A. (1999) ‘Bayesian detection of the fovea in eye fundus angiographies’,
Pattern Recognition Letters, Vol. 20, pp.229–240.
John, D., Braganza, A. and Kuriakose, T. (2006) ‘A study of the foveal avascular zone using the
heidelberg retina angiogram-2 in normal eyes’, 34th All India Optometry Conference (AIOC
2008) Proceedings, Amritsar, India, Vol. 1, pp.485–486.
Jun Lee, S. and Jun Koh, H. (2008) ‘Enlargement of the foveal avascular zone in diabetic
retinopathy after adjunctive intravitreal bevacizumab (avastin) with pars plana vitrectomy’,
Journal of Ocular Pharmacology and Therapeutics.
Karunakar, T.V.N. (2004–2006) Kuala Lumpur, Ophthalmology Department, General Hospital,
Kuala Lumpur, Malaysia.
Khurana, A.K. (2003) Ophthalmology, New Age International Publishers, New Delhi.
Khurana, A.K. (2008) Comprehensive Ophthalmology, Tunbridge Wells, Anshan, UK.
Li, H. and Chutatape, O. (2000) ‘Fundus image features extraction’, Engineering in Medicine and
Biology Society, 2000. Proceedings of the 22nd Annual International Conference of the IEEE,
Chicago, IL, USA, Vol. 4, pp.3071–3073.
Li, H. and Chutatape, O. (2001) ‘Automatic location of optic disk in retinal images’, International
Conference on Image Processing, 2001. Proceedings, Thessaloniki, Greece, Vol 2,
pp.837–840.
Mansour, A.M. (1990) ‘Measuring fundus landmark’, Invest. Ophthalmol Vis. Sci., Vol. 31,
pp.41–42.
Nguyen, H., Roychoudhry, A. and Shannon, A. (1997) ‘Classification of diabetic retinopathy
lesions from stereoscopic fundus images’, Engineering in Medicine and Biology society, 1997.
Proceedings of the 19th Annual International Conference of the IEEE, Chicago, IL, USA,
Vol. 1, pp.426–428.
Osareh, A. (2004) Automatic Identification of Diabetic Retinal Exudates and Optic Disk,
Department of Computer Science, University of Bristol, Bristol, UK.
Osareh, A., Mirmehdi, M., Thomas, B. and Markham, R. (2003) ‘Automated identification of
diabetic retinal exudates in digital colour images’, Br. J. Ophthalmol., Vol. 87, pp.1220–1223.
Parodi, M.B., Visintin, F., Rupe, P.D. and Ravalico, G. (1995) ‘Foveal avascular zone in macular
branch retinal vein occlusion’, International Ophthalmology, Vol. 19, pp.25–28.
Patton, N., Aslam, T.M., Macgillivray, T., Deary, I.J., Dhillon, B., Eikelboom, R.H., Yogesan, K.
and Constable, I.J. (2006) ‘Retinal image analysis: concepts, applications and potential’,
American Journal of Ophthalmology, Vol. 141, p.603.
250 M.H. Ahmad Fadzil et al.

Poonguzhali, S. and Ravindran, G. (2007) ‘Evaluation of feature extraction methods for

classification of liver abnormalities in ultrasound images’, International Journal of
Biomedical Engineering and Technology (IJBET), Vol. 1, pp.134–143.
Report No. 10, O. (1991) Grading Diabetic Retinopathy from Stereoscopic Color Fundus
Photographs-an Extension of the Modified Airlie House Classification, ETDRS (Early
Treatment Diabetic Retinopathy Study).
Report No. 12, O. (1991) Fundus Photographic Risk Factors for Progression of Diabetic
Retinopathy, ETDRS (Early Treatment Diabetic Retinopathy Study).
Richard, G., Soubrane, G.L., Yannuzzi, L.A. and Courland, S. (1998) Fluorescein and ICG
Angiography, Thieme Medical Publisher Inc., New York.
Saini, V.K., Varma, P., Bhaisare, V., Kulkarni, S., Sodani, P., Maheshwai, D. and Pawar, N. (2006)
‘Foveal avascular zone calculation and its variation with different posterior segment diseases
and analysis of its impact on best corrected visual acuity’, AIOC 2006, Bhopal, India, Vol. 4,
pp.538–539.
Sander, B., Larsen, M., Engler, C. and Lund-Andersen, H. (1994) ‘Absence of foveal vascular zone
demonstrated by laser scanning fluorescein angiography’, Acta Ophthalmol (Copenh), Vol. 72,
pp.550–552.
School of Medicine of the University of Birmingham (2001) Med. Web Technology Test Area,
School of Medicine of the University of Birmingham.
Sinthanayothin, C., Boyce, J.F., Cook, H.L. and Williamson, T.H. (1999) ‘Automated localisation
of the optic disc, fovea, and retinal blood vessels from digital colour fundus images’,
Br. J. Ophthalmol., Vol. 83, pp.902–910.
Sinthanayothin, C., Boyce, J.F., Williamson, T.H., Cook, H.L., Mensah, E., Lal, S. and Usher, D.
(2002) ‘Automated detection of diabetic retinopathy on digital fundus images’, Diabetic
Medicine, Vol. 19, pp.105–112.
Sinthanayothin, C., Kongbunkiat, V., Phoojaruenchanachai, S. and Singalavanija, A. (2003)
‘Automated screening system for diabetic retinopathy’, Proceedings of the 3rd International
Symposium on Image and Signal Processing and Analysis, 2003. ISPA 2003, Rome, Italy,
Vol. 2, pp.915–920.
Sopharak, A., Uyyanonvara, B., Barman, S. and Williamson, T. H. (2008) ‘Automatic detection of
diabetic retinopathy exudates from non-dilated retinal images using mathematical morphology
methods’, Computerized Medical Imaging and Graphics, Vol. 32, pp.720–727.
Teng, T., Lefley, M. and Claremont, D. (2002) ‘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.
Walter, T., Klein, J-C., Massin, P. and Erginay, A. (2002) ‘A contribution of image processing
to the diagnosis of diabetic retinopathy-detection of exudates in color fundus images of the
human retina’, Medical Imaging, IEEE Transactions on, Vol. 21, pp.1236–1243.
Wong, T.Y., Cheung, N., Tay, W.T., Wang, J.J., Aung, T., Saw, S.M., Lim, S.C., Tai, E.S.
and Mitchell, P. (2008) ‘Prevalence and risk factors for diabetic retinopathy: the Singapore
Malay eye study’, Ophthalmology, Vol. 115, pp.1869–1875.
Zeffren, B., Applegate, R., Bradley, A. and Van Heuven, W. (1990) ‘Retinal fixation point location
in the foveal avascular zone’, Invest. Ophthalmol. Vis. Sci., Vol. 31, pp.2099–2105.