COMPUTERIZED is unusual leakage and accumulation of fluid

COMPUTERIZED MACULAR EDEMA DETECTION USING
OCT IMAGES: A REVIEW

Athira S C1, Reena M Roy2

1P G Scholar, L.B.S Institute of Technology for women,
Poojappura, Trivandrum. [email protected]

 

2Assistant
Professor, L.B.S Institute of Technology for women, Poojappura, Trivandrum. [email protected]

 

Abstract— This
article reviews the existing approaches that are used for computerized
detection of Macular Edema (ME), which is one of the leading causes of
blindness among majority. It takes place when there is unusual leakage and
accumulation of fluid inside the macula from broken blood vessels within the
nearby retina. Any disorder that damages blood vessels inside retina can
reason. The signs of ME are not apparent in the early stages and it is
difficult to diagnose at its advanced stages. Optical coherence tomography
(OCT) is the modern day eye exam approach which gives cross sectional location
of retinal layers, and can be utilized for the earlier detection of ME.
Different algorithms were implemented to detect ME from OCT images. This article
focuses on identifying a computerized algorithm that could detect ME at earlier
stages with great level of accuracy.

Keywords—Macular Edema, Macular Disorder, Optical
Coherence Tomography, Central Serous Retinopathy, Macular Hole, Age-related
Macular Degeneration.

 

I.  INTRODUCTION

 

The
human eye has been called the most compound organ in our body. It’s an astonishing
fact that something so small can have so many working parts. There are three
layers which constitute human eye. Sclera, the outer layer which protects the
eye ball. Cornea is a transparent membrane which forms the outer coating. It
will provide a protection to iris and pupil. The main function is to focus
light onto the retina. Beneath sclera, the second layer exists which is called
choroid. It is a vascular layer which has blood vessels and responsible to provide
nourishment to the eye. Beneath choroid, there exists retina, the inner most
layer of an eye. Retina is responsible for eye vision and pigmentation with the
presence of photoreceptors 1. The macula is the small area at the centre of
the retina and is responsible for central vision. It
determines what we see straight in front of us, at the centre of our field of
vision. It is a very important part as it gives us the vision needed for basic
activities such as reading and writing, and the ability to appreciate color etc
2. It is an oval-shaped pigmented area near the center of the retina,
with a diameter of approximately 5.5 mm in human eyes 2. Near the center
of the macula is a tiny dip packed with light-sensitive cells called fovea 3.
The fovea picks up the finest details of central vision. The basic structure of
human Eye and macula is shown in Figure 1 and Figure 2 respectively.

 

 

 

 

 

 

 

 

 

 

 

Fig. 1. Eye structure.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Fig. 2. Basic structure of human macula.

 

 

 

 

 

 

A. Macular Disorders

 

    
Macular disorders mainly damage the macular region of the retina, and
can greatly affect the central vision of a person 3. There are different
types of macular disorders, but some common ones are Macular Edema (ME), Central Serous Retinopathy (CSR), Macular Hole (MH), Age
related Macular Degeneration (AMD), Choroidal Neovascularization,
Pigment Epithelium Detachment and the Förster–Fuchs retinal spot 3. Most of
these macular disorders are curable if diagnosed at an early stage. The common
symptoms of macular impairments includes

1) Issues with the central vision: This may be characterized
as something appearing to obstruct the central vision or a blurred patch 5.

 

2) Metamorphopsia: This
is characterized by a distortion of images – especially of straight lines 5.

 

3) Distortion of image size: Objects may appear bigger
(macropsia) or smaller (micropsia). This may in turn give rise to diplopia, as
there is a discrepancy between the image perceived in the healthy eye and in
the diseased eye 5.

 

                In the case of ME, the retinal layers are swollen due
to the leakage of fluid from retinal blood vessels. There are two major causes
of ME. The first one is diabetes, where small blood capillaries within the
retina start leaking fluid 2. In this case, ME is termed diabetic macular
edema (DME) 2. Eye (cataract) surgery may also increase the risk of
developing macular edema due to irritated blood vessels and fluid leakage 3.
In this case, ME is termed cystoid macular edema (CME). CSR occurs due to the
accumulation of serous fluid beneath the retina and causes the retinal layers
to detach 3. There are two types of CSR. In Type 1 CSR, the fluid accumulates
under the neurosensory retina. Type II is characterized by the accumulation of
fluid in the retina due to retinal pigment epithelium (RPE) leakage. Serous
fluid in such cases tends to be shallower rather than domed shaped. Common
symptoms of ME and CSR are dim, blurred, and distorted central vision 2.

 

 

 

 

 

 

 

 

 

 

 

Fig.
3. ( a) Healthy person vision. (b) ME affected vision.

 

 

 

 

 

                There are multiple techniques
that are used to detect retinal disorders. Some common techniques are fundus
photography, fundus fluorescein angiography and OCT 2. OCT is the recently
developed technology to detect macular diseases such as ME, Macular Hole,
Central Serous Retinopathy, Choroidal Neovascularization and Pigment Epithelium
Detachment. The principle advantage of using OCT images is that it can assist
in the detection of the diseases in the very earlier stages using appropriate
techniques 3.

 

B. Optical
Coherence Tomography:

 

     Optical Coherence Tomography, or OCT
technology, allows optometrist to take cross sectional images of your retina,
commonly referred to us OT scan 1. It’s a medical imaging technique that uses
light to capture micrometer-resolution, three-dimensional images from within
optical scattering media like biological tissues. It is based on low coherence
interferometry and utilizes near infrared light. It employs Michelsons
interferometer principle 2.OCT imaging technology mainly consists of OCT
camera which uses a low coherence interferometry in which low coherence visible
light is allowed to penetrate human retina and it is reflected back to
interferometer producing a cross sectional image of retina. The OCT imaging
technique and OCT images of the macula are shown in figure 4 and figure5
respectively.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Fig. 4.
OCT imaging technique is based on Michelson type interferometer.

 

 

 

 

 

 

 

 

 

Fig. 5. (a) OCT
image of healthy macula (b) OCT image of Macular edema

 

 

II. LITERATURE REVIEW

 

         This
section review the existing methods, which gives more details about automatic
detection .Most of the existing systems are based on figure 6.

 

 

Pre
processing
 
i

Segmentation

Feature extraction

Feature set formulation

Classification

ME detection

Input OCT images

 

 

 

 

 

 

 

 

 

 

Fig. 6. Common systems
for detection of ME

A.  Detection using Fundus photography

     Fundus photography is used to capture an
image of the back of the eye that is fundus. It gives the three dimensional
image of the retina and choroidal tissues. Ophthalmologists make use of the
fundus images to detect and diagnose patients with ME. This approach is highly
useful and accurate, provided there is a sharp fundus image obtained. But it
requires a sharp and well focused fundus camera image. To make proper judgments
the ophthalmic photographer must be well versed with the layered anatomy of the
posterior pole and its relationship to the disease. The adjustment of fundus
camera eyepiece and the different step to be followed for focusing the fundus
camera is relatively a difficult task. The coordination among the photographer,
patient, and camera determines the final sharpness of the image. Special
techniques must be designed for optimizing the sharpness in stereo images. The
detection of macular disorders using the fundus images is therefore not a
feasible option since it contains several parameters and predefined knowledge
that determines the accuracy of results. This approach guarantees a sensitivity
of 100% and specificity ranging between 74% and 90% 10.

 

B. Detection using
Texture features and Classification using SVM Classifier

 

     This approach concentrates on bringing
down the computational time. It utilizes a computerized method for texture
feature extraction. The extraction is carried within a specified radius with
macula as the centre. Proper segmentation techniques must be applied to extract
the features. The texture features can vary greatly since the extracted region
can contains a great amount of abnormalities like micro-aneurysms,
hard-exudates and hemorrhages. This methodology can be used to classify
diabetic macular edema

 

 

into stage 0
(Normal) and stage 2 (Abnormal) based on the extracted features. Accompanying
the texture feature extraction, it utilizes a Support Vector Machine (SVM)
classifier for grading.  Sensitivity,
Specificity and Accuracy are the parameters that determine the performance of
this system. Experimental results have proved that it can provide 91%, 75% and
86 % Sensitivity, Specificity and Accuracy respectively 8.

 

C. Detection of diabetic retinopathy with image processing

    

     Diabetic retinopathy can be detected by
the application of image processing techniques on the color fundus images. It
can be done through image enhancement, mass screening that includes detection
of pathologies and retinal features. It is followed by monitoring stage. The
feature detection and registration of retinal images is carried out during this
stage. To diagnose the disease properly, it needs the detection of exudates.
But this approach cannot distinct between hard exudates and soft Exudates with
the proposed technique. Hence the application of the image processing technique
in detecting Macular Edema is limited 6.

 

 

D.   Detection using automated feature extraction
technique.

      

       This technique focuses on localizing the
different features and lesions extracted from the fundus retinal image. It
places a constraint for optic disk detection where we first detect the major
blood vessels and use the intersection of these blood vessels to determine the
approximate location of the optic disk, which can be further assisted by the color
properties. The region around the optic disk can be further divided into four
quadrants in order to determine the severity of the disorders. The application of
thesis approach on a database containing 516 images with varied contrast,
illumination and disease stages posses 97.1% success rate for optic disk
localization 6.

 

E.  Detection
of ME on the based on OCT  

     This approach utilizes the OCT retinal images
along on the Support Vector machine classifier to detect Macular Edema
automatically. The Support Vector is trained using the distinct features
extracted from labeled images with macular disorders. The experiments were
carried on a local dataset acquired from AFIO and the results states that the
algorithm worked 88 out of 90 times. It provides an accuracy rate of 97.77 %,
sensitivity of 100 % and specificity of 93.33%. This approach reduces the
computational time to a significant level 1.

 

 

III. RESULTS AND DISCUSSION

 

                There
exist different approaches to detect Macular Edema. The different methods and
the experimental results are shown in the table I. The main parameters that
determine the effectiveness of a method are accuracy, sensitivity and specificity.
Based on the calculations, it can be noticed that Bilal Hassan et.al achieves the best results with
97.77% of accuracy, which is the structure tensor based Macular Edema and
Central Serous Retinopathy detection using the OCT images. A Support Vector
Machine classifier is trained with distinct features that are extracted from
numerous labeled images. Additionally it provides 93.35 % of specificity and
100% sensitivity. It requires computationally less time and is quite
fast.  Annu Anna Lal et al are not behind with 97.1 % accuracy rate. The
results becomes better based on the algorithm used for segmentation, feature
extraction, feature set formulation and classification which forms the backbone
of any technique.   

 

TABLE I:  Result analyses
of existing methods

Subjects

Specifications

Accuracy

Specificity

Sensitivity

 
Aditya   Kunwar et.al8

 
86%

 
75%

 
91%

 
Annu Anna Lal et.al6

 
97.1%

 
—–

 
—–

 
Bilal Hassan et.al1

 
97.77%

 
93.35%

 
100%

 

 

 

 

 

 

 

 

IV. CONCLUSION AND FUTURE SCOPE

In
this paper we discussed on the computerized detection of the various macular or
retinal diseases. Various techniques can be employed to detect the Macular
edema. From the analysis it is clear that the system proposed by Bilal Hassan
et.al 1 can detect ME with about 98 percent accuracy and 100 percent
sensitivity. But the earlier automatic and accurate detection is not defined so
far. Additionally, the major issue with the macular edema is that if they are
not diagnosed and treated at the initial stages then the chances of the
recovery is very feeble. Hence it demands an algorithm which can be deployed on
the OCT images of a retina to detect and classify the ME at early stages. We
have reviewed several techniques that can be utilized to detect ME and reached
a conclusion that the OCT image when treated with SVM provides much better
results. Going forward in future we can develop,

 1) A single algorithm that can be used to
detect all the types of macular disorders at one go such as, tractional retinal
detachment, PEDs, and choroidal neovascularization in one go.

 2) An algorithm should be developed for the
grading of the common retinal diseases.

 

 

 3) Additionally an approach can be made which
calculates the thickness level between the ILM and RPE from circular optic
nerve head scans in order to detect the ocular diseases.

 

 

REFERENCES

1     Hassan, Bilal, et al. “Structure tensor based automated
detection of macular edema and central serous retinopathy using optical
coherence tomography images.” JOSA A 33.4 (2016):
455-463.

2     Hassan, Taimur, et al. “Review of OCT and fundus images
for detection of Macular Edema.” Imaging Systems and Techniques
(IST), 2015 IEEE International Conference on. IEEE, 2015.

3     Novosel, Jelena, et al. “Joint Segmentation of Retinal
Layers and Focal Lesions in 3-D OCT Data of Topologically Disrupted
Retinas.” IEEE Transactions on Medical Imaging 36.6
(2017): 1276-1286.

4     Hassan, Bilal, and Gulistan Raja. “Fully automated
assessment of Macular Edema using Optical Coherence Tomography (OCT)
images.” Intelligent Systems Engineering (ICISE), 2016
International Conference on. IEEE, 2016.

5     Awan, Zahid Hussain, P. S. Mahar, and M. Saleh Memon.
“Blindness and poverty.” Pak J Ophthalmol 27.3
(2011): 165-170.

6     Lal, Annu Anna, and Jomina John. “A Survey on Diabetic
Retinopathy Detection Techniques.” International Journal of
Computer Technology and Applications 4.4 (2013): 705.

7     Deepak, K. Sai, and Jayanthi Sivaswamy. “Automatic
assessment of macular edema from color retinal images.” IEEE
Transactions on medical imaging 31.3 (2012): 766-776.

8     Kunwar, Aditya, Shrey Magotra, and M. Partha Sarathi.
“Detection of high-risk macular edema using texture features and
classification using SVM classifier.” Advances in Computing,
Communications and Informatics (ICACCI), 2015 International Conference on.
IEEE, 2015.

9     Massich, Joan, et al. “Classifying DME vs normal SD-OCT
volumes: A review.” Pattern Recognition (ICPR), 2016 23rd
International Conference on. IEEE, 2016.

10 
Saine, P. J. “Focusing the fundus camera:
a clinical approach.” J. Opthalmic Photography 14 (1992):
7-24.

11  Kunwar, Aditya, Shrey Magotra, and M. Partha
Sarathi. “Detection of high-risk macular edema using texture features and
classification using SVM classifier.” Advances in Computing,
Communications and Informatics (ICACCI), 2015 International Conference on.
IEEE, 2015.

12 
Massich,
Joan, et al. “Classifying DME vs normal SD-OCT volumes: A
review.” Pattern Recognition (ICPR), 2016 23rd International
Conference on. IEEE, 2016.

13  Faust, Oliver, et al. “Algorithms for the
automated detection of diabetic retinopathy using digital fundus images: a
review.” Journal of medical systems 36.1 (2012): 145-157.