layers such as the formation of cotton wool spots in the nerve fiber layer, it is ... These spots are often followed within a few days or weeks by a much greater ...
LESION DETECTION USING SEGMENTED STRUCTURE OF RETINA -A SVM CLASSIFIER APPROACH
By PROF ANOOP B K ASSISTANT PROFESSOR DEPARTMENT OF ECE VIMAL JYOTHI ENGINEERING COLLEGE KANNUR KERALA
PROF DR S PERUMAL SANKAR PROFESSOR DEPARTMENT OF ECE ToCH INSTITUTE OF SCIENCE AND TECHNOLOGY ERNAKULAM KERALA
CONTENTS Contents
Page No.
LIST OF TABLES
ii
LIST OF FIGURES
iii
ABBREVIATIONS
iv
NOTATION
v
Chapter 1. INTRODUCTION
1
Chapter 2. LITERATURE SURVEY
16
Chapter 3.LESION DETECTION USING SEGMENTED STRUCTURE OF RETINA 22 3.1 Pre-processing
25
3.1.1 Gaussian filter
26
3.1.2 Adaptive histogram equalization
28
3.2 Segmentation
29
3.2.1 fuzzy c means clustering
30
3.2.2 fuzzy c means algorithm
32
3.2.3 flowchart
33
3.3 Feature extraction and selection 3.3.1 Morphological feature extraction 3.4 Classification
34 35 37
3.4.1 SVM classifier
44
Chapter 4. RESULTS AND DISCUSSION
47
4.1 Performance evaluation
49
Chapter 5. CONCLUSION
52
REFERENCES
i
LIST OF TABLES No. 4.1
Title
Page No.
PSNR and RMSE value of images
ii
14
LIST OF FIGURES No.
Title
Page No.
1.1
Fundus image
1
1.2
An OCT(optic coherence tomography)
2
1.3
Normal vision
8
1.4
Same view with DR
8
1.5
Retinal image with lesion
1
1.6
Retinal image with related pathologies
24
3.1
Flow diagram
26
3.2
Structural element
35
3.3
Dilation
36
3.4
Erosion
37
3.5
Opening
38
3.6
Closing
41
3.8
SVM classification
42
4.1
Input image
43
4.2
Gaussian image
45
4.3
Histogram Image
48
4.4
Fuzzy C segmentation
48
4.5
Feature extraction
48
4.6
SVM output
48
4.7
Extracted abnormal image
48
iii
LIST OF ABREVIATIONS
AHE
DR
Adaptive Histogram Equalization Diabetic Retinopathy
FCM
Fuzzy C Means
MA
Microaneurysms
MSE
Mean Square Error
PSNR
Peak signal to noise ratio
RMSE
Root mean square error
SVM
Support vector machine
iv
NOTATIONS C
Cluster
I
Image
g(x,y)
Gaussian function
p
Pixel
*
Convolution
U(k)
Membership function
Weight vector
0 n
Bias No of data points
.
v
CHAPTER 1 INTRODUCTION Fundus photography is capturing the photograph of the back of the eye i.e. fundus. Specialized fundus cameras mainly consist of an intricate microscope attached to a flashed enabled camera, it is used for fundus photography. The main structures can be visualized on a fundus photo are the central and peripheral retina, optic disc and macula. Fundus photography can be performed with colored filters, with specialized dyes including fluorescein and indocyanine green. The models and technology of fundus photography has advanced and evolved suddenly over the last century. Since the equipment’s are sophisticated and challenging to manufacture to clinical standards, only a few manufacturers/brands are available in the market Topcon, Canon, CSO and Centervue are some example of fundus camera manufacturers.
Fig 1.1 fundus image
The optical design of fundus cameras is mainly based on the principle of monocular indirect ophthalmoscopy. A fundus camera provides an upright, magnified view of the fundus. A typical camera field of views 30 to 50° of retinal area, with a magnification of
1
2.5x, and allows some modification of this relationship through zoom or auxiliary lenses from 15°, which provides 5x magnification, to 140° with a wide angle lens, which minifies the image by its half. The optics of a fundus camera is similar to those of an indirect ophthalmoscope in that the observation and illumination systems.
The observation light is focused through a series of lenses through a doughnut shaped aperture, which then passes through a central aperture to form an annulus, before passing through the camera objective lens and through the cornea onto the retina. The light reflected from the retina passes through the un-illuminated hole formed by the illumination system. As the light paths of the two systems are independent, there are minimal reflections of the light source captured in the formed image.
The image forming rays continue towards the low powered telescopic eyepiece. When the button is pressed to take a picture, a mirror interrupts the path of the illumination system allow the light from the flash bulb to pass into the eye. Simultaneously, a mirror falls in front of the observation telescope, which redirects the light onto the capturing medium, whether it is film or a digital CCD. Because of the eye‟s tendency to accommodate while looking through a telescope, it is imperative that the exiting in parallel in order for an in focus image to be formed on the capturing medium.
Fig 1.2 An OCT (Optic Coherence Tomography)
2
The retina consists of ten semi-transparent layers that have specific functions in the process of visual perception. Fundus photography provides a bird‟s-eye view of the upper layers, the inner limiting membrane, as well as the other underlying layers. As retinal abnormalities often begin in a particular layer of the retina before encroaching into the other layers such as the formation of cotton wool spots in the nerve fiber layer, it is important to be able to appreciate depth when examining a fundus in order to provide an accurate diagnosis. However, despite recent advancements in technology and the development of stereo fundus cameras, which are able to provide three dimensional.
Images by superimposing two images, most fundus cameras in circulation are only able to provide two dimensional images of the fundus. This limitation currently prevents the technology from superseding the current gold standard which is indirect binocular ophthalmoscopy. The methodology described here, is evaluated on two publicly available databases. DRIVE and STARE. These databases have been widely used by other researchers to check the effectiveness of their methods.
The 40 color fundus retinal images consist in DRIVE database were captured with a Canon CR5 non-mydriatic 3CCD camera with a 45 degree field-of-view (FOV). Each image represented in 8 bits per color plane, captured at 768 x 584 pixels, and were saved in JPEG format. STARE database consists of 81 retinal color fundus images taken with a TopCon TRV-50 fundus camera at 35 degree FOV. The database is available in PPM format at 700 x 605 pixels, in 8 bits per color channel form.
DR, also known as diabetic eye disease, is when damage occurs to the retina due to diabetes. It can eventually lead to blindness. It is an ocular manifestation of diabetes, a
3
systemic disease, which affects up to 80 percent of all patients who have had diabetes for 20 years or more. Despite these intimidating statistics, research indicates that at least 90% of these new cases could be reduced if there were proper and vigilant treatment and monitoring of the eyes. The longer a person has diabetes, the higher chances of developing diabetic retinopathy. Each year in the United States, diabetic retinopathy accounts for 12% of all new cases of blindness. It is also the leading cause of blindness for people aged between 20 to 64 years .Diabetic retinopathy often has no early warning signs. Even macular edema, which can cause sudden vision loss, may not have any warning signs for some time. In general, however, a person with macular edema is likely to have blurred vision.
These spots are often followed within a few days or weeks by a much greater leakage of blood, which blurs the vision. In extreme cases, a person may only be able to tell light from dark in that eye. It may take the blood anywhere from a few days to months or even years to clear from the inside of the eye, and in some cases the blood will not clear. These types of large hemorrhages tend to happen more than once, often during sleep.
Diabetic retinopathy results damage to the small blood vessels and neurons of the retina. The earliest changes detected in the retina in diabetes leading to diabetic retinopathy include a narrowing of the retinal arteries associated with reduced retinal blood flow ,dysfunction of the neurons of the inner neurons, followed in later stages by changes in the function of the outer retina, associated with changes in visual function, dysfunction of the blood-retinal barrier, which protects the retina from many substances in the blood (including toxins and immune cells), leading to the leaking of blood constituents into the retinal neutrophil. Later, the basement membrane of the retinal blood vessels thickens, capillaries degenerate and lose cells, particularly pericytes and vascular smooth muscle
4
cells leading
to
loss
of
blood
flow
and
progressive ischemia,
and
microscopic aneurysms which appear as balloon-like structures jutting out from the capillary walls, which recruit inflammatory cells and advanced dysfunction and degeneration of the neurons and glial cells of the retina.
Fig 1.3 Normal vision
Fig 1.4 Same view with DR
Glaucoma is a group of eye diseases which result in damage to the optic nerve and leads to vision loss. The most common type is open-angle glaucoma . closed-angle glaucoma is less common compared to open angle glaucoma and normal-tension glaucoma. Open-angle glaucoma develops slowly over time and there is no heavy pain. Side vision may begin to decrease followed by central vision resulting in blindness if not treated. Closed-angle glaucoma can present suddenly. The rapid presentation may involve severe eye pain, blurred vision, mid-dilated pupil, redness of the eye, and nausea. Vision loss from glaucoma, once it has occurred, is permanent. It can‟t cure through medicine Risk factors for glaucoma leads increased pressure in the eye, a family history of the condition, migraines, high blood pressure, and obesity. For eye pressures a value of greater than 21 mmHg is often used with higher pressures leading to a greater risk. However, some may have high eye pressure for years and never develop damage. Conversely, optic nerve damage may occur with normal pressure, known as normal-tension glaucoma. The
5
mechanism of open-angle glaucoma is believed to be slow exit of aqueous tumour through the trabecular meshwork while in closed-angle glaucoma the iris blocks the trabecular meshwork. Diagnosis is by a dilated eye examination. Often the optic nerve shows an abnormal amount of cupping.
If treated early it is possible to slow or stop the
progression
of
disease
with
medication, laser treatment, or surgery. A number of different classes of glaucoma medication are available. Laser treatments may be effective in both open-angle and closedangle glaucoma. A number of types of glaucoma surgeries may be used in people who do not respond sufficiently to other measures. Treatment of closed-angle glaucoma is a medical emergency.
Robust screening methods increases the accessibility of eye care providers with timely intervention of to prevent the vision loss caused by diabetic retinopathy. Recently a study by International Diabetes Federation found diabetes will see an epidemic growth closing to 552 million people by 2030. Besides this complications arising from diabetes are also growing including DR, which is the root cause of blindness within the 20–74 age group in most of the developed countries and presently affect 2–4% of diabetic people. The most common signs of DR are red lesions(micro aneurysms, haemorrhages) and bright lesions (exudates, drusen and cotton wool spots).
The presence of red lesions and hard exudates (bright lesions) are indicative of early stage DR. Micro aneurysms (MAs) are focal dilatations of retinal capillaries and appear as red dots in retinal fundus images. Fig 1.5 shows retinal main regions and related pathologies.
6
Fig1.5 Retinal main regions and DR related pathologies
Light entering the eye through the pupil is focused on the retina. The retina is a multilayered sensory tissue that lines the back of the eye. It contains millions of photoreceptors for capturing light rays and convert them into electrical impulses. These impulses travel along the optic nerve to the brain where they are turned into images. Optic disk is brighter than other part of the retina and is normally circular in shape. It is also the entry and exit point for nerves entering and leaving the retina to and from the brain. Retina represents the start of the optic nerve and is the entry point of the major blood vessels to the eye. Macula is the dark region devoid of vessels with fovea at its center is responsible for central and high resolution vision. The blood vasculature is a tree like structure having a high frequency component exhibited more clearly at high contrast spanning across the fundus image. The retinal blood vessels are the most important component of the retina as they provide the blood supply to the retina and also transmit the information signals from retina to the brain .The diseased eye can be detected by examining the changes in the retinal vasculature. The different types of diseases can be detected by examining the retinal structure such as diabetic retinopathy, cardiovascular disease,
7
hypertension and stroke. The retinal diseases, if not detected and diagnosed at early stage can further lead a person to a vision loss. The complete vision loss of a person can be prevented by diagnosing these diseases at initial stage. Further, for detection of these disorders segmentation of blood vessels is necessary.
Diabetic Retinopathy (DR) is a sight threatening disease which causes blindness among the working age people. It involves the identification of abnormalities like cotton wool spots, hard exudates, hemorrhages, red lesions and soft exudates .Hence, for detection of diabetic symptoms, segmentation of blood vessels is important as few morphological changes like diameter, branching angle, length or tortuosity occur in retinal blood vessels. The symptom of diabetic retinopathy which can be noticed in retinal images is the variation in width of retinal vessels
Bright lesions or intra retinal lipid exudates results from the breakdown of blood retinal barrier. Excluded fluid rich in lipids and proteins leave the parenchyma, leads to retinal edema and exudation. Lastly, wherever capillary walls are weak inside the retina, dot hemorrhages lesions are found which are slightly larger than MAs. On rupturing it will cause intra-retinal hemorrhages. Progression of DR also causes macular edema, neo-vascularization and in later stages, retinal detachment. All these abnormalities are with main retinal structures highlighted. Systematic screening by eye care specialists of diabetic patients is a cost-effective health care practice that can diagnose the pathology at the initial stage. In order to accommodate the screening and annual reviews requisite of a large number of patients, an automated screening tool is useful adjunct in diabetes clinics. At present, there are several methods which can accurately diagnose specific DR related lesions. Fig 1.6 shows retinal images with lesions.
8
Fig.1.6 Retinal images with lesions The most important internal components of eye is called retina, which covers all posterior compartment. Any damage in retina leads to severe diseases. Disorders in retina resulted from special diseases are diagnosed by special images which are obtained by using optic imaging called fundus image. They are captured by using special devices called ophthalmoscopes. Each pixel in the fundus image consists of three values namely Red, Green and Blue, each value being quantized to 256 levels. The blood vessels are the important parts of the retinal images consisting of arteries and arterioles. Checking the obtained changes in retinal images can help the physician to diagnose the diseases.
Retinal hemorrhages is a disorder of the eye in which bleeding occurs into the light sensitive tissue on the back wall of the eye. A retinal hemorrhage can be caused by hypertension, retinal vein occlusion , or diabetes mellitus which causes small fragile blood vessels to form, which are easily damaged). Retinal hemorrhages can also occur due to shaking, particularly in young infants (shaken baby syndrome) or from severe blows to the head. Retinal hemorrhages that take place outside the macula can go undetected for many years, and may sometimes only be picked up when the eye is examined in detail
9
by ophthalmoscopy, fundus photography, or a dilated fundus exam. However, some retinal hemorrhages can cause severe impairment of vision. They may occur in connection with posterior vitreous detachment or retinal detachment.
Applications of retinal images are diagnosing the progress of some cardiovascular diseases, diagnosing the region with no blood vessels (Macula), using such images in helping automatic laser surgery on eye, and using such images in biometric applications, etc. Diabetic retinopathy (DR) is damage to the retina caused by complications of diabetes mellitus, which can eventually lead to blindness. There may exist different kinds of abnormal lesions caused by diabetic retinopathy, the most frequent being dark lesions such as micro aneurysms, haemorrhages and bright lesions such as hard and soft exudates.
The severity depends on the years that the specific patient has experienced the disease, and in worst case, can eventually lead to blindness. The early stage is further classified as mild NPDR (Non-Proliferative diabetic retinopathy) and moderate to severe NPDR. In mild NPDR, signs such as micro aneurysms dot and blot haemorrhage‟s and hard or intraretinal exudates are seen in the retinal images. Micro aneurysms are small, round and dark red dots with sharp margins and are often temporal to macula. Their size ranges from 20 to 200 microns i.e., less than 1/12th the diameter of an average optic disc and are first detectable signs of retinopathy. In today‟s world diabetes has become a very common disease and so is diabetic retinopathy, an eye disorder caused by abnormalities in the retina due to inadequate amount of insulin in the body. Diabetic Retinopathy is affecting more than 80% of all diabetic
10
patients since past a decade and May also result in an unprecedented number of persons becoming blind. Diabetic Retinopathy is identified by the exudates formation in the retina. The stipulated method followed by ophthalmologists is the legitimate supervision of the retina. Protein and cellular debris which has escaped from blood vessels and had been deposited in tissues or on tissue surfaces of an eye. As it grows, DR can subsequently decrease visual acuity. The systematic screening method includes diluting the pupil of an eye with chemical solution in order to identify the exudates manually. These are the visible sign of DR and a major cause of visual loss in Non-Proliferative forms of DR. Objective of this project is lesion detection using segmented structure of retina. Presence of Lesion in retina is an early sign of DR
Retinal images play a major role in the ocular fundus operations and detection of diabetes in early stages (by comparing the states of retinal blood vessels and optic disc). The present work developed a system to identify patients with proliferate diabetic retinopathy (PDR) from the retina. The different diabetic retinopathy diseases that are of interest include red spots, micro aneurysm, neovascularisation and exudates are fall between BDR and PDR stages of the disease.
To detect the PDR, blood vessels of the fundus image are removed by curvelet transform and morphological erosion and dilation process. Then optics disc in the image are removed by using circular fitting method and blood vessels are separated by using canny edge detection. Finally remaining portion of image are referred to as exudates region.
11
Retinal hemorrhages are signs of retinal disease or injury, observable as dark patches indicative of bleeding. On a retinal image, hemorrhages appear as dark, reddish structures of various sizes dot hemorrhages are small isolated red spots, blot hemorrhages are irregular, dotted textured, localized circular structures. Hemorrhages are a clinical sign of diseases like diabetic retinopathy, and hypertensive retinopathy. Clinical guidelines specify that the location and extent of hemorrhages is a direct indicator of the severity of disease. Detection and segmentation of hemorrhages is therefore an important component for computer-assisted screening and grading.
Detection and quantitative measurement of variations in the retinal blood vessels can help diagnose several diseases including diabetic retinopathy. Intrinsic characteristics of abnormal retinal images make blood vessel detection difficult. The major problem with traditional vessel segmentation algorithms is producing false positive vessels in the presence of diabetic retinopathy lesions. To overcome this problem, a novel scheme for extracting retinal blood vessels based on morphological component analysis (MCA) algorithm is presented inthis paper.
MCA was developed based on sparse representation of signals. This algorithm assumes that each signal is a linear combination of several morphologically distinct components. In the proposed method, the MCA algorithm with appropriate transforms is adopted to separate vessels and lesions from each other. Afterwards, the Morlet Wavelet Transform is applied to enhance the retinal vessels. The final vessel map is obtained by adaptive thresholding. The performance of the proposed method is measured on the publicly available DRIVE and STARE datasets and compared with several state-of-the-art methods.
12
An accuracy of 0.9523 and 0.9590 has been respectively achieved on the DRIVE and STARE datasets, which are not only greater than most methods, but are also superior to the second human observer‟s performance. The results show that the proposed method can achieve improved detection in abnormal retinal images and decrease false positive vessels in pathological regions compared to other methods. Also, the robustness of the method in the presence of noise is shown via experimental result
Retinal vessel segmentation and quantitative measurement of vessel variations is crucial in many research efforts related to vascular features. Analysis of vascular structures could be used to diagnose several diseases such as diabetic retinopathy. glaucoma and hypertension. In many clinical investigations, segmentation of retinal blood vessel becomes a pre requisite for the analysis of vessel parameters such as tortuosity and vessel width. Manual segmentation of blood vessels is a time consuming task that requires remarkable skills. Therefore, the development of algorithms for automatic vessel segmentation and vessel diameter estimation is of paramount importance.
Retinal vessel segmentation is still a challenging issue that has been widely studied in the literature. These studies can be classified into six categories: (1) Matched filtering, (2) multi-scale algorithms, (3) pattern recognition methods, (4) vessel tracking, (5) model-based techniques and (6) mathematical morphology .In matched filtering methods; retinal images are filtered by various vessel-like kernels which are designed to model a specific feature in the image at different positions and orientations. The presence of the desired feature is recognized using the matched filter response and this method based on a verification-based multi-threshold probing scheme.
13
In above method, the image was probed with different thresholds and a vessel map was obtained by combining images derived from probed thresholds followed by post-processing algorithms. The method was evaluated on the DRIVE dataset, reporting an average accuracy of 0.92 and an area of 0.94 under the ROC curve.
In a 2D Gaussian matched filter was used to enhance the retinal images and simplified pulse coupled neural network (PCNN) was employed to segment the blood vessels by firing neighbourhood neurons. Then, a 2D Otsu thresholding was used to search for the best segmentation results. The final vessel map was obtained through the analysis of regional connectivity. The evaluation of the methodology yielded a true positive rate of 0.80 and a false positive rate of 0.02 on the STARE dataset.
Diabetic retinopathy (DR), the major cause of poor vision, is an eye disease that is associated with longstanding diabetes. If the disease is detected in its early stages, treatment can slow down the progression of DR. However, this is not an easy task, as DR often has no early warning signs. Earliest signs of DR are damages of the blood vessels and then formation of lesions.
Lesions such as exudates are normally detected and graded manually by clinicians in time consuming and it is susceptible to observer error. Diabetic retinopathy results from the leakage of small vessels in the retina correlated to a prolonged period of hyper glycaemia. In the early stages of the disease, known as non-proliferative retinopathy, there may be hemorrhages due to bleeding of the capillaries or exudates resulting from protein deposits in the retina. There is usually no vision loss unless there is a build-up of fluid in the center of the eye. As the disease progresses, new abnormal vessels grow in the retina, known as
14
revascularization. This stage of the disease is called proliferative retinopathy and may cause severe visual problems.
15
CHAPTER 2 LITERATURE REVIEW Retinal image analysis is increasingly prominent as a nonintrusive diagnosis method in modern ophthalmology. Since the morphology of the blood vessel and the optic disk is an important indicator for diseases like diabetic retinopathy, glaucoma, and hypertension. Presence of lesion in retina is an early sign of DR Anasagaret.al (2014)[3] In this paper method first step is the extraction of the retina vascular tree using the graph cut technique. The blood vessel information is then used to estimate the location of the optic disk. The optic disk segmentation is performed using two alternative methods. The Markov random field (MRF) image reconstruction method segments the optic disk by removing vessels from the optic disk region, and the compensation factor method segments the optic disk using the prior local intensity knowledge of the vessels. Initially apply a contrast enhancement process to the green channel image. The intensity of the image is inverted, and the illumination is equalized. The resulting image is enhanced using an adaptive histogram equalizer. The optic disk segmentation starts by defining the location of the optic disk. This process used the convergence feature of vessels into the optic disk to estimate its location. The disk area is then segmented using two different automated methods (MRF image reconstruction and compensation factor). Both methods use the convergence feature of the vessels to identify the position of the disk. The MRF method is applied to eliminate the vessel from the optic disk region.
16
Ismeetkauret.al (2010)[12] Diabetic Retinopathy is a disease which causes a menace to the eyesight. The detection of this at an early stage can aid the person from vision loss. The examination of retinal blood vessel structure can help to detect the disease, so segmentation of retinal blood vessel vasculature is important and is appreciated by the ophthalmologists. In this paper, they proposed the approach of blood vessel segmentation using computer intelligence by deploying fuzzy c-means and neutrosophic set. Further, the input image is scrutinized and the result achieved is whether the image is diseased or not. The various diseases detected in this technique are cotton wool spots, exudates and lesions with the help of region growing and neural network classification method. Huan wang et.al (2010)[13]The contrast of the image is enhanced with the help of contrast limited adaptive histogram equalization (CLAHE) algorithm which is applied to increase the contrast as a result of which the blood vessels appear more distinguished from the background. The features of the image are more enhanced after the magnitude of the gradient is calculated which helps further in easier segmentation of the blood vessels. Diabetic-related eye diseases are the most common cause of blindness in the world.Theypropose a novel approach that combines brightness adjustment procedure with statistical classification method. Objects in an image usually can be described in terms of some features f1, f2, …, fk such as colour, size, shape, texture and other more complex characteristics. These features, f1, f2, …, fk, form a k-dimensional feature space, F. Ideally, they would like to find a space F such that different objects map to different, non-intersecting clusters in this feature space Though the minimum distance discriminant function approach seems to work well for images under the same illumination conditions these lesions would be wrongly classified as “background” instead Of “lesion” by MDD classifier. On the basis of color information, the presence of
17
lesions can be preliminarily detected by using MDD classifier based on statistical pattern recognition techniques. Gayathri et.al(2014)[2]Automatic extraction of dark lesions from retinal images can assist in early diagnosis and screening of a common disease such as Diabetic Retinopathy. A robust and computationally efficient approach for the localization of the different features and lesions in a fundus retinal image is presented in this paper. Proposed method consists of preprocessing, contrast enhancement, blood vessels extraction and dark lesions detection stages. In the pre-processing stage, since the green channel from the coloured retinal images has the highest contrast between the sub-bands, so the green component is selected. The Green channel shows well contrasted arteries and veins with a clear dark fovea in the centre. Adaptive Histogram used for contrast the image features. Curvelet transforms is developed to overcome the limitation of wavelet and Gabor transforms. Although, wavelets are widely used in feature extraction but it fails to handle randomly oriented edges of the object and the singularities of the object. Gabor filters overcome the limitation of wavelet transform and deal with the oriented edges, but it loses the spectral information of the image. Curvelet transform is used to overcome these problems of the wavelet and Gabor filters. It can obtain the complete spectral information of the image and handle with the different orientations of the image edges.
Shradha thripathiet.al(2014)[5]Digital fundus imaging in ophthalmology plays an important role in medical diagnosis of several pathologies like hypertension, cardiovascular disease and diabetes. Retinal vessel segmentation is a primary step towards automated analysis of the retina for anomaly and also image registration. Automated assessment of the retinal vasculature morphology can be used in screening tool for early detection of diabetic
18
retinopathy pre-processing operation. Image quality is of central importance to the success of retinal examination. Hence noise suppression by means of digital image processing should improve image quality and diagnostic potential of fundus image enhancement. The histogram of an image represents the relative frequency of occurrence of various gray levels in the image.
UsmanAkramet.al(2016)[9]The early detection and diagnosis of diabetic retinopathy is important to protect the patient‟s vision. The accurate detection of micro aneurysms (MAs) is a critical step for early detection of diabetic retinopathy because they appear as the first sign of disease. In this paper, they propose a three-stage system for early detection of MAs using filter banks. In the first stage, the system extracts all possible candidate regions for MAs present in retinal image. In order to classify a candidate region as MA or non- MA, the system formulates a feature vector for each region depending upon certain properties, i.e. shape, color, intensity and statistics. they present a hybrid classifier which combines the Gaussian mixture model (GMM), support vector machine (SVM) and an extension of multimodel mediod based modeling approach in an ensemble to improve the accuracy of classification. Given the features representation of candidate regions, now describe the proposed hybrid classifier for the detection of MAs.
S Rathinam et.al(2013)[8] This Glaucoma is one of the major causes of blindness. Glaucoma is a group of conditions, in which high pressure inside the eye damages the optic nerve of the eye. Glaucoma usually affects both the eyes. It commonly occurs in adults above 40 years of age, but can even occur in newborn babies. The vision lost due to glaucoma is irreversible and cannot be regained. Hence it is very important to detect this disease as early as possible and treat early to preserve vision. Pre-processing of eye fundus image is a crucial
19
initial step before further analysis is performed. Many pre-processing techniques are available in the literature. In this paper, the performance of five pre-processing techniques are compared namely Contrast adjustment, Adaptive Histogram equalization, Median filtering, Average filtering and Holomorphic filtering. The performance of these techniques is evaluated using Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR).The Median filtering and Average filtering give suitable results and Median filter is found to be better with high PSNR and low MSE values
Mega Nathanet.al(2013)[1]Diabetes has become a new global challenge. If not diagnosed and treated in time, diabetes can encourage other illnesses in the body of patients. One such illness is related to the retina of human eyes that affects the retina and retinal structure in certain ways. The screening for detection of such abnormalities in the retina is called Diabetic Retinopathy (DR). Latest technological advances in the image processing helps auto detection of diabetic retinopathy based on the analysis of feature extractions. This analysis not only helps diagnose the disease but also helps detecting the severity of the disease.Filtering is used to remove the noise which gets added into the fundus image. Here median filtering is quite useful as it is very robust and has the capability to filter any outliers. It is also the preferred choice for removal of salt and pepper noise. Median filtering effectively suppresses isolated noise without fading sharp edges. It replaces a pixel by the median of all pixels in the neighbourhood of small sliding window.
Garimaguptha et.al(2014)[11] Segmentation of hemorrhages helps in improving the efficiency of computer assisted image analysis of diseases like diabetic retinopathy and hypertensive retinopathy.Region growing method used for segmentation. Identify the centroids of the connected set of pixels with same intensity, surrounded by pixels having
20
greater intensity as local minima. The local minima with the intensity lying in the darkest 10% range of the contrast-normalized image are used as initialization points for region growing process. They use an iterative intensity-based region growing where at each iteration the region adds neighbour pixels whose intensity is within a certain contrast limit from the current region mean. For restricting the growth of vessel seeds into branches and bends, we set stopping criteria based on the displacement of the centre of mass of the currently growing region.
Adalarsanan et.al(2016)[7]They present learning focus is developing the extraction of normal and isolated characteristics or marks in colour retinal images. The adaptive filters are tuned to match the lump (part) of vessel to be extracted in green channel images. To classify the pixels into vessels and non-vessels the Biogeography Based Optimization Algorithm is applied. Thresholding based method is used for segmentation.
Adithya kusuma wardanaet.al(2014)[4]Optic disc is one of the main keys of the retina, because of the difficult position in determining the area because the area is very close to the blood vessels, so that the optical disk intersect with the blood vessels, the optic disc becomes difficult in the segmentation process because the area is very difficult to be detected in the OD localization of the optic disc area using thresholding.Optic disc is one of the main keys of the retina, because of the difficult position in determining the area because the area is very close to the blood vessels, so that the optical disk intersect with the blood vessels, the optic disc becomes difficult in the segmentation process because the area is very difficult to be detected in the OD localization of the optic disc area using thresholding. K-mean clustering used for segmentation
21
Meera walvekar et.al(2014)[6]Median filtering is used for noise removal and adaptive Histogram equalization is a constant enhancement technique which provides an enhanced method for modifying the dynamic range and contrast of an image by altering the image. It is finding of cumulative distribution function for a given probability density function. The small area of pixels, considered to be noise, is removed after applying morphological operations. Post the transformation, the probability density function of the output will be uniform and the image will have high contrast.The features such as blood vessels, exudates, micro-aneurysms and optic discs are extracted for further analysis. In this extraction process the morphological operations such as opening, closing, erode and dilate are used.
Adithya kusuma whardana et.al (2014) in this paper all the images were resized to 720 × 576 pixels while maintaining the original aspect ratio prior to analysis. Following this, green color plane was used in the analysis since it shows the best contrast between the background retina and the vessels. The grey levels were normalized by stretching the image contrast using CLAHE to cover the full pixel dynamic range, excluding the surrounding dark border pixels and any image labels. CLAHE limits amplifying any noise that might be present in the low contrast area of the image. Initially, the green component„s intensity is inverted. After inverting the green component„s intensity, edge detection is performed using canny method. Exudates appear as bright yellow-white deposits on the retinal layer. Their shape and size varies gradually with different stages of retinopathy. Initially extracted green channel image is converted into grayscale image and then pre-processed for uniformity. Then morphological closingoperation is carried out to remove the blood vessels
22
CHAPTER 3 LESION DETECTION USING SEGMENTATION STRUCTURE OF RETINA
Fig3.1 flow diagram
Image acquisition is the retrieval of image from a particular source. Hardware based sources are particularly used for image acquisition. A digital image is produced by one or several image sensors, which, besides various types of light-sensitive cameras, include range sensors, tomography devices, radar, ultra-sonic cameras, etc. Depending on the type of sensor, the resulting image data is an ordinary 2D image, a 3D volume, or an image sequence. The pixel values typically correspond to light intensity in one or several spectral bands (gray
23
images or colour images), but can also be related to various physical measures, such as depth, absorption or reflectance of sonic or electromagnetic waves, or nuclear magnetic resonance.
Image pre-processing improves the image data by removing unwanted distortions and enhances some image features. Pre-processing depends on computational cost, computational time, noise removal and also quality of the denoised image. Image pre-processing can use linear method and nonlinear method. In linear method the filter can be apply linear to all pixels without defining the image corrupted or uncorrupted. Then corrupted image is filtered by specific algorithm and uncorrupted image is retained. Nonlinear filter produce better result compared to linear filter.
Segmentation is the process of dividing an image into regions with several properties such as colour, texture, brightness, contrast and gray level. The input is a digital gray scale image. The output of the process consist of abnormalities. The use of segmentation is to give greater information than which exists in medical images. Image segmentation is the operation of partitioning an image into a collection of connected group of pixels into regions, linear shapes or 2D shapes
Feature
extraction
is
extraction
of
particular
feature
from
the
segmented
image.Classification includes a broad range of decision-theoretic approaches to the identification of image. All classification algorithms are based on the assumption that the image in question depicts one or more features and that each of these features belongs to one of several distinct and exclusive classes. The classes may be specified by an analyst (as in supervised classification) or automatically clustered (i.e. as in unsupervised classification) into sets of prototype classes, where the analyst merely specifies the number of desired
24
categories. (Classification and segmentation have closely related objectives, as the former is another form).
3.1 PREPROCESSING Gaussian filter and adaptive histogram equalizer uses for pre-processing of the image. 3.1.1 Gaussian Filter It is a linear filter which can applied to the all pixels in the image without defining which pixel is corrupted or uncorrupted. The fundus image is a color image containing three different bands; red, green and blue. It is observed that the red and blue bands do not have significant information for exudates detection and thus it is sufficient to use the green band
only I I R I G I B
(1)
Here I is the input RGB image and others represents the red, green and blue layers respectively. In the green layer, exudates appear brightest as compared to the other two layers thus the green layer is chosen for exudates detection. The optic disc also appears bright but in the green layer it appears fragmented due to the high contrast of the blood vessels. Thus the optic disc can be easily removed. The fundus image is a color image containing three different bands; red, green and blue. It is observed that the red and blue bands do not have significant information for exudates detection and thus it is sufficient to use the green band only.
25
Here * denotes convolution and g(x, y) is a Gaussian function and remove high frequency component from the image (low pass filter)Gaussian smoothing very effective for removing gaussion noise.
I S (x, y) IG (x, y) * g(x, y)
(2)
x2 y2
2 2 1 g(x, y) 2 2
3.1.2 Adaptive Histogram Equalization Adaptive histogram equalization is a technique used to improve the contrast in images. It differs from ordinary histogram equalization in the respect that the adaptive method computes several histograms each corresponding to a distinct section of the image, and uses them to redistribute the lightness values of the image. It is therefore suitable for improving the local contrast and enhancing the definitions of edges in each region of an image.
Applying adaptive histogram equalization to an image to enhance the contrast between the contrast between the back ground pixels and the information contained in the image also lead to enhancement of the noisy pixels. The noisy pixels appear with the background information, which consists of an effective adaptive histogram equalization This operation improves the robustness and the accuracy of the algorithm adaptive histogram equalization is performed to improve the contrast of the image and to correct uneven illumination
26
I Enhanced
sI (P) I (P) r . M h2
(4)
where I is the green channel of the fundus retinal colour image, p denotes a pixel, and p is the neighbourhood pixel around p. p_ ∈ R(p) is the square window neighbourhood with length h. s(d) =1 if d >0, and s(d) = 0 otherwise with d = s (I (p) − I (p_)). M = 255 value of the maximum intensity in the image. r is a parameter to control the level of enhancement. Increasing the value of r would also increase the contrast between vessel pixels
3.2 SEGMENTATION In this project fuzzy c means algorithm is used for segmentation. Fuzzy logic is a form of many-valued logic in which the truth values of variables may be any real number between 0 and 1, considered to be "fuzzy". By contrast, in Boolean logic, the truth values of variables may only be the "crisp" values 0 or 1. Fuzzy logic has been employed to handle the concept of partial truth, where the truth value may range between completely true and completely false
Cluster analysis or clustering is the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters). It is a main task of exploratory data mining, and a common technique for statistical data analysis used in many fields
27
3.2.1Fuzzy C Means Clustering
This algorithm works by assigning membership to each data point corresponding to each cluster center on the basis of distance between the cluster center and the data point. More the data is near to the cluster center more is its membership towards the particular cluster center. Clearly, summation of membership of each data point should be equal to one. After each iteration membership and cluster centers are updated according to the formula.
ij c
1
(6)
2 m1
dij / dik
k 1 m
n
ij .xi vj
j 1,2,3...c
i 1
ij
(7)
m
'vj' represents the jth cluster 'm' is the fuzziness index [1,∞]. 'c' represents the number of cluster center. 'µij' represents the membership of ith data to jth cluster center. 'dij' represents the Euclidean distance between ith data and jth cluster center.
3.2.2 Fuzzy C-Means Algorithm
Let xi be a vector of values for data point gi. 1. Initialize membership U(0) = [ uij ] for data point gi of cluster clj by random 2. At the k-th step, compute the fuzzy centroid C(k) = [ cj ] for j = 1, .., nc, where nc is the number of clusters, using
(8)
(uij ) m x n
i
c j i1n
(u i1
ij
)
m
28
where m is the fuzzy parameter and n is the number of data points. 3. Update the fuzzy membership U(k)= [ uij ], using 1
m1 1 x c i j uij 1 nc m1 1 j 1 xi c j
4. If ||u(k) – u(k-1)||
