Automatic Identification Of Diabetic Maculopathy Stages Economics Essay

Published:

Diabetes mellitus is a major cause of visual impairment and blindness. Almost all patients with type 1 diabetes and over 60% of patients with type 2 diabetes will develop some degree of retinopathy after having diabetes for twenty years. The damage of the normal vision, depending on the severity of damage of the macula, is due to prolong of diabetes retinopathy which leads to maculopathy. The purpose of this work is to present a computer-based intelligent system for identification of clinically significant, non-clinically significant maculopathy and normal fundus eye images. The features were extracted from the raw fundus images using morphological image processing technique and fed to the two classifiers, the feed-forward architecture artificial neural network (ANN) and probabilistic neural network (PNN), for comparison. 90 subjects consisting of three different types of maculopathy eye disease conditions have been studied in this work. Result showed that p value of less than 0.001 is obtained when subjected to ANOVA (Analysis Of Variance) test, which indicates that it is clinically significant. Our results showed that ANN has the best classification results of 96.67%. Hence, the ANN classifier performs better than PNN and achieves with a sensitivity and specificity of 96.67% and 100% respectively.

1.1. Introduction

Lady using a tablet
Lady using a tablet

Professional

Essay Writers

Lady Using Tablet

Get your grade
or your money back

using our Essay Writing Service!

Essay Writing Service

One of the greatest concern and immediate challenges to the current health care is the severe progression of diabetes. People can often have diabetes and be completely unaware as the symptoms seem harmless when they seen on their own. Thus serious complications can result from having diabetes. Many of the complications of diabetes do not show until after many years of having the disease. They usually develop silently and gradually over time, although people with diabetes are not having any signs of complications, they may still eventually develop them.

The parts of the body that can be most affected by diabetes complications are the kidneys, nerves, heart and blood vessels, gums, feet and eyes. Diabetic retinopathy is an eye disease that associated with long-standing diabetes. Retinopathy can occur with all types of diabetes and can lead to blindness if left untreated. If the condition is detected early enough for laser treatment, most of the blindness can be prevented but many patients remain undiagnosed even as their disease is causing severe retinal damage [1]. The World Health Organization (WHO) has estimated that diabetic retinopathy is responsible for 4.8% of the 37 million cases of blindness throughout the world [2].

In Singapore, the prevalence of diabetes is 9% people between the age of 18 and 69 years. This figure, 32.4%, is much higher in the 60 to 69 years age group. With such statistics, the diabetic retinopathy becomes an important health problem in Singapore. In 1992, 13,296 patients screened in a diabetic retinal photography (DRP) programme were studied and 21.8% of the patients have some form of diabetic retinopathy with half of them (10.8%) sightthreatening [3]. Thus a "screenable" disease is an important requirement for health problem.

Blindness is often a late symptom of advanced diabetic maculopathy.

Normal Human Eye

The function of our eyes is to enable us to see clearly the objects in our surroundings at variable distances and under various conditions of lights. The eye works like a camera.

The rays of light enter the eye and passes through cornea and lens, which converges them so that it gets focused at the retina and a sharp image is formed. The lens of the eye has the property to automatically adjust its power depending upon the location of the object of interest. Therefore, whenever we see from distance to near object, the lens of eye increases its curvature and thus is able to focus the image clearly onto the retina. The aperture (pupil) in the colored part (iris) of the eye is also adjustable according to the illumination of the surroundings. The retina, which is situated towards the back of the eye, acts as the film in the camera. The image is formed there and then the signal is sent from there to the brain through the optic nerve, and thus we perceive the objects around us [4]. Figure 1.1 shows the cross section of the human eye.

Macula is the sensitive area of the retina which is used for reading vision and color vision. It is located almost in the center of the retina and near the optic nerve. The optic disc, also called optic nerve head, is the brightest region on the fundus and usually round in the back of the inside of the eye where the optic nerve connects to the retina. The optic nerve takes the visual image from the retina as electrical signals and transmits it to the brain. The optic disc is placed approximately 3 ~ 4mm to the nasal side of the fovea with a diameter of about 1.5 to 2mm. Figure 1.2 shows a normal retina with blood vessels that branch from the optic nerve, cascading toward the macula. Figure 1.3 shows the macula contains the following components from center to periphery Foveola, Fovea, Parafovea, and Perifovea [5].

Lady using a tablet
Lady using a tablet

Comprehensive

Writing Services

Lady Using Tablet

Plagiarism-free
Always on Time

Marked to Standard

Order Now

Virtreous Body

Aqueous Body

Retina

Optic Nerve

Central Fovea

Cornea

Lens

Iris

Figure 1.1 Cross section of the human eye

Optic Nerve Head

Macula

Figure 1.2 Human Retina

Foveola (R1)

Fovea (R2)

Parafoveal area

(R3)

Perifoveal area

(R4)

1.5mm

1.5mm

0.5mm

0.5mm

1.55mm

Figure 1.3 Normal fundus image showing major vascular arcade

0.35mm

Retinopathy is usually occurred due to damage to the tiny blood vessels next to the retina which has some tiny microaneurysms, and tiny leaks of fluid (exudates) or bleeds (haemorrhages). The macula area may only affected by retinopathy. If it becomes more severe, the name of this condition is called diabetic maculopathy.

In diabetic maculopathy, fluid rich in fat and cholesterol leaks out of damaged vessels.  The central vision will be distorted if the fluid accumulates near the center of the retina (the macula). In worst case, it can cause permanent loss of central vision if too much fluid and cholesterol accumulates in the macula. There are four different categorize of eye disorders, namely normal, mild maculopathy, moderate maculopathy and severe maculopathy retina. Mild and moderate maculopathy are classified as Non-clinically significant maculopathy (Non-CSME) whereas severe maculopathy is classified as clinically significant maculopathy (CSME). Both non-clinically significant maculopathy (Non-CSME) and clinically significant maculopathy (CSME) are two types of maculopathy eye disease.

Non-Clinically significant maculopathy (Non-CSME)

In Non-CSME stage, exudates start to leaks out from the damaged vessels which results from diabetes. Areas of leakage develop in retina, and the retina can become boggy like a sponge. In this stage, the patient's vision is not seriously affected because the exudates locations are far away from the fovea.

Clinically significant maculopathy (CSME)

In CSME stage, most of the retinal blood vessels are damaged and the leakage area increases. This result exudates leak out and deposit very close to the fovea. Hence, this affects the very centre of the macula area and will affect the visibility.

The fundus images of normal eye, Non-CSME fundus image and CSME fundus image is shown in Figure 1.4.

Exudates

Figure 1.4 Fundus images: (a) normal (b) non-clinically significant maculopathy

(c) clinically significant maculopathy

The current methods to detect and assess diabetic maculopathy are manual, expensive, potentially inconsistent, and require highly skilled personnel to facilitate the process by obtaining a large amount of fundus images [1]. Skilled ophthalmologists do not have the time for laborious photograph examination and the trained personnel are difficult to recruit, train and retain because of the nature of the work.

Hence, there is an urgent need to have a good, automatic method based on modern digital image processing techniques that can perform faster and yield consistent results, will need less or no human intervention [1]. It can also extend the capabilities and productivity of the ophthalmologist and also provide decision support to physicians. It can also help to early detect maculopathy and identify patients for early treatment to prevent or delay visual loss.

Many investigations on the computer assisted analysis [6,7,8,9,10] have been carried out, and automatic detection of microaneurysms, exudates, optic disc, blood vessels and hemorrhages for the pathology detection was studied [9,10,11,12,13]. The feature extraction is necessary for the identification and classification of the pathologies. There are different algorithms and techniques employed to extract the features from the fundus images [10,11,12,13,14].

Osareh et al [1] has detected the exudates regions by using image colour normalization, enhancing the contrast between the objects and background, segmenting the colour retinal image. The exudates and non exudates regions were determined by neural network. The colour normalization was done by using histogram specification. Local contrast enhancement was applied to improve both the contrasting attribute of lesions and the overall colour saturation in the image. This operation was performed on the intensity channel of the image after it was converted to the Hue-Saturation-Intensity (HSI) colour space. Colour segmentation was performed based on a FCM clustering algorithm after the image has been pre-processed. Due to the similarity of optic disc colour to the yellowish exudate regions, the optic disc was segmented as fragmented exudate regions. The false detected optic disc regions were ignored. The segmented objects were classified as exudate or non-exudate regions by a neural network. In the testing stage, the result gave a sensitivity of 92% and specificity 82%.

Lady using a tablet
Lady using a tablet

This Essay is

a Student's Work

Lady Using Tablet

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Examples of our work

Sinthanayothin et al [14] has used 112 retinal images to preprocess via adaptive, local, contrast enhancement. The main features are identified as the optic disc, fovea, and blood vessels. The optic discs were located by identifying the area with the highest variation in intensity of adjacent pixels. Blood vessels were identified by means of a multilayer perceptron neural net, for which the inputs were derived from a principal component analysis (PCA) of the image and edge detection of the first component of PCA. The foveas were identified using matching correlation together with characteristics typical of a fovea.

Walter et al [15] has presented and discussed an algorithm for detection of exudates, as well as detection of optic disc which was essential for this approach. Exudates were found using their high grey level variation, and their contours were determined by means of morphological reconstruction techniques. The optic disc was detected by means of morphological filtering techniques and the watershed transformation was to find its contours. The algorithm has been tested on an image data base of 30 images and compared with the performance of a human grader. A mean sensitivity of 92.8% and a mean predictive value of 92.4% were achieved.

Walter et al [16] also introduced mathematical morphology for detection of the optic disc and the vascular tree in noisy low contrast color fundus photographs. The approximate position was first found for detection of the optic disc and then the exact contours were found by means of the watershed transformation. In order to distinguish the vessels from other blood containing features, the algorithm for vessel detection consists in contrast enhancement, application of the morphological top-hat-transform and a post-filtering step.

Gagnon et al [17] has located the optic disc through a pyramidal decomposition and its contour search technique is based on the Hausdorff distance. The global optic disc region was found by means of a multi-scale analysis (pyramidal approach) using a simple Haar-based wavelet transform. The brightest pixel that appears in a coarse resolution image (at an appropriate resolution level depending on the initial image resolution and the optic disc average dimension) was assumed to be part of the optic disc. This global optic disc localization serves as the starting point for a more accurate optic disc localization obtained from a template-based matching that uses the Hausdorff distance measure on a binary image of the most intense Canny edges. An average error of 7% on optic disc center positioning was reached with no false detection. In addition, a Dempster-Shafer confidence level was associated to the final detection that indicates the level of difficulty the detector has to identify the optic disc position and shape. Once the optic disc was detected, the macula is localized by finding the darkest pixel in the coarse resolution image following a priori geometric criteria based on the eye's anatomy (macula position and distance with respect to the OD is relatively constant).

Abraham et al [18], a simple clustering mechanism and optic disc diameter are considered to determine the candidate region and optic disc center during the preprocessing stage which having pixels with the highest 2% gray levels. Genetic algorithm explores the combinatory space of possible contours (solutions) by means of crossover and mutation, followed by the evaluation of fitness and the selection of a new set of contours. The cumulative local gradient is used as a fitness function to find the fittest contour.

Singh et al [19] has developed to localize the fovea using modified IReS approach alone, without prior knowledge and relationship of other structure in retinal image.

The detection of exudates in the macular region will be employed in our current work. The measurement of severity was carried out based on the four regions of the macular area as show in Figure 1.3. The distribution of the exudates in the macular region decides the severity of the maculopathy. The automatic identification of the severity of the diabetic maculopathy system was studied.

Jayakumari et al [20] applied Histogram equalization independently for each RGB channel followed by the colour enhancement technique. The contrast enhancement technique not only enhances the brightness of lesions but also arguments the brightness of some surrounding pixels so that these may be recognized as class lesion. There is a possibility that segmented regions contain the varying of one or more retinal lesions including hard exudates, drusen and optic disc. The automatic optic disc localization algorithm has been carried out for location and elimination of the same. The segmented regions are discriminated to locate the hard exudates using features such as 'Convex Area', 'Solidity' and 'Orientation'. The exudates were detected with 93.4% sensitivity and 80% specificity. This system cannot detect soft exudates.

The layout of this chapter is as follows: section (1.2) presents the data acquisition process, preprocessing, detection of optic disc, fovea and exudates, and discusses the two classifiers, Artifical Neural Network (ANN) and Probabilistic Neural Network (PNN), used for the classification. Section (1.3) presents and compares the performance results of the two classifiers. Discussion of the performance results is presented in section (1.4) and followed by conclusions are presented in section (1.5).

1.2. Data Acquisition and Processing

90 retinal photographs of normal, non-clinically significant and clinically significant diabetic maculopathy have been studied in this work. These patient data were provided by the Kasturba Medical Hospital, Manipal, India. The images were stored in 24-bit JPEG format with an image size of 576x720 pixels. Figure 1.4 shows the fundus images of normal, non-clinically significant and clinically significant diabetic maculopathy.

Figure 1.5 shows the block diagram of the proposed system for identification of the diabetic maculopathy stages. The fundus images are converted to either green component or grayscale for features extraction of texture analysis. Image processing techniques are used to detect the location of the optic disc, fovea and entire macular region and these are discussed in the following section. The exudates in the macular regions are identified as the features to feed into the feed-forward neural network and probabilistic neural network classifiers for classification of normal, non-CSME and CSME maculopathy images. The performances of the two classifiers are compared.

Image pre-processing and segmentation

Local contrast enhancement

Detection of optic disc, fovea and exudates in macular area

Artifical neural network and Probabilistic neural network

Normal

Non-CSME

CSME

Feature extraction - area of the exudates (R1, R2, R3 and R4)

Figure 1.5 Proposed system for identification of the diabetic maculopathy stages

1.2.1 Image Pre-processing

There is an extensive dissimilarity in the colour of the fundus taken from different patients. This dissimilarity is strongly correlated to the person's skin pigmentation and iris colour. Other reasons like intrinsic attribute of lesions, decreasing colour dispersion at the lesion periphery and lighting disparity, etc. This may result colour of the lesion of some images to be lighter than the background colour. Under these conditions, there is every possibility that these lesions may erroneously be classified as background colour. Therefore, colour normalization is necessary to be performed.

Pre-processing consists of the following steps:

The RGB image is converted to a green channel or grayscale image.

Adjust the intensity values in the image, using 'imadjust' function.

Figure 1.6 shows (a) the original image (b) the image after intensity values adjustment.

Figure 1.6 (a) Original image Figure 1.6 (b) the image after intensity values adjustment.

1.2.2 Image Segmentation

Image segmentation is a process partitioning image pixels based on image feature/s. This is to separate pixels that have different colours into different regions, group the pixels hat are spatially connected and have similar colour into different region. The selection criterion is referred as the threshold value and 'im2bw' function used this value to convert the image pixels. Figure 1.7 shows the image segmentation.

Figure 1.7 Image segmentation

1.2.3 Detection of Optic Disc

The optic disc is the exist point of retinal nerve fibers from the eye and the entrance and exist point for the retinal blood vessels. It appears with similar intensity, colour and contrast to other features on the retinal image. While blood vessels also appear with high contrast as the optic disc, the green channel of the image with morphological closing operator on the intensity channel will help to eliminate the vessels which may remain in the optic disc region. A flat, octagonal structuring element with a fixed radius of fifteen (SE - morphological structuring element) was used. Figure 1.8(a) shows the result after closing operator was applied.

A columnwise neighborhood operations was applied to set each output pixel of the image to the variance value of the input pixel's 8-by-8 sliding neighborhood, as shown in Figure 1.8(b). The resulting image was binarized by thresholding of 0.95, shown in Figure 1.8(c).The location of the maximum of the image was taken as the centre of the optic disc. Figure 1.8(d) shows the detection of the optic disc. A circular mask is then created with a radius to cover the optic disc region, shown in Figure 1.8(e).

(a)

(b)

(c)

(c) (d)

(e)

Figure 1.8 (a) Image after morphological closing, (b) columnwise neighborhood operations, (c) Image segmentation, (d) Optic disc detection, (e) Circular mask on optic disc

1.2.4 Detection of Fovea

Fovea is near to the centre of the macula which responsible for our central, sharpest vision. The centre of the fovea is usually located at a distance of approximately 2.5 times the diameter of the optic disc, from the centre of the optic disc [14]. It is the darkest part of the fundus images which some images are not obvious to human eyes due to bright lighting or being covered by lesions.

After locating the optic disc, the macula region can be determined by setting an area of restriction in the vicinity of the image centre, as determined by the optic disc centre. Two circle binary masks from the centre of the optic disc are draw and the restriction area is set as 478x216 pixels to identify as the macula region. Once the macula region is identified, the location of the minimum of this region was taken as the centre of the fovea. Figure 1.9 shows the area restriction and detection of fovea.

Fovea region

(a) (b)

Figure 1.9 (a) Restriction of area to identify macula area (b) Fundus image with center of fovea marked as circle

1.2.5 Detection of Exudates

Exudates appear as bright patterns in colour fundus images and they are well contrasted with respect to the background that surrounds them. The shape and size of the exudates vary considerably and their borders are mostly irregular [1,15] during the progress of the disease. There are other features such as optic disc and blood vessels in the images that cause difficulty to detect exudates. They have high level variation and brightness patterns as compare to the exudates.

Morphological image processing techniques are used for detection of exudates. Dilation and erosion are the two fundamental morphological operations [21,22]. Closing and opening are applied extensively for detecting the exudates. The algorithm developed uses a morphological operation to smooth the background, allowing exudates to be seen clearly. Two types of structuring elements (SE) are used. They are octagon SE used to remove the vessels from the image as discussed in section 2.3 and disk-shaped SE used to identify the exudates. Figure 1.10 shows the two types of structuring elements.

Origin SE =

0

0

1

1

1

0

0

0

R=31

1

1

1

1

0

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

0

1

1

1

1

1

0

0

0

1

1

1

0

0

(a)

SE =

Origin0

0

0

1

0

0

0

0

R=31

1

1

1

1

0

0

1

1

1

1

1

0

1

1

1

1

1

1

1

0

1

1

1

1

1

0

0

1

1

1

1

1

0

0

0

0

1

0

0

0

(b)

Figure 1.10 Structuring elements: (a) octagon (b) disk-shaped

The regions of the exudates are obtained after the removal of the optic disc, circular border and the four side of the image border. Morphological closing is then applied to the image. The dilation function is to fill the exudates while erosion function is to expand their sizes. Figure 1.11(a) and (b) shows the image after morphological closing operation and exudates area in the macular region.

(a) (b)

igure 1.11 Exudates results: (a) Morphological closing operation, (b) Exudates in macular region

Images

Normal

Non-CSME

CSME

Original

Exudates

Figure 1.12 Original images and Exudates results for the three classes

1.2.6 Features used for Classification

The purpose of feature extraction is to reduce representation set of features, which distinguish input data. In this work, the exudates in the macular region are focus to identify the diabetic maculopathy stages. The below four main features are chosen for identifying the severity in the maculopathy images. These four regions in the macular area are stated and shown in section 1, Figure 1.3.

R1 - the area of the exudate in the Foveola region;

R2 - the area of the exudate in the Fovea region;

R3 - the area of the exudate in the Parafovea region;

R4 - the area of the exudate in the Perifovea region.

1.2.7 Classifiers Used

The Artifical Neural Network (ANN) was used for classification. An Artificial Neural Network is data information processing paradigm that is inspired by biological nervous systems, for instance the brain process data information. The novel structure of the information processing system is an essential element of this paradigm. The network function is determined by the connections between elements, identified as neurons working to solve specific problems. An ANN is defined for a specific application such pattern recognition, system identification or data classification through a learning process [21, 22, 23].

There are two types of network in neural network architecture and they are Feed-forward network and Feedback network [22, 23].

i.) Feed-forward network

Feed-forward neural network is an artificial neural network where the data moves in one direction from input to output, with no cycles or loops in the network. Figure 1.13(a) is an example of the feed-forward network (single layer perception).

x1

x2

xn

y1

y2

yn

Figure 1.13(a) Feed-forward network (single layer perception)

ii.) Feedback network

A feedback network has a closed loop in the network architecture where the data is moved in both directions. The weights are controlled by the activity of the outputs. Figure 1.13(b) is an example of the feedback network (single layer perception).

x1

x2

xn

y1

y2

yn

Figure 1.13(b) Feedback network (single layer perception)

Feed-forward neural network is the most popular and widely used network in many practical applications. The data processing in feed-forward network is faster than feedback network due to the data moves only in one direction and with guarantee in system stability.

The ANN used for this project is the feed-forward network and uses supervised learning to train the neural network. Supervised learning is a technique in which the network is trained by providing it with input and matches it with a desired output. Feed-forward network often has one or more hidden layers of sigmoid neurons followed by an output layer of linear neurons [23]. The network will only be tested for accuracy when it is undergoes training with the weights being adjusted according to its learning rules. A three layers network with sigmoid activation function is chosen as the classifier for this work. The input layer has four neurons corresponding to four features that were extracted from the data. The input neurons feed the values to each of the neurons in the hidden layer which has a total of eleven neurons. The resulting values from the hidden layer are fed into the three output neurons. Figure 1.13(c) shows the three layer feed-forward neural network used for classification. The three different classes for the output are represented as shown in Table 1.1.

R1

R2

R3

R4

Normal

Non-CSME

CSME

Input Layer

(4 neurons)

Hidden Layer

(11 neurons)

Output Layer

(3 neurons) Figure 2.4(a) ANN structure for classification

Figure 1.13(c) Three layer feed-forward neural network classifier

Classes

Binary Ouput

Normal

00

Non-CSME

01

CSME

10

Table 1.1 Binary output for three classes

1.3. Results

Table 1.2 shows the ranges of four parameters used to feed as input to the classification. 'p-value' of the features is obtained. A value of less than 0.001 is obtained which indicates that it is clinically significant. [24]

A data base of 90 samples is used for training and testing the classifiers. These samples are divided into 90% training and 10% for testing. The performance measures such as sensitivity, specificity and positive predictive accuracy are used to evaluate the performance of the systems.

The terms used to measure the test performance are as follow:

- True positive (TP): abnormal subjects with positive test results and are correctly diagnosed.

- True negative (TN): normal subjects with negative results and are diagnosed correctly.

- False positive (FP): normal subjects with negative test results and diagnosed wrongly as abnormal subjects.

- False negative (FN): abnormal subjects with positive results and diagnosed wrongly as normal subjects.

The sensitivity, specificity and positive predictive accuracy are computed as below.

Sensitivity = TP / (TP + FN)

Specificity = TN / (TN + FP)

PPA = TP / (TP + FP)

ANN performance

This training consisted of 5,000 iterations. During the training phase, we have observed that the output is an analog value of 0 to 1, whereas the 'desired' output is either 0 or 1. The mean square error (MSE) was set to 0.001. This MSE is high during the early part of the training and it falls gradually in the later part of the training. Table 1.3 shows the no of training and testing data set used in ANN and it indicates the correct classification of average 96.67%. From Table 1.4, it can be seen that the sensitivity of the ANN system are evaluated 96.67%, and specificity and positive predictive accuracy are evaluated 100%.

A Graphical User Interface (GUI) is created by using the 'GUIDE' function in MATLAB to process the preprocessed images [22] and allows user to perform interactive tasks. Figure 1.14 shows the diagram of GUI interface for this project work.

Classes

R1

mean ± std dev

R2

mean ± std dev

R3

mean ± std dev

R4

mean ± std dev

p value

Normal

0.0±0.0

0.0±0.0

0.0±0.0

0.0±0.0

<0.001

Non-CSME

0.0±0.0

0.0±0.0

91.333 ±135

397.27 ±491

<0.001

CSME

22.467 ±52.5

238.90 ±216

390.47 ±464

1095.1±.579E+03

<0.001

Table 1.2 Range of input features to classification model

Classes

No. of data set used for training

No. of data set used for testing

Percentage (%) of correct classification

Normal

20

10

100

Non-CSME

20

10

93.33

CSME

20

10

96.67

Average

96.67

Table 1.3 Training and testing data set used in ANN

Classifiers

TP

TN

FP

FN

Accuracy (%)

Sensitivity (%)

Specificity (%)

Positive Predictive Accuracy (%)

ANN

58

30

0

2

96.67

96.67

100.00

100.00

Table 1.4 Results of accuracy, sensitivity, specificity and positive predictive values of ANN

Figure 1.14 Graphical User I Interface

1.4. Discussion

Hence, we can conclude that the feed-forward neural network is greatly performing better. ANN system can give preliminary diagnostics in evaluating the stage of the maculopathy and also aid physicians in clinical diagnosis.

In this work, I have studied and proposed a morphological step to automatically detect optic disc and exudates in the macular area from the images in attempting to detect the maculopathy earlier. The optic disc was detected and removed prior to the exudates detection because both appear with similar intensity.

There are misclassified from the abnormal images and might need further diagnosis. This system intends to help ophthalmologists in detecting the diabetic maculopathy faster and more easily. The results show here indicated that the automatic diagnosis of diabetic maculopathy can be very successful. It is not a final result application but this system can give a preliminary diagnosis tool for ophthalmologists in evaluating the stages of maculopathy. This system is helpful in diagnosing the non-clinically significant maculopathy. So, the patient can detect maculopathy at an early stage and hence the loss of vision can be prevented.

The accuracy and performance of the classifiers can be further improved by increasing the size of the training data. By extracting the proper features from the optical images can enhance the classification results. The software for feature extraction and the program for classification of retina images are written in MATLAB 7.4.0 which can be enhanced with future version.

1.5. Conclusion

Diabetic maculopathy is resulted from the prolonged diabetes retinopathy and is a leading cause of blindness. It occurs when the retinal blood vessels are damaged and the exudates leakage area increases, deposit very close to the fovea. Hence, an automatic system for identification of normal, Non-CSME and CSME fundus eye image is proposed. The features from the raw images are extracted using image processing techniques, and fed into the feed-forward neural network classifier for classification. We have concluded that feed-forward neural network classifier perform with an accuracy of more than 96% of correct classification, sensitivity of more than 96% and specificity of 100%. The accuracy of the system can further be improved using proper input features such as microaneurysms and Hemorrhages, and the size of the training data.