International Journal of Computer Engineering and Applications, Volume X, Issue IX, Sep. 16 www.ijcea.com ISSN 2321-3469

OPTIC DISC AND OPTIC CUP SEGMENTATION FOR GLAUCOMA DETECTION Pranchal Chaudhari 1, Asst. Prof. Rupali Nikhare 2 1

Department of Information Technology Department of Computer Engineering Pillai Institute of Technology , New Panvel, Navi Mumbai, India

2

ABSTRACT: Glaucoma is an eye disease that leads to vision loss. The symptoms of the glaucoma occurs when disease is quite advance. Glaucoma is called ‘silent thief of sight’ so, that early detection of glaucoma is very essential. In existing approach three methods are used to detect glaucoma namely Assessment of raised intraocular pressure (IOP).Assessment of abnormal visual field. Assessment of damaged optic nerve head. These method are not sensitive enough to be an effective screening tool because glaucoma can be occurs with or without intraocular pressure. A functional test through vision loss requires special equipments which is available only in territory hospitals and therefore unsuitable for screening, so assessment of optic nerve head would be beneficial. Proposed approach based on optic disc and optic cup segmentation for glaucoma screening. In optic disc segmentation, histograms and center surrounding statistics (CSS) are used to classify each superpixel which is either disc or non-disc. For optic cup segmentation in addition to histogram and CSS, the location information is also included into the feature space. After obtaining the optic disc and optic cup, cup-to-disc ratio is calculated. Glaucoma risk can be evaluated by identifying stages of glaucoma.

Keywords: Optic Disc, Optic Cup, Nero Retinal Rim, CDR.

[1] INTRODUCTION Glaucoma is an eye disease. According to world health organization glaucoma is second leading reason of blindness. A major apprehension with glaucoma detection is that the disease has no particular set of physical sign or symptoms that doctors can identify to detect the disease in an early stage. The main focus in glaucoma diagnosis is to detect changes in the visual operation of the eye at early stages of the disease so that vision can be protected and preserved through medical treatment. In existing approach, there are three clinical methods to detect the glaucoma are raised intraocular pressure, evaluation of abnormal visual field and assessment of damaged optic nerve head. These methods are not sensitive enough to be an efficient screening tool, because glaucoma can be there in the eye with or without intraocular pressure. Practical test during vision loss requires special equipments which are available only in territory hospitals therefore these methods for screening.

Pranchal Chaudhari and Rupali Nikhare

1

OPTIC DISC AND OPTIC CUP SEGMENTATION FOR GLAUCOMA DETECTION

Figure: 1. Structure of Normal and Glaucoma Effected Eye

Glaucoma is incurable, but its development can be slowed down through appropriate treatment. Hence detection of glaucoma in time is necessary to avoid criticality. However, many glaucoma patients are unaware of the disease until it has reached its higher stage. Subjective examination of the disease is usually time consuming. Thus, the information obtained may not be reliable. Optic nerve assessment by specialist is subjective and the availability of Heidelberg Retinal Tomography and Ocular Computing Tomography equipment is limited due to the high cost involved. Therefore, an automatic and economic system is highly desirable for detection of glaucoma in large-scale screening programs. Manual analysis of eye images is fairly time consuming and the accuracy of parameters measurements varies between experts. Hence, there arises the need for an automated technique. Automatic analysis of retina images is becoming an important screening tool now days. In early stage of glaucoma, Patients do not usually have any visual signs or symptoms. As the disease progress it causes of losing the vision and the patients may suffer from tunnel vision. Therefore early detection of this disease is essential to prevent the permanent blindness.

2. PROBLEM STATEMENT The idea is to classify the image either normal eye image or glaucoma affected eye image by optic disc and cup segmentation, from which CDR value is calculated and glaucoma risk can be evaluated by finding the severance of the disease. Providing the best and accurate result by using the different algorithms is the challenging task here. The proposed approach attempts to develop an algorithm which directed towards an efficient system for detection of glaucoma. This proposed system describes which is mainly based on image processing and classification technique for detection of glaucoma by comparing and measuring different parameters of funds image of glaucoma patient and normal patient.

3. METHODOLOGY Optic disc cupping is one of the most important risk factors in the diagnosis of glaucoma. It is defined as the ratio of the vertical cup diameter over the vertical disc diameter. The optic disc (OD), also known as the optic nerve head, is the location where the optic nerve connects to the retina. Quantitative analysis of the optic disc cupping can be used to evaluate the progression of glaucoma. More optic nerve fibers die; the optic cup becomes larger with respect to the OD which corresponds to an increased CDR value. For a normal subject, the CDR value is typically around 0.2 to 0.3. If the CDR value is 0.3 or less, then the optic nerve is relatively healthy. System architecture of proposed system is as shown in figure below which consist of above mentioned phases, pre-processing is done by converting input retinal color image into gray scale image, median filter is applied on it to reduce noise. After segmenting Optic disc and optic cup value of CDR is determined. Then value of CDR is compared with the threshold value if it is

2

International Journal of Computer Engineering and Applications, Volume X, Issue IX, Sep. 16 www.ijcea.com ISSN 2321-3469

greater than threshold, then image is glaucoma affected otherwise healthy. Then Glaucoma risk is evaluated.

Figure: 2. System Architecture for glaucoma detection

3.1 Input for system Single retinal image is given as input to the system, to check weather given input image is glaucoma affected or normal retinal image. Retinal image is basically contains the optic disc, optic

cup and neuroretinal rim which we have to segmented for the glaucoma detection. 3.2 Preprocessing

Pranchal Chaudhari and Rupali Nikhare

3

OPTIC DISC AND OPTIC CUP SEGMENTATION FOR GLAUCOMA DETECTION

Input retinal image is preprocessed by converting input image into gray level and after applying filter to that image. First, we need to enhance the image that we are going to use, for that apply basic filter techniques that is median filter to emphasize certain features or remove other features. The median filter is a non-linear filter type and which is used to reduce the effect of noise without blurring the sharp edge. The operation of the median filter is – first arrange the pixel values in either the ascending or descending order and then compute the median value of the neighborhood pixels.

Figure: 3. (a) Input Retinal Image

Figure: 3. (b) Grayscale Image

Smoothening is done through filtering technique, the median filter which is a non-linear filter type is used to reduce the effect of noise without blurring the sharp edge. The operation of the median filter is first arrange the pixel values in either the ascending or descending order and then computes the median value of the neighborhood pixels.

Figure: 4. (a) Image with noise

Figure: 4.(b) Result of median filter

3.3 Optic Disc Segmentation In the proposed method, superpixel classification is used for disc boundary initialization and deformable model is used to fine tune the disc boundary, optic disc segmentation include a superpixel generation step is used to divide the image into superpixel, feature extraction step to compute features from each superpixel as disc or non-disc superpixel to estimate the boundary, deformation step is used to fine tune the disc boundary. After that we will get the optic disc diameter. Pre-Processed Image Circle detection Using Circular Hough Transform Superpixel Generation Using SLIC Feature Extraction Classification Optic Disc Diameter Figure: 5. Flow diagram for optic disc segmentation

4

International Journal of Computer Engineering and Applications, Volume X, Issue IX, Sep. 16 www.ijcea.com ISSN 2321-3469

3.3.1 Circle detection using Hough Transforms Pre-processed image is taken as input, in which initial circle is detected using circular hough transform. The Hough transform has been used to identify circles and other parameterized geometrical shapes. The points lying on the circle: (x-a)2+(y-b)2 =C2…………..(1) are represented by a single point in the three-dimensional (3D) parameter space (a,b,c) with an accumulator of the form A(a,b,c), which is also known as the Hough space. Here, (a, b) is the center and c is the radius of the circle. The procedure to detect circles involves the following steps: 1. Obtain a binary edge map of the image. 2. Set values for a and b. 3. Solve for the value of c that satisfies above equation. 4. Increment the accumulator that corresponds to (a, b, c). 5. Update values for a and b within the range of interest and go back to step 3. 3.3.2 Superpixel generation using SLIC algorithm Superpixel algorithms makes group of pixels into perceptually meaningful atomic regions, which can be used to change the rigid structure of the pixel grid. They capture image severance, provide a suitable primitive from which to compute image features, and greatly decrease the difficulty of successive image processing tasks. A superpixel generation step to divide the image into superpixel, in this research work Simple linear iterative clustering algorithm (SLIC) is used to combined nearby pixels into superpixels retinal fundus images. Compared with other superpixel methods, SLIC is fast, memory proficient and has excellent edge adherence. Simple linear iterative clustering (SLIC) is an adaptation of k-means for superpixel generation.

Figure: 6. (a) Input Retinal Image Figure: 6.(b) Superpixel Image

3.4 Feature Extraction Feature extraction is the method of generating features to be used in the selection and classification tasks. Feature extraction consists of simplifying the quantity of resources required to describe a huge set of data accurately. Image features can refer to: Global properties of an image: average gray level, shape of intensity, histogram etc. Local properties of an image: refers to some local features as image. Feature extraction is a general term for methods of constructing combinations of the variables to get around these problems while still describing the data with sufficient accuracy. 3.5 Classification

Pranchal Chaudhari and Rupali Nikhare

5

OPTIC DISC AND OPTIC CUP SEGMENTATION FOR GLAUCOMA DETECTION

In the training part, reference images are stored for comparing the features with the test image given as an input to the program by the user. In the training, we randomly obtain the same number of superpixels from the disc and non-disc region from a set of training images with manual disc boundary. One challenge to find a good classifier is that samples from the non-disc area are often from different clusters with unbalanced numbers. One typical example is PPA. 3.5.1 Classification using SVM (Support Vector Machine) Support vector machine is a machine learning tool which is based on the data classification idea. Classification is performed by constructing N-dimensional hyper plane which separate the data in two categories. This classifier is used because of its generalization ability is high with small number of training samples. Steps for SVM Classifier 1. Calculate the image features form the reference images and store them in the array. 2. Then the test image is given to the program, the feature values are calculated and compared with the set of values already stored in the cells. 3. The set of value which is nearest to the set of feature values obtained from the input image is then mapped to group set.

Figure:7.(a)Disc Region Figure:7.(b)Cup Region Figure:7.(c) Rim Region

3.5.2 Classification using K-Mean clustering K-Mean algorithm is an unsupervised clustering algorithm that classifies the input data points into multiple classes based on their inherent distance from each other. The algorithm assumes that the data features from a vector space and tries to find clustering in them. Step for K-Mean clustering Classifier 1. Number of cluster k is taken as four. Lower value of k leads to an increase in the cup size. Higher value results in the predominance of blood vessels. An incorrect value of k gives a sub optimal result. 2.Initialize cluster centers µ1…µk. Choose k data points and set cluster centers to these points and make them as initial centroids.The data points are grouped into k clusters such that similar pixels are grouped together in the same cluster. 3. For each data point, nearest centroid is found and the pixel is assigned to the cluster associated with the nearest centroid. Centroid is the mean of the pixel in the cluster. 4. Update the centroid of each cluster based on the pixels in that cluster. The new centroid will be the mean of all pixels in the cluster.

6

International Journal of Computer Engineering and Applications, Volume X, Issue IX, Sep. 16 www.ijcea.com ISSN 2321-3469

Fig 8(a) K-Mean

Fig 9(c) Cup Region

Fig 8(b) Disc Region

Fig 9(d) Rim Region

3.5.3Classification using KNN Classifier K-nearest neighbor algorithm is a method for classifying objects based on closest training examples in the feature space. K-nearest neighbor algorithm is among the simplest of all machine learning algorithms. Training process for this algorithm only consists of storing feature vectors and labels of the training images. In the classification process, the unlabelled query point is simply assigned to the label of its k nearest neighbors. Typically the object is classified based on the labels of its k nearest neighbors by majority vote. If k=1, the object is simply classified as the class of the object nearest to it. When there are only two classes, k must be a odd integer. However, there can still be ties when k is an odd integer when performing multiclass classification. After we convert each image to a vector of fixed-length with real numbers, we used the most common distance function for KNN which is Euclidean distance.

Figure: 10. (a) Disc Region Figure: 10. (b) Cup Region Figure: 10. (c) Rim Region

3.4 Optic Cup Segmentation The LIBSVM (library support vector machine) with linear kernel is used again to randomly obtain the same number of superpixels from the cup and non-cup regions in the training step from a set of training images with manual cup boundary. Similarly, the output values from the LIBSVM decision function are used. The output value for each superpixel is used as the decision values for all pixels in the superpixel. A mean filter is applied on the decision values to compute smoothed decision values. Then the smoothed decision values are used to obtain the binary decisions for all pixels. The largest connected object is obtained and its boundary is used as the raw estimation. 3.5 Cup to Disc Ratio

Pranchal Chaudhari and Rupali Nikhare

7

OPTIC DISC AND OPTIC CUP SEGMENTATION FOR GLAUCOMA DETECTION

After obtaining the disc and cup, various features can be computed. The clinical convention CDR is used.CDR is an important indicator for glaucoma screening computed as. CDR=VCD/VDD, Where, CDR=Cup-to-Disc ratio VCD=Vertical cup diameter VDD=Vertical disc diameter The computed CDR is used for glaucoma screening. When CDR is greater than a threshold, it is glaucomatous, otherwise, healthy 3.5 Glaucoma Risk Evaluation From optic disc segmentation and optic cup segmentation we will get the value of vertical disc diameter (VCD) and vertical cup diameter (VDD).Then we are going to find the stages of glaucoma disease by setting the threshold. The cup-to-disc ratio (CDR) expresses the proportion of the disc occupied by the cup and it is widely accepted index for the assessment of glaucoma. For normal eye it is found to be 0.3 to 0.5. As the neuro-retinal degeneration occurs the ratio increases and at the CDR value of 0.8 the vision is lost completely. If the Cup-to-Disc ratio compared with the threshold if CDR is grater then, input retinal image is glaucoma affected. To finding the severance of disease stage I CDR value is varied in 0.59 to 0.66. For stage II CDR Value is varied in range 0.67 to 0.85.If CDR is greater than 0.8 then glaucoma is at stage III. Table 1 Glaucoma Stages

Sr

CDR Value

Stage

1

0.59-0.66

Stage I

2

0.65-0.85

Stage II

3

0.9

Stage III

No.

CONCLUSION A robust and efficient system is presented. The objective of optic disc and optic cup segmentation for glaucoma detection is to find out early detection of glaucoma because as the symptoms occurs when the dieses is quite advance, it leads to loss of vision, so early detection of glaucoma is very essential. This approach presents efficient and objective method for automatically classifying digital fundus images into either normal or glaucomatous types in order to facilitate ophthalmologists, for that we are going to present superpixel classification based methods for disc and cup segmentations for glaucoma screening. For evaluation we are used retinal images from clinical database. The system has been implemented using superpixel generation technique with three different classifier are used to classify image as glaucoma affected or normal. Comparison being made between K-NN, K-Mean and SVM classifier among those SVM gives best result which has better true positive rate than other. The system mainly provides an efficient method for glaucoma detection and is aimed to be highly beneficial for any person or

8

International Journal of Computer Engineering and Applications, Volume X, Issue IX, Sep. 16 www.ijcea.com ISSN 2321-3469

ophthalmologist can help to find the severance of the system.Experimnetal result shows the proposed method is more effective and helpful to user and ophthalmologist as it reduces false positive rate.

REFERENCES [1] Screening Jun Cheng, Jiang Liu, Yanwu Xu, Fengshou Yin, Damon Wing Kee Wong, Ngan-Meng Tan, Dacheng Tao,Ching-Yu Cheng, Tin Aung, and Tien Yin Wong “Superpixel Classification Based Optic Disc and Optic Cup Segmentation for Glaucoma,” IEEE transactions on medical imaging, vol. 32, no. 6, pp.0278-0062,june 2013. [2] Joshi GD, Sivaswamy J, Krishnadas “Optic disk and cup segmentation from monocular color retinal images for glaucoma assessment,” SR. IEEE Trans. Med. Image., vol. 30, no. 6, pp. 1192– 1205,Jun. 2011. [3] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine S¨usstrunk “SLIC Superpixels Compared to State-of-the-art Superpixel Methods journal of latex class files”, vol. 6, no. 1,pp.2274-2281, december 2011. [4] Chalinee Burana-Anusorn1, Waree Kongprawechnon1, Toshiaki Kondo1, Sunisa “Image Processing Techniques for Glaucoma Detection Using the Cup-to-Disc Ratio,” International Journal of Science and Technology, Vol. 18, No. 1, pp.365-373 January-March 2013. [5] Brintha Therese, Ph.D, S.J. Grace Shoba “Detection of glaucoma based on superpixel classification and feature extraction,” IJCA (0975 – 8887) Volume 106 No. 16, pp.0975 – 8887 November 2014. [6] Madhusudan Mishra, Malaya Kumar Nath and Samarendra Dandapat “Glaucoma Detection from Color Fundus Images,” IJCCT Volume-2, Issue-VI, 2011. [7] R. Bock, J. Meier, L. G. Nyl, and G. Michelson, “Glaucoma risk index: Automated glaucoma detection from color fundus images,” Med. Image Anal., vol. 14, pp. 471–481, 2010.2007. [8] Saja Usman1, Dimple Shajahan “ a review on different glaucoma detection methods,” International Journal of Advanced Research in Engineering Technology (IJARET),ISSN 0976– 6480(Print), ISSN 0976 – 6499 (Online) Volume 5, Issue 2, February(2014),pp.95-100.

[9] R. Bock, J. Meier, L. G. Nyl, J. Hornegger, and G.Michelson, “Glaucoma risk index: Automated glaucoma detection from color fundus images,” Medical Image Analysis, vol. 14, pp. 471–481, June 2010.

Pranchal Chaudhari and Rupali Nikhare

9