Differential Entropy in Wavelet Sub-Band for Assessment of Glaucoma Malaya Kumar Nath, Samarendra Dandapat Department of Electronics and Electrical Engineering, IIT Guwahati, Guwahati, Assam, India Received 23 April 2012; revised 18 May 2012; accepted 24 May 2012 ABSTRACT: Glaucoma is an eye disease that causes progressive optic neuropathy and vision loss due to degeneration of the optic nerves. Cup to disc ratio (CDR) is the standard measure for evaluation of glaucoma. It is difficult to estimate the value of CDR if the bounda- ries of cup and disc are not well defined. In this work, we propose a novel method based on differential entropy (DE) for evaluation of glaucoma. DE can be used as a measure of glaucoma as it is propor- tional to the probability of number of glaucoma pixels in wavelet sub- bands. It has been shown that DE or negentropy value of 0.25 as an optimum threshold for glaucoma detection. This method is evaluated and its performance is compared with three existing methods using 54 retinal images. The proposed method shows the best result with an accuracy value of 92.59%. V V C 2012 Wiley Periodicals, Inc. Int J Imag- ing Syst Technol, 22, 161–165, 2012; Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/ima.22017 Key words: glaucoma; differential entropy; cup to disc ratio; wavelet sub-band I. INTRODUCTION Glaucoma (Leite et al., 2011) is the leading cause of blindness after diabetic retinopathy (Niemeijer et al., 2010) worldwide. It occurs due to the elevated intraocular pressure (IOP) exerted by aqueous humor of the eye (Schacknow and Samples, 2010). IOP can be measured by Tonometry, Goldman Applanation Tonometry, and Tonopen. These methods of defining glaucoma do not provide accu- rate measurement always, as tissue damage does not have a direct relation with IOP. So, IOP cannot be used as a standard measure for glaucoma. It is a critical job for the doctors to detect the vision loss or glaucoma by measuring IOP (Jonas et al., 1999). Visual field is more specific indicator than IOP. Visual field test is done by perim- etry, which document the level of peripheral vision. In this case, the patient responds to a perceived flash of light by looking at it every time. The accuracy of the testing depends on patient’s patience, attention, and retinal sensitivity. Limitations in IOP measurement and visual field test are reduced by looking at the appearance of optic disc (OD). Glaucoma affects the structure of optic nerve head by reducing the neuroretinal rim. Glaucoma minimally affects the other regions except the OD. Figures 1a and 1b show the cup and disc regions in fundus images. Figures 1c and 1d show the OD with glaucoma. Cup to disc ratio (CDR) increases with the progression of the disease. CDR is used as a measure for progression of glau- coma (Mishra et al., 2011). Bock et al. (2010) used the concept of principal component analysis (PCA) and support vector machine (SVM) for glaucoma prediction. Mishra et al. (2011) used active contour-based method for finding CDR value for evaluation of glau- coma (Mishra et al., 2011). In some glaucoma cases, the cup and disc are not distinguished clearly as shown in Figures 1c and 1d. In such cases, CDR cannot be used as a measure for glaucoma. The in- formation regarding the changes in cup size is most vital for assess- ment of glaucoma. In this article, we propose a novel method which is based on evaluation of differential entropy (DE) in wavelet sub- band-5 for prediction of glaucoma. Wavelet sub-band-5 highlights the OD region by suppressing the blood vessels. Blood vessels are minimally affected by glaucoma. The rest of the article is organized as follows. Methodology is explained in Section II. Results are pre- sented in Section III, and conclusions are discussed in Section IV. II. METHODOLOGY The proposed method for glaucoma prediction is shown in Figure 2. It consists of preprocessing, wavelet decomposition, segmentation- of-cup, and measurement of information by DE. A. Preprocessing. OD appearance changes during glaucoma. Gradually, the cup size increases with the progression of glaucoma. Preprocessing is applied to the green channel of the color fundus image as green channel provides higher contrast between the features and the background. Preprocessing consists of cropping, illumination correction, and histogram equalization to provide the details of OD. Cropping provides the details about the OD. Illumi- nation correction is performed to have homogeneous background, which is obtained by subtracting the mean value of the image from the original image. Histogram equalization is applied to all the images in order to spread energy of all the pixels inside the image and then normalize them to equalize the amount of energy related to each image. Figure 3 shows the different preprocessing stages for glaucoma detection. Figure 3a is the cropped color fundus image with 128 3 128 pixels in bit-map format taken from OD organization (Mishra Correspondence to: Malaya Kumar Nath; e-mail: malaya@iitg.ernet.in ' 2012 Wiley Periodicals, Inc.