Automated Detection of Drusens to Diagnose Age Related Macular Degeneration Using OCT Images Samina Khalid Dept. Software Engineering, Bahria University, Islamabad Pakistan Samina.csit@must.edu.pk M. Usman Akram College of E&ME NUST Rawalpindi, Pakistan usman.akram@ceme.nust.edu.k Amina Jameel Dept. Software Engineering, Bahria University, Islamabad Pakistan amina@bahria.edu.pk Tehmina Khalil Dept. Software Engineering, Bahria University, Islamabad Pakistan tehmina_khalil08@yahoo.co m Abstract— We have developed an algorithm for detection and classification of progressive retinal disease and normal subjects into their respective classes on the basis of drusen detection. This algorithm uses intensity based thresholding and poly fitting curve strategies for the purpose of drusen detection. The spectral domain OCT images dataset was used for cross validation consist on volumetric scans of dry ARMD affected and normal eyes named as 2014_BOE_Srinivasan - Modified2 dataset [1] of Duke University. This dataset consists on OCT volumetric scans: 15 patients each from normal and dry ARMD patients consist on 30 volumes. The proposed algorithm was successfully run on all normal OCT volumes and 12 out of 15 dry ARMD volumes. The proposed algorithm successfully classified 28 volumes out of 30 volumes with 92 % accuracy for all dry ARMD and Normal classes. The results indicate that proposed algorithm can be a supportive tool for early detection of dry ARMD retinal disease. Keywords—Age Related Macular Degeneration ARMD/AMD; Retinal Pigment Epithelium RPE; Optical Coherence Tomography OCT; Gaussian Kernel. I. INTRODUCTION Age-related macular degeneration (ARMD) the leading cause of worldwide blindness in the elderly age is a bilateral ocular condition that affects the central area of retina known as the macula. Although the macula comprises only four percent of retinal area, it is responsible for the majority of useful photonic vision [2]. ARMD is the main cause of the aged blindness in developed countries e.g. Australia, United Kingdom, and America. Identifying and segmenting drusen on a retinal image is central in the classification of ARMD [3]. That’s why their measurements and quantification are important. There are multiple techniques that are used to detect retinal disorders from human retina. Several algorithms have been proposed for detection and classification of ARMD from retinal images based on statistical measures from pixel based features. Lots of useful techniques have been developed for automated detection of eye diseases and retinal abnormalities using pattern recognition techniques and image processing. However almost 95% of these techniques have been developed for fundus images most of which have used pixel based features. OCT is a relatively modern technique and very little research has been done on this. OCT imaging technology has many advantageous on other techniques in detecting and diagnosing retinal diseases. One of the major advantages of OCT imaging system is that, it can provide an early detection of all retinal disorders as compared to other techniques as shown in Figure 1. OCT imaging technology shows the cross sectional region of retina in which the retinal layers, RPE and choroid can be seen. Automated or computer-assisted analysis of ARMD affected patients retina, can help eye care specialists to screen larger populations of patients. ARMD can be classifies into two types; Dry macular degeneration can be characterized by thinning of the retina and drusen as shown in Figure 2 (drusen can be seen clearly in left image as a ripple in RPE (Retinal Pigment Epithelium) layer) It results in slow, gradual progressive “dimming” of the central vision. Dry ARMD can be sub divided into further three categories e.g. early, intermediate, and advanced. Dry ARMD can change into wet ARMD at any stage and can be characterized by abnormal growth of new blood vessel under the retina called neovascularization. International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 10, October 2016 1 https://sites.google.com/site/ijcsis/ ISSN 1947-5500