AUTOMATIC EXTRACTION OF THE OPTIC DISC BOUNDARY FOR DETECTING RETINAL DISEASES Muhammad Salman Haleem, Liangxiu Han, Baihua Li and Andy Nisbet School of Mathemetics, Computing and Digital Technology Manchester Metropolitan University John Dalton Building Manchester, England, M1 5GD, United Kingdom email: muhammad.s.haleem2@stu.mmu.ac.uk, l.han, b.li, a.nisbet@mmu.ac.uk Jano van Hemert and Michael Verhoek Optos plc Queensferry House, Carnegie Campus, Enterprise Way Dunfermline, Scotland, KY11 8GR, United Kingdom email: jvanhemert, mverhoek@optos.com ABSTRACT In this paper, we propose an algorithm based on active shape model for the extraction of Optic Disc boundary. The determination of Optic Disc boundary is fundamental to the automation of retinal eye disease diagnosis because the Op- tic Disc Center is typically used as a reference point to lo- cate other retinal structures, and any structural change in Optic Disc, whether textural or geometrical, can be used to determine the occurrence of retinal diseases such as Glau- coma. The algorithm is based on determining a model for the Optic Disc boundary by learning patterns of variability from a training set of annotated Optic Discs. The model can be deformed so as to reflect the boundary of Optic Disc in any feasible shape. The algorithm provides some initial steps towards automation of the diagnostic process for reti- nal eye disease in order that more patients can be screened with consistent diagnoses. The overall accuracy of the al- gorithm was 92% on a set of 110 images. KEY WORDS Optic Disc Boundary extraction, fundus image, Automatic Feature Detection, Glaucoma, Active Shape Model 1 Introduction Early detection and treatment of retinal eye disease is criti- cal to avoid preventable vision loss. Conventionally, retinal disease identification techniques are based on costly and time consuming manual observations. However, the de- ployment of automatic detection techniques can aid the di- agnosis of eye disease in a time and cost effective manner. Consequently, we expect more patients can be screened with greater consistency in their diagnoses via automation of steps in the diagnostic process. Retinal disease gives rise to changes in anatomical structure that can be observed on retinal images. The key anatomical structure of interest for diagnosis is the Optic Disc that provides the path to trans- fer visual information to the brain via Optic Nerves [7]. Its boundary detection is fundamental to the automation of retinal eye disease diagnosis because the Optic Disc Center is typically used as a reference point to locate other reti- nal structures, and any structural change in the Optic Disc, whether textural or geometrical, can be used to determine the occurrence of retinal diseases such as Glaucoma. Glaucoma is one of the most common retinal disease and it is the second most frequent cause of blindness with 12.7% of the cases being affected by it [19]. Glaucoma leads to irreversible structural changes in the Optic Nerve Head (ONH). Clearly, early detection and subsequent treat- ment is essential if patients are to preserve their vision. The most common type of Glaucoma, i.e. Open Angle Glaucoma is usually suspected if the eye pressure i.e. the Intra-Ocular Pressure (IOP) as measured by Tonometer is IOP>21 Hg [7]. However, this method is not reliable [6] and optometrists usually refer for a second opinion before making their decision. The measurement of IOP is not possible directly from a retinal image but the diagnosis of Glaucoma is possible by using indirect measurements such as an increased Optic Cup to Disc Ratio, ONH shift and vessel shift in a retinal image. In order to determine these measurements, we need reliable and accurate techniques to segment the Optic Disc in an image. Therefore, the extraction of the Optic Disc is important in the diagnosis of Glaucoma. However, automating the process accurately and reliably is not a simple task as the Disc boundary can be partially obscured by blood vessels and furthermore, the Optic Nerve is a 3-Dimensional structure of which a fundus image only shows a 2-Dimensional view. Optic Disc segmentation is usually performed after identifying the approximate center of the Disc. The nat- ural variation in the characteristics of the Optic Disc, in- cluding the variations in pigmentation and myelination of the nerve fiber layer are significant problems for defining the Optic Disc boundary. The occlusion of the Disc’s rim by blood vessels are also significant distractors. This paper presents the use of an Active Shape Modeling (ASM) [4] algorithm in the determination of the Optic Disc bound- ary. ASM involves shape approximation of the object to be determined (in our case it is the Optic Disc) using a mathe- matical model. The model deforms to reflect the boundary shape of the Optic Disc in ways that are consistent with shapes presented in a training set of images. In this pa- Proceedings of the IASTED International Conference Computer Graphics and Imaging (CGIM 2013) February 12 - 14, 2013 Innsbruck, Austria DOI: 10.2316/P.2013.797-015 40