Analysis of retinal fundus images for glaucoma screening

Abstract

The processing of a Fundus image using different computer aided techniques are used for the extraction of its distinct features for the diagnosis and screening of the deformities related to human body. The eye's fundus is the only part of the human body where the microcirculation is be observed directly. Several medical signs that can be detected from funduscopy include hemorrhages, blood vessel abnormalities; exudates and pigmentation. The Optic Disc (OD), Optic Cup (OC), Neuro-Retinal Rim Area (NRRA), ISNT (Inferior-Superior-Nasal-Temporal) Quadrants are the main contrasting features of a Fundus image. There are some diseases which do not show any symptoms at early stages but if not detected earlier they may lead to severe consequences. One such disease is Glaucoma. If not detected, diagnosed and cured at prior stages it may lead to complete vision loss. The shape, size and orientation of the Fundus image features are altered in a pathological condition that can be observed during early stages. These changes can be detected and identified by using the image processing techniques and hence preventive measures can be taken to avoid any further damage. The methodology has been applied on a database acquired from a local physician. In this dissertation work, the size and shape of the Optic Disc and Cup are used as the major features for calculating the Cup-to-Disc Ratio which classifies the image samples as normal or suspicious for Glaucoma. The methodology proposed extracts out the Optic Disc and Cup from a color retinal Fundus image. The value of Cup-to-Disc Ratio is calculated by estimating the diameters of the extracted Disc and Cup. This ratio then screens the images for Glaucoma. The ratios obtained are compared with the gold standard values and the algorithm showed a detection rate of 80% and classification accuracy of 95%, sensitivity and specificity of 100% and 91% respectively.