Publication

Abstract : Glaucoma is a chronic eye disease and a leading cause of blindness, after cataract. It can be diagnosed using Cup-to-Disc ratio (CDR) which can be assessed using a few imaging modalities. Digital Fundus Photography (DFP) is a widely employed for imaging the interior surface of the eye and hence popular in glaucoma diagnosis. Computer Aided Diagnosis (CAD) can help ophthalmologists in diagnostic decision making, reducing their burden and the cost and time involved. However, automatic extraction of CDR depends on localization and segmentation of landmarks such as Optic Nerve Head (ONH) localization, optic disc (OD), optic cup (OC). The segmentation procedure is affected by many factors including blood vessel occlusion, non-uniform illumination and poor contrast. The error in segmentation will propagate to rest of the stages and reflect in poor diagnosis and the use of colour information is not fully exploited in state-of-art techniques for glaucoma diagnosis. This has motivated the need for new techniques for clinical feature extraction that reduces segmentation dependent errors. In this work, clinical feature is extracted using vector based colour derivative approach in CIE L*a*b* colour space using CIE94 and CIE2000 colour distance. A comparative analysis between the two colour metrics and the resulting classification accuracies using bagged tree classifier are compared. Publicly available benchmark databases (RIM ONE, HRF) will be used for validating the algorithm.