Model Based Retinal Analysis for Retinopathy Detection

This paper provides a novel combination of methods to preprocess and extract features from retinal images, in order to get structures such as the vessel tree and the optic disc, and diagnose retinopathy. A model based approach was developed, making extensive use of LBG vector quantization and elements provided by mathematical morphology as investigation tools. The proposed algorithm was tested against the DRIVE database hand-labeled ground-truth, scoring an overall 93.2% correspondence on vessel tree search algorithm, 81.3% on the optic disc localization, 85% specificity and 78% correspondence on the disease detection.