A proposed diabetic retinopathy classification algorithm with statistical inference of exudates detection

An automated image processing system has the potential to aid ophthalmologist in diagnosing eye diabetic retinopathy (DR) diseases better, by detecting changes in retina features. This paper introduces the development of detection and classification system that provides DR stage classification based on exudates quantification in digital fundus images. The system can help ophthalmologist to perform early screening on diabetes patients. The exudates detection methods consist of two steps; rough and fine exudates segmentation. Rough segmentation is performed using morphology operation and column-wise neighborhoods operation, while fine segmentation is done using morphological reconstruction. As for DR stage classification, the segmented image is translated into DR numerical index, which later classified into the respective stage based on inference study using statistical analysis. The proposed method used fundus image database from Sungai Buloh Hospital, Malaysia. Together with other suitable retinal features extraction and classification methods, this segmentation method can form the basis of a fast and easy to use diagnostic support tool for diabetic retinopathy, which will give a huge advantage in terms of improved access to mass screening people for risk or presence of diabetes.