Detection of Red lesions in diabetic retinopathy affected fundus images

Computer based automated system is one of the important diagnostic tools in medical field. Diabetic Retinopathy is an eye disorder in which red lesions due to blood leakages can be spotted on retinal surface. This disease is commonly observed in long term diabetic patients. Ignorance to this disease can result into permanent blindness. Early stage signs of diabetic retinopathy are called as Red lesions viz. microaneurysms and hemorrhages. This paper presents a unique methodology for automatic detection of red lesions in fundus images. Proposed methodology employs modified approach to matched filtering for extraction of retinal vasculature and detection of candidate lesions. Features of all candidate lesions are extracted and are used to train Support Vector Machine classifier. In turn support Vector Machine classifies input image object into lesion or non-lesion category. The method is tested on 89 fundus images from DIARETDB1 database. The proposed algorithm gives performance as sensitivity 96.42%, specificity 100% and accuracy 96.62%.