Fundus image lesion detection algorithm for diabetic retinopathy screening

An eye disease that damages the retina of diabetic patients is known as diabetic retinopathy (DR). The severity of the disease is found by different lesions such as hemorrhages, microaneurysms, exudates etc., these are the early stage symptoms of non-proliferative DR for early analysis of DR. A single framework for instinctive Lesion Detection used for diagnosis of the disease easily by screening is proposed. It consists of four steps: luminosity and contrast enhancement, removal of extracted blood vessels and optic disc (OD), lesion detection and classification based on lesions. Gamma correction and CLAHE for luminosity and contrast enhancement. Principle component analysis for vessel extraction and using convex hull transform for OD detection. After background subtraction, lesions are detected using morphological operations and classification based on count of lesions. The proposed algorithm is analyzed using the publically available datasets and evaluated using the metrics of specificity, sensitivity and accuracy.