Retinal Abnormality Risk Prediction Model: A Hybrid Approach Based on Vessel Characteristics and Exudates

A few systemic diseases, for example, hypertension, diabetes, and vascular disorders will first affect the retinal vessels. When affected with these diseases, the retinal vessels show some sort of vascular changes according to the severity of the conditions. So, in order to diagnose this kind of diseases an efficient system that can detect the retinal abnormalities is required. This paper presents a hybrid approach for the automatic retinal vessel classification, vascular caliber estimation, and exudate detection in retinal images. The retinal vessel classification and caliber estimation is done by exploiting both visual and geometric features that enable discrimination between vein and arteries. Exudates are identified by extracting the yellow pixel level in the retinal image. Based on these three parameters, the retinal abnormality risk prediction model predicts whether the input retinal image is normal or abnormal.