Multiclass SVM-based automated diagnosis of diabetic retinopathy

Diabetes related eye disease is growing as a major health concern worldwide. Diabetic retinopathy is an infirmity due to higher level of glucose in the retinal capillaries, resulting in cloudy vision and blindness eventually. With regular screening, pathology can be detected in the instigating stage and if intervened with in time medication could prevent further deterioration. This paper develops an automated diagnosis system to recognize retinal blood vessels, and pathologies, such as exudates and microaneurysms together with certain texture properties using image processing techniques. These anatomical and texture features are then fed into a multiclass support vector machine (SVM) for classifying it into normal, mild, moderate, severe and proliferative categories. Advantages include, it processes quickly a large collection of fundus images obtained from mass screening which lessens cost and increases efficiency for ophthalmologists. Our method was evaluated on two publicly available databases and got encouraging results with a state of the art in this area.