Automated Diagnostic System for Grading of Diabetic Retinopathy Stages from Fundus Images Using Texture Features

Computational methodologies and medical imaging are become an important part of real time applications. These techniques transform medicine by providing effective health care diagnosis in all major disease areas. This will allow the clinicians to understand life-saving information using less invasive techniques. Diabetes is a rapidly increasing worldwide disease that occurs when the body is unable to metabolize glucose. It increases the risk of a range of eye diseases, but the main cause of blindness associated with diabetes is Diabetic retinopathy (DR). A new feature based automated technique for diagnosis and grading of normal, Nonproliferative diabetic retinopathy (NPDR) and Proliferative diabetic retinopathy (PDR) is proposed in this paper. This method involves preprocessing of retinal images, detection of lesions, extraction of blood vessels and extraction of texture features such as local binary pattern, Laws texture energy and Fractal Dimension. These features were used for classification of DR stages by means of supervised classifiers namely Support vector machine (SVM) and Extreme Learning Machine (ELM). In this work, in addition to morphological features, statistically significant texture features were also used for classification. It was found that the average classification accuracy of 98.88%, sensitivity and specificity of 100% respectively achieved using ELM classifier with texture features. The results were validated by comparing with expert ophthalmologists. This proposed automated diagnostic system reduces the work of professionals during mass screening of DR stages