Diabetic Retinopathy Detection based on Hybrid Feature Extraction and SVM

Diabetes is a disease caused by high blood sugar levels in the body. Diabetic retinopathy (DR) is a vision-threatening disease that primarily affects people who have diabetes for many years. It is the major cause of blindness in people with diabetes. Medical work in this domain indicated that blindness could be prevented by providing proper treatment by diagnosing DR at the initial stage. The proper screening requires the training of manual graders to understand the type of DR. However, the overall cost of this screening program increases due to the complexity of this process and workload on pathologists. State of the art methods has focused on simple retinal image analysis to eliminate the patients who are not affected by this disease. Therefore, reducing the overall cost of this process by decreasing the workload of pathologists. The focus of this research work is to automatically detect the severity level of DR instead of just providing information about its presence that can further reduce the DR costs. Therefore, we designed an automated framework to extract the anatomy independent features and trained the SVM classifier to detect different DR stages. We used the Kaggle DR-data set to evaluate the performance of the proposed method. For each stage of DR, which indicates the effectiveness of the proposed technique, an average accuracy of 96.4% was achieved. Experimental results show that the proposed method can efficiently and reliably detect DR in large image data sets. The main contribution of the proposed work is to design efficient, cost-effective and fully automatic DR screening techniques.