Automatic Detection of Diabetic Retinopathy using Neural Networks and Support Vector Machine

Diabetic Retinopathy (DR) is the most common leading cause of preventable blindness for a large percentage of people in the world. Standard retinal cameras are too costly. Moreover, there is a significant shortage of trained retinal specialists and graders to capture the retinal images, grade and treat them. The early diagnosis and treatment of DR can aid patients in avoiding vision loss. Consequently, the development of an automated detection system is of a great value to help providing early and accurate detection. In this paper, we have proposed a computer based method that minimizes SVM drawbacks in classification by incorporating them into neural networks for the detection of diabetic retinopathy using the fundus images. Using Image pre-processing, segmentation techniques involves processing of fundus images to detect features, such as blood vessel area, micro aneurysms, and texture. The proposed techniques have been tested on the images of IDRiD database for DR. So, a total of 516 images were tested and classified into normal, abnormal and microaneurysm. The detection results were judged by some expert ophthalmologists and a classification sensitivity, specificity and accuracy of 98%, 67 %s and 96%, are respectively obtained.