Classification of Diabetic Retinopathy based on Hybrid Neural Network

Diabetic retinopathy is a diabetic disorder associated with human eye and it also remains as a common cause for blindness across the globe. Diabetic retinopathy is still diagnosed using fundus images from eye doctors due to insufficiently reliable automated DR detection mechanism. However, manual screening is the weakest, complex and time consuming process. Therefore, a computer based recognition technique with a deep learning approach is proposed to automatically recognize the deiabetic retinopathy images by classifying retinal fundus images into two different types. In this regard, this paper proposes an algorithm that incorporates an improved DR detection method aimed at increasing the efficiency. The deep neural network method called inception v3 is used for DR feature extraction. By adopting the transfer learning method, different image classification functions of the Inception V3 can be customized. For testing and training, totally 48 and 90 fundus images are used. The final classification is done by using the support vector machine algorithm, where input images are classified into two types namely, diabetic retinopathy image and normal retinal image.