An Effective Diabetic Retinopathy Detection System using Deep Belief Nets and Adaptive Learning in Cloud Environment

Abstract

The major reason behind the blindness of the diabetes patients is diabetic retinopathy. It can be characterized as an eye disease that affects the retina of eye due to diabetes mellitus. The detection of diabetic retinopathy in early stage is a challenging task to ophthalmologists. This paper presents a diabetic retinopathy detection system for accurate detection of DR in the patients. The proposed diabetic retinopathy detection system is the combination of several preprocessing technique and deep belief nets. The aim of preprocessing technique is to enhance the images, edge detection, and segmentation. Further, the deep belief nets are adopted for the accurate detection of DR. But, the parameter tuning of weight, bias and learning rate have significant impact on the performance of deep belief nets. This work also addresses these issues of deep belief nets though an adaptive learning strategy for learning rate and updated mechanism for weight and bias issues. The proposed system is implemented in cloud environment. It is utilized to store the information regarding DR and communication between doctors and patients. Further, the efficacy of the proposed diabetic retinopathy detection system is tested over an image dataset and it comprises of three thousand two hundred eye images include with diabetes retinopathy and no diabetes retinopathy. The results are evaluated using accuracy, sensitivity, specificity, F1-Score and AUC parameters. The results of proposed system are compared with KNN, SVM, ANN, InceptionV3, VGG16 and VGG19 techniques. The results showed that proposed diabetic retinopathy detection system obtains 91.28% of accuracy, 93.46% of sensitivity, 94.84 of specificity and 94.14 of F1-Score rates than other techniques using 10-cross fold validation method. Hence, it is stated that proposed system detects diabetes retinopathy more accurate than other techniques.