Deep Convolutional Neural Network-Based Diabetic Retinopathy Detection in Digital Fundus Images

摘要: Diabetic Retinopathy (DR) is a common medical disorder damaging the retinal blood vessels of diabetic patients. Regular screening fundus images and timely detection initial symptoms DR, namely microaneurysms hemorrhages, are important to reduce possibility vision impairment. The proposed work explores power Convolutional Neural Network (CNN) in analysis disorders. An automated deep learning model named Deep Network-based Detection (DCNN-DRD) has been analyze classify them as healthy or defective based on DR symptoms. A image fed into DCNN-DRD which consists five convolution pooling layers followed by dropout layer three fully connected layers. linear output data produced every represents weighted value gradient descent graph for refinement improve accuracy through several iterations. Thus, does not require any preprocessing learns high-level discriminative features from pixel intensities categorize either defective. trained with subset MESSIDOR dataset ROC dataset. Experimental results show that successfully predicts 97% accuracy.