Diabetic Retinopathy Detection using Convolutional Neural Networks on Retinal Fundus Imaging

The importance of early and standardized detection of diabetic retinopathy as one of the major causes of blindness to diabetic patients is proposed within the work using convolutional neural network. The paper proposes deep learning methodology that is compared and proposed the best configuration of its hidden layers' configuration, activation function, optimization algorithm and input layer parameters. The work was trained and tested on a medium size of retinal fundus imaging (inside surface of the eye) images to classify the present of retinopathy with robustness and promising mean accuracy of 92% in several generations of epochs. It describes optimum configuration of the adjusted deep learning algorithm's hyperparameters in solving the adversity of the case to recognize early symptoms retinopathy. The study as well contributes to the advancement of imaging technology in the domain of ophthalmology by adopting machine learning and computer vision to support further medical justification, learning process and remote medical support.