Early Deep Detection for Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) is a diabetic condition that affects the eyes and it could lead to blurry vision or complete vision loss. Convolutional neural networks (CNNs) have been used increasingly for computer vision projects and medical image analysis. Past work has been done using deep learning models and frameworks to automatically detect diabetic retinopathy. However, such techniques used very large CNNs requiring enormous computing resources. Therefore, it is necessary to develop more computationally efficient deep learning frameworks for automated DR diagnosis. The main objective of this project is to build a reliable and computationally efficient deep learning model for the automated DR diagnosis. In this paper, a computationally efficient deep learning CNN is presented based on the DenseNet-121 neural network architecture that provides very deep CNN with lower computational resources using the concept of transfer learning. The model also detects the severity of the disease. The proposed deep learning model is trained and tested using the commonly used labeled retinal images data set and the cloud GPU provided by the community of data scientists and machine learners, Kaggle.