Diabetic Retinopathy Detection Using Convolutional Neural Networks (CNNs)

One of the leading diseases that cause blindness in people of working-age adults is diabetic retinopathy, which gives rise to the importance of early diagnosis for effective treatment. This article brings forth a deep learning approach towards automation in diabetic retinopathy detection using Convolutional Neural Networks. For robustness, in the proposed model, advanced preprocessing techniques are incorporated in image normalization and data augmentation. Following the standardized classification scheme from the publicly available Kaggle Diabetic Retinopathy Dataset (Kaggle 2015), it automatically learns spatial hierarchies in images and classifies retinal images according to severity level. This model has shown very competitive experimental results, surpassed traditional methodologies and provided very sound platforms for automated diagnostic systems in healthcare environments. While the CNN-based model performs quite excellently, this paper discusses some future work opportunities: augmenting the model with RNNs for temporal analysis and transfer learning through ResNet architectures. These developments may allow for increased accuracies in detecting, as well as future scalability, within DR diagnostic systems.