An Automated Detection and Multi-stage classification of Diabetic Retinopathy using Convolutional Neural Networks

Vision-impairing lesions on the retina are a common consequence of diabetes mellitus known as Diabetic Retinopathy (DR). Failure to diagnose it early can result in blindness. If DR is diagnosed and treated early on, the risk of permanent vision loss can be drastically reduced. Unlike computer-aided diagnosis systems, the time, effort, and expense involved in manually diagnosing DR retina fundus images by ophthalmologists is significant. Medical image analysis and classification are two domains where deep learning has recently become widespread. Convolutional neural networks are the preferred deep learning method when it comes to evaluating medical images. In this study, a method for detecting diabetic retinopathy was presented using DiaNet Model (DNM). The Gabor filter is employed in the retinal Image Pre-processing phase for the purpose of improving the visibility of blood vessels as well as for texture analysis, object recognition, feature extraction, and image compression. In Image Augmentation stage, the dataset's input dimensions are reduced using Principal Component Analysis (PCA). The DNM Model can benefit from a reduction in the number of attributes under certain conditions. A mean classification accuracy of 90.02% was observed, which is significantly higher than state-of-the-art methods.