Comprehensive Deep Learning Framework For Diabetic Retinopathy

Diabetic Retinopathy is one of the foremost attractions to impaired vision across the world and especially in working-age adults. While still among the major reasons for distorted vision worldwide and more so among working-age adults, diabetic retinopathy, without having any early detection, is responsible for some of the most severe forms of blindness. Such fields in coastal Karnataka, India, where the art of retinal fundus imaging still lacks technical knowledge, experience, and skills, might have been able to avert such outermost disasters of patients. The mission aims closer to the development of a deep learning model to create an automated class framework that generates the DR stage, i.e. moderate, mild, severe, and proliferative retinopathy, from fundus pictures. One of the core architecture models noted to be very efficient while analyzing medical images is the CNN. Some data preprocessing techniques also enhance the performance which includes resizing, normalization, and augmentation. This classification process is performed in three phases: first with real-time data, second with the publicly available dataset on Kaggle, and then the combination of real-time as well as Kaggle images to further classify. These phases are compared in terms of performance metrics, such as accuracy, precision, and recall with cross-validation. This project has the potential to revolutionize diabetic eye care by offering speedy and accurate DR classification.