Optimizing Diabetic Retinopathy Detection with Deep Learning and Advanced Image Preprocessing Techniques for High-Accuracy Classification

Eye disease is a big concern across the globe. Early identification of eye diseases improves survival rates via effective treatment. This work also compares different ML and DL approaches for the categorisation of eye ailments, such as diabetic retinopathy, glaucoma, and macular degeneration, from the fundus images. A comprehensive dataset of labelled fundus images is utilised for training and evaluating multiple models. This work aims to classify eye diseases by Developing ML models with an aid of the Messidor dataset taken from the UCI ML Repository. Other sophisticated ML models which were tested are DenseU-Net, DenseNet-201, CNN, ANN, KNN, ResNet where performance indicators including specificity, accuracy, sensitivity, and F1 score were used. The DenseU-Net model outperformed the others with the following metrics: accuracy (99.01%), specificity (99.06%), sensitivity (99.56%), and f1-score (99.64%). This emphasises the efficiency of DenseU-Net in identifying eye diseases and its superiority over other techniques examined in the research. This work focuses on the analysis of the role of DL in the classification of medical images and provides recommendations for choosing the best approaches for diagnosing eye diseases.