Deep Convolutional Neural Networks for Automated Diagnosis of Diabetic Retinopathy using Retinal Fundus Images

A primary cause of blindness is DR, must be detected early to prevent vision loss. An automated DR diagnostic method is presented in this research using a DCNN trained on retinal fundus images. To extract and categorize DR characteristics, the suggested DCNN architecture consists of convolutional, pooling, and fully connected layers. Preprocessing methods such as noise reduction, contrast enhancement, and picture normalization are used for the 88,702 photos that make up the Kaggle Diabetic Retinopathy Detection dataset. Data augmentation is used to rectify class imbalances, and model performance is enhanced via the Adam optimizer. The system attains 92.5% classification accuracy, 90.8% precision, 91.3% recall, and 91.0% F1 score. Furthermore, 94.2% of patients may be distinguished as impacted from healthy by the model, which has a 95.6% sensitivity in identifying severe DR cases. These results demonstrate the model's potential for accurate, real-time DR diagnosis, facilitating early identification and treatment.