Automated Detection of Diabetic Retinopathy Segmented Images using ResNet50 and VGG16 Deep Learning Algorithms

Diabetic retinopathy (DR), a severe complication arising from diabetes, make a significant threat to vision due to the deterioration of retinal blood vessels. This research work proposes a comprehensive methodology for the automated detection, grading, and segmentation of DR, leveraging advanced image processing , deep learning techniques and machine learning. The study utilizes the Indian Diabetic Retinopathy Image dataset (IDRID), comprising 81 fundus images and labels, to rigorously evaluates the proposed methodology. Key steps include detailed image preprocessing, VGG16-based feature extraction, Random Forest classifier-based grading, and innovative segmentation techniques for lesion localization. The evaluation demonstrates exceptional performance, with both VGG16 and ResNet50 architectures achieving over 99% accuracy. The process of semantic segmentation enhances interpretability, supporting clinical decision-making in retinopathy diagnosis. While the results are promising, future validation on diverse datasets and careful consideration of ethical implications are essential for responsible deployment in clinical settings. The proposed methodology signifies a significant step toward precise diagnostics and improved patient outcomes in diabetic retinopathy and holds potential for broader applications in retinal disease diagnosis.