Integrating IDNET and Residual Blocks for Efficient Diabetic Retinopathy Grading

This project introduces a novel approach for diagnosing Diabetic Retinopathy (DR) using deep learning models, specifically the Inception-Deconvolutional (IDNet) and Residual networks (ResNet). DR, a significant ocular condition linked to Diabetes mellitus, necessitates early detection for effective intervention. Manual diagnosis is resource-intensive, prompting the exploration of automated methods. The proposed model integrates fundamental data augmentation techniques to address data imbalances and enhance model resilience. By optimizing image processing and incorporating residual blocks, the model aims to ensure robust and efficient diagnosis. Additionally, preprocessing techniques improve the quality of fundus images. The evaluation metrics for this model include Precision (0.924), Recall (0.955), F1-score (0.931), and ROC AUC (0.94), highlighting its potential to reduce healthcare burdens and improve patient outcomes by enabling quicker and more accurate detection and management of this sight-threatening condition.