Automated Grading of Diabetic Retinopathy Using Stacked DR-CNN: A Deep Learning Approach

There is an immediate need for effective automated screening tools for diabetic retinopathy (DR) since it is a predominant cause of avoidable blindness. This study introduces a Stacked Diabetic Retinopathy Convolutional Neural Network (Stacked DR-CNN), which is intended to more accurately and consistently score the severity of DR. The suggested design allows for the incremental learning of retinal characteristics including microaneurysms, hemorrhages, and exudates by combining many convolutional modules placed in a stacked hierarchy. In difficult fundus pictures, preprocessing techniques like light balance, noise reduction, and contrast enhancement reinforce feature extraction even more. The model is tested across the five conventional severity categories using publicly accessible DR datasets. According to experimental results, the Stacked DR-CNN outperforms conventional single-stage $C N N$ techniques in terms of accuracy, sensitivity, and specificity. The results demonstrate that the suggested method offers a reliable and expandable automated DR grading solution, facilitating teleophthalmology and extensive screening.