An Ensemble Approach to Detect Diabetic Retinopathy using the Residual Contrast Limited Adaptable Histogram Equalization Method

If untreated, diabetic retinopathy (DR), an irreversible disease of the retina brought on by high blood sugar, can cause blindness. In this research, a unique automated method for DR detection is suggested. Fundus Images (FIs) that had been cleaned up using Residual Contrast Limited Adaptable Histogram Equalization (RCLAHE) showed the spots. The proposed approach seeks to detect retinopathy symptoms and solve the issue of low fundus image quality. RCLAHE images' fundus-related characteristics are extracted using a weighted, fine-tuned VGG-16 algorithm, and the ultimate detection rate is determined using a soft-max classifier. The suggested network can accurately identify the DR phases, according to experimental results. The proposed method's accuracy level was 95.0%, precision was 91.0%, F1-Score was 90.0%, and recall was 87.0% when evaluated on the Messidor dataset. Our suggested network gets comparable performance when compared to other models.