Improving Automated Diagnosis of Diabetic Retinopathy: Exploring the Influence of Segmented Retinal Blood Vessel Images Through Deep Learning

The present investigation focuses on using original and segmented retinal images to apply deep learning models, particularly DenseNet 121 and Inception V3, for the detection of diabetic retinopathy. According to the experimental results, DenseNet 121 achieved 97.7% accuracy for original images and 95.5% accuracy for segmented images, while Inception V3 demonstrated 96.6% accuracy for original images and 92.2% accuracy for segmented images. Precision, recall, and F1 score values support the models’ strong diagnostic capabilities. DenseNet 121 consistently demonstrated excellent precision and recall in its results, while Inception V3 exhibited commendable performance. The confusion matrices underscored the models’ effectiveness in minimizing false positives and false negatives, reinforcing their reliability in clinical settings. These results imply that deep learning models show significant potential for enhancing the diagnosis of diabetic retinopathy, providing medical professionals with precise tools for early detection and management. This study contributes to the expanding body of research on the application of deep learning in medical diagnostics, holding implications for improved patient care and outcomes not only in ophthalmology but also in various other medical fields.