Introduction:
Diabetes, or more accurately, Diabetes Mellitus (DM), is a metabolic disease that results from elevated blood sugar levels. Diabetes eventually results in diabetic retinopathy (DR), an eye condition that significantly impairs vision. Diabetic retinopathy is one of the most serious retinal disorders that can cause blindness. Thus, obtaining a timely diagnosis of the illness is essential. This research introduces an advanced system for categorizing diabetic retinopathy (DR) by utilizing deep learning (DL) techniques, including convolutional neural networks (CNN), VGG-16, ResNet-50, and other related methodologies in deep learning. The suggested system can help ophthalmologists reach a preliminary decision by classifying patients as having no DR, mild DR, moderate DR, severe DR, or proliferative DR.
Methodology:
To determine the degree of diabetic retinopathy severity level, we have employed deep learning classification algorithms, Convolutional Neural Network (CNN), VGG-16, and ResNet50 using transfer learning. This includes the process of data collection, preprocessing steps including data augmentation and normalization, the selection of the ResNet-50 model for its deep learning capabilities, transfer learning with pre-trained weights, model training with hyperparameter optimization, validation through cross-validation techniques, and interpretability analysis for clinical insights.
Results:
Our model, which we trained using many retinal images taken using fundus photography from the APTOS 2019 Blindness Detection dataset, achieves 96.93% training accuracy and 93.59% test accuracy for the ResNet-50 model. Using the VGG-16, we achieved an accuracy of 67%.
Conclusion:
The results of this extensive study are intended to provide a substantial contribution to the field of diabetic retinopathy diagnosis by providing a reliable, scalable, and automated approach. The use of deep learning methodologies in retinal image processing has promise for transforming the initial identification and categorization of diabetic retinopathy. This would enable prompt intervention and avert irreversible visual impairment in individuals with diabetes.
