Advancing Diabetic Retinopathy Detection: Leveraging Deep Learning for Accurate Classification and Early Diagnosis

DR is a prevalent diabetes-related complication that can result in vision impairment if not identified and addressed promptly. However, the limited availability of large-scale annotated datasets and the high inter-observer variability of manual diagnosis by ophthalmologists pose challenges for automated DR detection. To achieve the objective, the proposed method involved incorporating transfer learning into a deep learning model, which can improve the accuracy of the model while reducing the amount of training data required. The method involved three main steps: image resizing as preprocessing, feature extraction using Haralick texture features and Hu moments, and classification using various machine learning algorithms including K-NN, Naive Bayes, decision tree, and Support Vector Machine (SVM). The evaluation of the proposed method was conducted on a publicly available Kaggle dataset. The results showed that SVM outperformed the other machine learning algorithms, achieving an accuracy of 52.75%. Furthermore, a comparison was made with an Efficient-Net deep learning model, and the outcomes demonstrated that the deep learning technique achieved the highest accuracy of 91.8% and AUC of 0.91, surpassing all other techniques. This suggests that deep learning techniques, such as Efficient-Net, are well-suited for the detection of diabetic retinopathy.