Diabetic Retinopathy Detection Using Hybrid CNN - SVM Model

Diabetic retinopathy (DR) is not promptly diagnosed and treated, and it can result in irreversible vision loss. This research work introduces a composite model for the diagnosis of DR that combines CNN (Convolutional Neural Network) and SVM (Support Vector Machine). Combining the efficiency of CNN feature extraction with the accuracy of SVM classification, the proposed model achieves state-of-the-art performance in disease diagnosis. The model is trained using a massive dataset of annotated retinal images, and its hyper-parameters are fine-tuned using cross-validation. In experimental evaluations on benchmark datasets such as the Kaggle Diabetic Retinopathy Detection Challenge dataset, the hybrid CNN-SVM model outperforms both CNNs and SVMs on their own. Accuracy, specificity, sensitivity, and the Area Under the Receiver Operating characteristic curve, which is also known as AUC-ROC, are used to evaluate the model's performance. The hybrid model outperforms pure deep learning in terms of classification accuracy and interpretability, making it a potent tool for preventive medicine.