CNN and SVM Hybrid Model: A Robust Solution for Diabetic Retinopathy Classification

Diabetic Retinopathy (DR) is a common problem of diabetes mellitus, which causes lesions on the retina that affect vision. If it is not detected early, it can lead to blindness. Unfortunately, DR is not reversible, and treatment only sustains vision. Early detection and treatment of DR can significantly reduce the risk of vision loss. Unlike computer-aided diagnosis systems, ophthalmologists' manual diagnosis process of DR retina fundus images is time-, effort-, and cost-consuming and prone to misdiagnosis. Recently, deep learning has become one of the most common techniques that have achieved better performance in many areas, especially in medical image analysis and classification. Convolutional neural network models are more widely used as a deep learning method in medical image analysis, and they are highly effective. In this context, this work proposes and investigates hybrid CNNs using support vector machines and compares them with state-of-the-art CNN architectures. To select which models to use we tested 10 state-of-art CNN architectures: EfficientNetV2S, EfficientNetB0, ResNet50, DenseNet121, MobileNetV2, InceptionV3, Xception, VGG16, VGG19 and NASNetMobile. We formed the 9,815 DR dataset with images from the Indian Diabetic Retinopathy Image Dataset (IDRiD), Kaggle’s Diabetic Retinopathy dataset, and images from American Eye Hospital Hyderabad. The results showed that the hybrid CNNs using support vector machines tend to present the best results. The experimentation outcome showed that the proposed approach classifies all the classes of Diabetic Retinopathy and performs better compared to other methods with an accuracy of 90.02%.