Hybrid Deep Learning Model with ResNet50 and SVM for Diabetic Retinopathy Classification

Diabetic eye diseases such diabetic retinopathy, cataracts, and macular edema are the primary risks to visual acuity if they are not appropriately detected. In order to guarantee appropriate action and lower the risk of visual impairment, early diagnosis is crucial. This paper introduces a novel hybrid deep learning framework for classifying diabetic eye disorders by integrating Convolutional Neural Networks (CNNs) with Support Vector Machines (SVMs). The model aims to enhance the classification by incorporating an SVM algorithm at the last layers of a deep neural network. This framework basically consists of three major steps: data preprocessing, feature extraction, and then classification. The ResNet50 architecture underwent systematic optimization and calibration to enhance detection accuracy across various stages of diabetic retinopathy. The findings indicate that the proposed method surpasses traditional techniques in achieving higher accuracy in the identification of diabetic ocular conditions, thereby facilitating improved management and treatment strategies.