Year: 2026 | Month: March | Volume: 13 | Issue: 3 | Pages: 318-326
DOI: https://doi.org/10.52403/ijrr.20260337
Early Diabetes Prediction Using Hybrid Deep Learning on Retinal Images
Putti Venkata Siva Teja1, Duba Thrisha1, Begum Faridha2, Chennu Swetha3, Garikimukku Pravallika4, Kolli Dedeepya Krishna5
1,2,3,4,5Department of Information Technology,
Dhanekula Institute of Engineering & Technology, Vijayawada, Andhra Pradesh, India.
Corresponding Author: Putti Venkata Siva Teja, Duba Thrisha
ABSTRACT
Diabetes is a chronic condition that, if left untreated, can cause major health issues. Retinal blood vessel alterations may be a symptom of diabetes in its early stages. This project proposes a hybrid deep learning model to use retinal pictures to identify diabetes. To enhance prediction performance, the system integrates CNN, RNN, and U-Net. CNN extracts significant visual features, RNN examines the connections between the extracted features to predict risk, and U-Net highlights retinal vascular architecture following preprocessing and normalization, the model was trained on a dataset of retinal images. Class weights were employed to address data imbalance during training. With an AUC score of roughly 0.98, the final validation accuracy attained is about 93%. The confusion matrix demonstrates that, with very few misclassifications, the model accurately classifies the majority of diabetic and non-diabetic instances. The suggested hybrid strategy offers a straightforward, non-invasive, and trustworthy way to use retinal pictures for early diabetes prediction. The system can assist medical practitioners in early detection and screening.
Keywords: Retinal Images, Deep Learning, Diabetic Retinopathy, U-Net, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Hybrid Model, Early Detection
[PDF Full Text]