Detection and Classification of Diabetic Retinopathy Using EfficientNetB0: A Lightweight Model With Transfer Learning Approach
Students & Supervisors
Student Authors
Supervisors
Abstract
Diabetic retinopathy is a gradual consequence of diabetes that destroys the retinal blood vessels and can lead to irreversible blindness if not detected early. Automated and accurate screening is therefore critical to enable timely clinical intervention. In this study, an EfficientNetB0-based deep learning model is introduced using transfer learning and fine-tuning. The model was initialized with ImageNet-pretrained weights and trained on the APTOS-2019 dataset with extensive data augmentation to enhance robustness. The proposed model achieved an accuracy of 96.0\%, with precision, recall, and F1-scores of 96.09\%, 96.0\%, and 95.98\%, respectively. Compared to widely used architectures such as MobileNetV2, RetinaRes, and Capsule Network, the EfficientNetB0-based model demonstrated superior performance. These results highlight EfficientNetB0 as a lightweight yet powerful solution for diabetic retinopathy detection, with strong potential for real-world clinical deployment.By enabling early and precise identification of DR, the proposed method has the potential to reduce preventable blindness and enhance accessibility to retinal disease screening.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 2025 International Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (3ICT)
- Date of Conference: 17/11/2025 - 17/11/2025
- Venue: Bahrain (online)
- Organizer: University of Bahrain (Hybrid)