Diabetic Retinopathy Classification with EfficientNetB0 Model with Attention Mechanism: An Approach on OCT Images
2026 4th Cognitive Models and Artificial Intelligence Conference (AICCONF), Praha, Çek Cumhuriyeti, 24 - 25 Nisan 2026, ss.1-5, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/aicconf69182.2026.11600669
- Basıldığı Şehir: Praha
- Basıldığı Ülke: Çek Cumhuriyeti
- Sayfa Sayıları: ss.1-5
- Yozgat Bozok Üniversitesi Adresli: Evet
Özet
Diabetic retinopathy (DR) is a significant contributor to visual impairment globally, impacting more than 103 million individuals as of 2020. Timely identification is essential to avert visual impairment; yet, conventional diagnosis depends on the manual assessment of retinal images by experts. In this study, we propose an attention-enhanced transfer learning framework based on EfficientNetB0 for binary diabetic retinopathy classification using OCT images. To improve the model’s ability to focus on clinically relevant regions, Convolutional Block Attention Modules (CBAM) are integrated after the MBConv blocks of the backbone network. The proposed model is trained and evaluated on a publicly available OCT dataset consisting of 2,838 images categorized as diabetic retinopathy and non-diabetic retinopathy. Hyperparameters are optimized using grid search, and the model performance is assessed through accuracy, precision, recall, F1-score, and ROC-AUC metrics. Experimental results demonstrate that the proposed EfficientNetB0+CBAM architecture achieves a test accuracy of 97%, outperforming several existing approaches on the same dataset. The findings indicate that incorporating spatial and channel-wise attention mechanisms significantly enhances feature representation for OCT-based DR classification. This work highlights the effectiveness of lightweight attention-augmented models for medical image analysis and their potential to support early diagnosis in clinical decision-making systems.