Attention-Driven Hybrid U-Net Framework for Explainable Retinal Disease Diagnosis Using OCT Images

The prevention of vision loss requires the diagnosis of retinal disease with the help of Optical Coherence Tomography (OCT). This paper presents an Attention-Based Hybrid U-Net that can be used to analyze retinal diseases accurately and in ways that can be explained. The model combines U-Net based segmentation with an attention mechanism to boost the extraction of subtle pathological information. Thereafter, a CNN classifier recognizes four retinal types Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), DRUSEN, and Normal based on the features of the attention-enhanced segmentation. The proposed system was trained on 84,495 OCT images with preprocessing and augmentation to achieve a 97.8 % accuracy, which was better than the baseline deep learning models. Findings show that attention addition will enhance the feature localization and classification consistency, and this provides a reliable instrument to assist ophthalmologists in early diagnosing.