A Fusion-Based Approach to Multiclass Retinal Disease Classification: Transfer Learning Strategy with CBAM Integration

Eye diseases pose significant global public health challenges, often resulting to irreversible blindness if untreated. Early, precise diagnosis of these disorders is critical. In this paper, we address this crucial issue using transfer learning methods for the categorization of multiple eye disorders. The proposed architecture features a fusion-based technique that combines EfficientNetB0 and MobileNetV3Large convolutional neural networks (CNNs) for feature extraction and incorporates Convolutional Block Attention Module (CBAM), which dynamically focuses on significant regions of the images, enhancing the model's discriminative capability. Our proposed model achieves outstanding performance on two separate datasets: 97.85% accuracy on the Eye Disease Classification (EDC) dataset and 91.96% on the Ocular Disease Intelligent Recognition (ODIR) dataset, surpassing existing state-of-the-art approaches in ocular disease classification. By developing the proposed model, we intend to contribute to the creation of more effective diagnostic tools and eventually enhance patient outcomes in ophthalmology.