Empowering DR Assessment: A DenseNet-CBAM Driven Approach

Diabetic Retinopathy (DR) is leading cause of blindness that requires timely and accurate diagnosis for effective treatment. This paper presents a novel DR classification method using a DenseNet-CBAM-based deep learning framework. Retinal images are first pre-processed with Gabor filters to enhance vascular and textural features, followed by data augmentation to address dataset limitations. Texture features are then extracted using the Gray Level Co-occurrence Matrix (GLCM), providing critical input for the DenseNet-CBAM model. By integrating DenseNet’s efficient feature reuse with CBAM’s attention mechanisms, the proposed model effectively captures subtle retinal variations across DR stages. Evaluated on the MIT-BIH dataset using Python, the approach achieved a maximum classification accuracy of 93%. This method enhances the reliability of automated DR detection and demonstrates strong potential for clinical integration, supporting ophthalmologists in early and accurate diagnosis.