Explainable Triple Attention Dual Scale Residual Network for Diabetic Retinopathy Detection Using Retinal Fundus and OCT Images

Diabetic retinopathy (DR), long recognized as an important cause of blindness, with timely detection essential for preventing visual loss. DR is traditionally diagnosed by manual evaluation of retinal images, which is fully, labor involvement and major error causing system. Deep learning models are promising for automation of DR detection but many are not interpretable. This paper, we proposed an Explainable Triple Attention Dual Scale Residual Network (ET-DSRN) for GLIVEK classification using retinal fundus and OCT image datasets. Triple Attention, Dual Scale and Residual Learning Combined for Enhanced Predictive Power and understandability ET-DSRN alleviates the vanishing gradient problem and improves feature extraction at varying scales. We show from experimental results that the proposed method outperforms existing methods. The clarity of the model provides trust and transparency for its clinical use.” The method is authenticating through publicly available datasets demonstrating enhanced detection accuracy. The techniques are based on Butt et al. (2023); the greatly improved the DR detection process.