Attention-Enhanced Ensemble Learning for Diabetic Retinopathy Classification with Interpretability

Diabetic Retinopathy (DR) is a severe complication of diabetes that significantly impacts vision and can potentially lead to blindness. The effective diagnosis and timely treatment of DR is crucial preventing vision loss. Convolutional Neural Networks (CNNs) have emerged as a prominent tool in the field of medical image analysis for this purpose. This paper proposes a novel ensemble residual deep learning model that integrates parallel convolutional filters with shallow convolutional neural network enhanced by a channel attention mechanism for the classification of DR. The proposed model leverages the strength of residual learning and attention mechanism to improve feature extraction and classification performance. To validate the effectiveness of the model, this study conducted extensive experiment on DR data, where the model demonstrated robust diagnostic capabilities. The results were remarkable, achieving accuracy of 97.69<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, F1-score of 97.77<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, sensitivity of 97.02 %, specificity of 98.18%, precision and recall both at 97.77%. Furthermore, the model attained an area under curve of 98%, highlighting its potential to serve as an effective supplementary tool for the analysis of DR images. These findings underscore the model's potential to assist healthcare professionals in the early and accurate diagnosis of DR, ultimately contributing to better patient outcome.