Explainable Artificial Intelligence (AI) Powered Deep Learning Model for Interpretable Retinal Fundus Diagnosis using Biomedical Images

Image-based diagnosis of Retinal disease using fundus images with automated procedures is very important in early diagnosis of diseases like Diabetic Retinopathy (DR), Age-related Macular Degeneration (AMD), and Glaucoma. The majority of deep learning models, however, are not interpretable, restricting their credibility and clinical implementation. In this paper, a new Explainable Artificial Intelligence (XAI) driven deep learning model, which combines ResNet50 and Vision Transformer (ViT) layers, is proposed to achieve and diagnose retinal diseases in an accurate and interpretable way. The model uses pixel-based explanation methods like Grad-CAM, SHAP and Layer-wise Relevance Propagation (LRP) to produce clinically significant visual explanations of the predictions. The images of the eye with the EyePACS and APTOS datasets were preprocessed and segmented into areas of interest in the anatomy and trained with a hybrid loss function that is enhanced with XAI. The metrics of evaluation have shown that the proposed model has a better classification accuracy of 96.8, F1-score of 96.2 and AUC of 98.1, which is better than a number of baseline CNN and transformer architectures. Also, the model achieves a high level of compliance between attention heatmaps and expert-labeled retinal lesions with a Trust Score of 84.9%. This publication contributes to the creation of credible AI systems in ophthalmology through its ability to provide both predictive and visual interpretability, and thus be able to be deployed to real-world screening, telemedicine, and clinical decision support settings.