Explainable FL-Based Framework for Diagnosis of Diabetic Retinopathy in Healthcare 4.0 Environment

Ensuring interpretability and transparency in AIdriven decision-making is crucial, especially in healthcare. However, the resource-intensive nature of training complex AI models can often be a significant obstacle, particularly for smaller healthcare institutions and remote clinics. This is where Federated Learning (FL) and Explainable AI (XAI) come into play. This paper proposes a novel Explainable Federated Learning (EFL) framework for diagnosing Diabetic Retinopathy in a Healthcare 4.0 environment. Our approach combines the privacy-preserving benefits of FL with the enhanced transparency of XAI techniques, addressing the challenges of data privacy, resource optimization, and interpretability. By training lightweight local models on client devices and aggregating their weights to update a global model, our framework eliminates the need for high-resource centralized training, making it accessible to a wider range of healthcare providers. Furthermore, we integrate XAI methods, such as Local Interpretable Model-Agnostic Explanations (LIME), to provide visual explanations of the model's decision-making process. This enhances trust and transparency for medical professionals and patients while offering insights into the model's learning process. Our performance evaluation demonstrates the effectiveness of this EFL approach, with local models achieving high accuracy and low loss over multiple communication rounds and the global model exhibiting robust performance metrics, including accuracy, precision, recall, and F1-score. This framework can be readily extended to other healthcare applications where interpretability and privacy are paramount.