A trust-based global expert system for disease diagnosis using hierarchical federated learning

Many healthcare institutions have leveraged expert systems to assist in disease diagnosis and treatment recommendations. However, regulations require protecting patient data, this hinders the ability of these institutions to collaborate on developing a robust and comprehensive expert system. To address this concern, we proposed integrating Federated Learning (FL) and the joint expert system development. Each institution trains its local model on its data, and only the updated model parameters are sent to a central server. The server aggregates these updates to enhance a global model, thus enabling the institutions to contribute to a collective knowledge base while maintaining compliance with privacy regulations. However, a significant challenge in the federated learning solution is the presence of free riders. These are clients who benefit from the shared global model without truly contributing to its development. They may withhold their expertise, opting instead to rely on the contributions of more engaged clients. To handle this challenge, we propose a novel trust-oriented, coalition-based client selection process, ensuring that only clients who actively contribute to the construction of the global expert system are allowed model updates. Our experimental results demonstrate the effectiveness of this approach. Our solution accelerates the global model's convergence and improves its accuracy, which makes federated learning more resilient and effective in sensitive domains such as healthcare, where privacy concerns and collaborative knowledge sharing must coexist.