Abstract Background Conversational agents are increasingly used in healthcare, but most systems are designed for clinicians and insufficiently support patient-centered decision-making. Limited patient understanding of medical decisions may reduce adherence and lead to doctor-patient conflicts. Existing dialogue systems either lack interpretability or suffer from rigid architectures and limited scalability. This study aims to develop an interpretable, patient-centered medical dialogue system to improve communication and decision support. Methods We propose an intention-oriented multi-agent dialogue system consisting of an intention understanding agent, an argument reasoning agent, and a template-based interaction agent. A medical knowledge ontology was constructed using a hybrid expert-data-driven approach, and a knowledge base was built from recent biomedical literature. The system performs intent recognition, entity extraction, and argument-based reasoning using weighted clinical evidence. A Graves’ eye disease dialogue platform was implemented to evaluate system performance through objective metrics and clinician-based user studies. Results The system achieved an average F1 score of 85% for intent recognition and over 95% response accuracy in evaluations by physicians. User experience assessments showed performance comparable to physician-led online systems and superior to traditional medical Q&A systems. In comparative experiments, the proposed system significantly outperformed ChatGPT in personalized reasoning and interpretability across multiple evaluation dimensions. Conclusions The proposed multi-agent dialogue system improves patient-centered communication by providing interpretable and personalized decision support. It demonstrates strong potential to enhance healthcare dialogue systems and assist clinical decision-making.
Liu et al. (Tue,) studied this question.