Accurate Insights, Trustworthy Interactions: Designing a Collaborative AI-Human Multi-Agent System with Knowledge Graph for Diagnosis Prediction

Healthcare question-answering (QA) systems can assist physicians in making medical decisions. However, traditional medical QA systems struggle with multi-agents interaction and domain-specific knowledge processing, thereby reducing the accuracy and credibility of clinical decision-making. We thus develop a multi-agent decision-making system by combining a fine-tuned medical model, biomedical knowledge graphs, and PubMed data. By summarizing the symptoms described by users, our system can automatically convene clinical experts from various fields, retrieve domain knowledge, and provide clinical decision support for users. We have validated the system performance using both technical and user-centric approaches in terms of information accuracy, user satisfaction, user trust, ect. We thus provide an effective tool for healthcare professionals to make accurate and timely decisions. Furthermore, this study also reveals new design and research opportunities, including (1) optimizing multi-agent collaboration mechanisms for more complex medical decision-making, (2) improving interaction design to enhance system transparency and explainability, and (3) expanding the system to support a broader range of medical issues and multimodal data.