Application and Analysis of AI Large Models in Health Science Popularization Q&A and Medical Image Interpretation

Current large language and multimodal foundation models are transforming medical question answering from search engine–based information retrieval to model-centered, dialogic consultations. While this shift enhances interactivity and generative capability, it also increases the risk of inaccurate or misleading responses. Therefore, this paper organizes the task lineage, method paradigms, and evaluation points along two lines: "Multimodal medical image analysis-Text-based health science popularization questions - Safety, ethics, and compliance governance". The paper is organized according to the "Subject domain-Task domain-Governance domain", prioritizing high-quality review articles, policy regulatory documents, and studies with comparable evaluation results to support the closed-loop argument from "Technology -Application-Risk-Governance", summarizing key risks such as hallucinations, overstepping boundaries, bias, and privacy, and forming a system-level mitigation closed-loop. At the same time, it clearly deploys the task list for future research: evidence-driven and verifiable generation, workflow integration, continuous evaluation and monitoring, and auditable governance.