Retrieval-Augmented Generation (RAG) is widely employed to mitigate risks such as hallucinations and knowledge obsolescence in medical question answering, yet its predominantly single-round, static retrieval paradigm misaligns with the multi-stage process of clinical reasoning. This compressed workflow induces two structural deficiencies: question-to-query translation often lacks clinically grounded semantic interpretation, and retrieval lacks iterative sufficiency feedback, making it difficult to form reliable evidence chains. We argue that both issues stem from a deeper cause—overloading a single reasoning chain with heterogeneous tasks of interpretation, exploration, and adjudication—and that the remedy is to reconstruct the workflow via task decoupling and dynamic multi-round exploration. To this end, we propose the Self-Evolving Multi-Agent framework SEMA-RAG, which assigns these roles to three specialist agents: Interpreter Agent for clinical schema interpretation, Explorer Agent for sufficiency-driven self-evolving retrieval, and Arbiter Agent for evidence adjudication and answer selection. Across five benchmarks and five LLM backbones, SEMA-RAG improves the strongest baseline by +6.46 accuracy points on average, measured per backbone.

Hallucination detection in clinical large vision-language models requires cross-verifying heterogeneous evidence, including patient text and images for faithfulness, and medical knowledge for factuality. However, existing pipelines either verify these dimensions in isolation or conflate them into a uniform process, potentially causing evidence misalignment that limits detection precision. Consequently, we posit that the essence of black-box hallucination detection is an evidence routing problem and propose MedF²Route, an Evidence-Routed Graph Verification Framework for both Faithfulness and Factuality hallucination detection. Specifically, it first transforms free-text clinical queries and model responses into structured graphs, partitioning the response graph into independent subgraphs to separate patient-specific grounding from knowledge-based diagnostic inference. An evidence-aware routing and verification mechanism then dynamically dispatches response triplets in these subgraphs to their dedicated heterogeneous evidence sources for targeted verification. Experiments on our MedF²Bench diagnostic benchmark show that our framework outperforms state-of-the-art baselines, notably improving detection precision. Overall, MedF²Route enables professional and precise detection of clinical hallucinations, underpinning the trustworthiness of clinical LVLMs. The code and data are available at https://github.com/Agentic-CliniAI/MedF2Route.