The widespread dissemination of health misinformation on social media platforms poses a significant challenge to public health governance. Existing deep learning methods suffer from two limitations: (1) These models often fail to capture critical information in lengthy texts, which results in reduced detection performance, and (2) insufficient integration of domain-specific medical knowledge hinders the identification of misleading entities and illogical relationships in health misinformation claims. To address these challenges, we propose a Dual-Stage Framework Integrating Large Language Model (LLM) Summarization with Knowledge Graph Embeddings (DS-SumKG) for enhanced health misinformation detection. This framework effectively combines LLM-generated summaries with health-related knowledge graph embeddings to reduce noise and enhance semantic understanding of medical entities. Experimental results demonstrate that DS-SumKG achieves 95.37% F1-score, significantly outperforming state-of-the-art models.