Purpose/Significance To systematically review the key technologies, applications, and challenges of large language models(LLM) in the electronic medical record(EMR) domain. Method/Process The core technologies including pre-training, fine-tuning, and retrieval-augmented generation are summarized; three application paradigms such as information structuring, clinical text generation, and clinical decision support are analyzed; and the core challenges such as factual consistency and knowledge timeliness are analyzed. Result/Conclusion Three core application paradigms have been established for LLM in the EMR domain, but scalable adoption remains constrained by technical and ethical limitations. Future evolution should shifts towards trustworthy reasoning, autonomous agents, and collaborative ecosystems to unlock the potential of intelligent healthcare.