Leveraging LLMs for Efficient Data Structure Standardization in Chinese Medical Examination Reports

This study investigates the potential of large language models (LLMs) in standardizing Chinese medical examination reports, which are crucial for health assessment but often suffer from non-standardized terminologies across different hospitals. We utilized a dataset of over 900,000 Chinese reports and explored the effectiveness of prompt engineering and LoRA fine-tuning techniques on various LLMs. Our results demonstrate that fine-tuned LLMs, particularly Qwen2.5-14B, achieved high accuracy in predicting standardized department and item detail names, significantly outperforming traditional methods like BERT. Notably, the model exhibits strong generalization capabilities, achieving high accuracy on unseen hospital data with an average accuracy of 98.34% in cross-validation experiments. This research highlights the potential of LLMs in medical data standardization, laying the foundation for automated processing and analysis of large-scale medical data.