摘要
Abstract
The current analysis,evaluation,and diagnosis of medical process flows in hospital buildings rely heavily on manual labor,which suffers from low efficiency,strong professional dependence,susceptibility to omissions,and difficulty in coordinating multiple specifications.This paper proposes a system based on the Dify low-code platform,integrating the Qwen/Qwen2.5-VL multimodal parsing model and the Deepseek-V3 report generation model.Multi-source data,including textual standard specifications,case documents,and image parsing results,are uniformly converted into high-dimensional vectors using the Qwen/Qwen3-Embedding-8B model.Based on the efficient matching capability of the Reranker vector retrieval model,a distributed vector database index is built to construct a medical process knowledge base and a seven-step closed-loop diagnostic workflow,thereby achieving automated diagnosis of primary-level processes.The results show that,compared with manual diagnostic methods,the intelligent assisted diagnostic system reduced the average time spent from 210.0 minutes to 14.7 minutes,and the report generation time from 30 minutes to within 2 minutes,providing a reference for the optimization of hospital building medical process design.关键词
医疗工艺一级流程/AI辅助诊断/多模态图纸解析/Dify平台/知识库构建Key words
primary medical process flow/AI-assisted diagnosis/multimodal drawing parsing/Dify platform/knowledge base construction分类
建筑与水利
