中国循证儿科杂志2025，Vol.20Issue(3)：217-222,6.DOI:10.3969/j.issn.1673-5501.2025.03.010
DeepSeek赋能的儿科全流程智慧医疗系统的构建和应用效果评价
A DeepSeek-enabled intelligent pediatric healthcare system:Construction and application effectiveness evaluation
摘要
Abstract
Background The Children's Hospital of Fudan University developed an intelligent diagnostic assistance system by integrating its self-developed diagnostic tool with locally deployed DeepSeek large language models to enhance pediatric medical service efficiency and physician-patient satisfaction.Objective To evaluate the application effectiveness of the constructed intelligent healthcare system based on DeepSeek large language models in real-world pediatric patient care scenarios across the complete medical service workflow.Design Cross-sectional survey.Methods Integrating medical knowledge bases,knowledge graphs,and retrieval-augmented generation technologies,the team upgraded the original"Xiaobu AI Doctor,"creating a comprehensive intelligent medical system covering pre-consultation,consultation,and post-consultation workflows.After clinically implementation,performance metrics were automatically collected via the hospital's big data platform.Usability was evaluated through surveys of 50 randomly selected outpatient physicians and 200 patients' families.Main outcome measures Overall system performance and evaluation metrics for the pre-consultation,consultation,and post-consultation phases.Results During clinical implementation(March 3 to May 11,2025),the system served 11,957 pediatric patient visits with 86,533 cumulative interactions.Core performance showed 5％peak CPU utilization,5.9-second inference latency,and 81.5％medical reasoning accuracy.Phase-specific results included:82.3％triage recommendation utilization(pre-consultation);92.4％diagnostic accuracy and 96.4％information extraction accuracy(consultation);74.0％follow-up compliance(post-consultation).In comparative evaluation,DS-Xiaobu Doctor 2.0 outperformed GPT-4 Med(OpenAI)and BioMedLM(Stanford University)across key metrics:accuracy(92.4％),BLEU-4(0.87),ROUGE-L(0.73),and CEM(0.92).Physicians(n=50)reported high usability in system quality,information quality,interface quality,and overall satisfaction via the Post-Study System Usability Questionnaire(PSSUQ).Patient families(n=200)gave a Net Promoter Score(NPS)of+78.Conclusion DS-Xiaobu Doctor 2.0 delivered high-precision intelligent pediatric medical services across the complete care workflow with high user satisfaction,providing an effective technological solution for addressing shortages in pediatric medical resources.关键词
DeepSeek/医学大语言模型/智慧医疗/儿科Key words
DeepSeek/Medical large language model/Intelligent healthcare/Pediatric引用本文复制引用
张晓波,祁媛媛,张玉蓉,安海龙,王艺,李倩,冯瑞,杨睿,叶成杰,王新,葛小玲,史雨,王立波,傅唯佳..DeepSeek赋能的儿科全流程智慧医疗系统的构建和应用效果评价[J].中国循证儿科杂志,2025,20(3):217-222,6.基金项目
上海申康发展中心市级医院新兴前沿联合攻关项目:SHDC12024136 （）
上海市促进产业高质量发展专项资金产业战略关键领域技术攻关人工智能专题项目:2024-GZL-RGZN-01013 （）
复旦大学附属儿科医院"医+X"交叉创新团队孵化项目:EKYX202409 （）
