基于置信规则库推理方法（RIMER）的院前创伤评估决策支持系统研究

Pre-hospital trauma assessment is a key step in trauma care chain. Trauma incidents always happen in complex and uncertain environment, and many factors may affect pre-hospital traumatic condition of a trauma victim. How to use computers to model trauma expert knowledge, and to reach a comprehensive and reliable trauma assessment while available clinical signs and symptoms from the injured contain uncertainties? This remains a pressing issue in research area of decision support systems (DSSs) for pre-hospital trauma assessment. In this project, we will make some extensions and improvements to current belief rule-base inference methodology using the evidential reasoning approach (RIMER). Then we will employ the improved RIMER to construct a DSS for pre-hospital trauma assessment. The RIMER-based DSS can model trauma expert knowledge containing fuzzy values or intervals, infer with inputted uncertain clinical information, and train its knowledge base via accumulated trauma data to provide evidence-based decision support. The RIMER-based DSS for pre-hospital trauma assessment can help pre-hospital doctors to comprehensively and rationally assess trauma, to send the injured to appropriate hospitals or trauma centers, and can further help doctors in emergency department to speed up their operation preparation, and can eventually result in an improved success rate of trauma treatment, a reduced mortality rate and a reduced physical disability rate.

院前创伤评估是提高创伤救治质量的关键一环。影响院前创伤伤情的因素众多，且院前创伤环境复杂、不确定因素很多，如何对模糊的或包含区间值的创伤专家知识进行计算机建模，并在创伤信息不确定的环境下，对院前创伤进行全面可靠的综合评估？这是院前创伤评估决策支持研究领域亟待解决的问题。本项目提出对现有专家系统建模方法--置信规则库推理法（RIMER）进行扩充和改进研究，并对院前创伤评估进行决策系统建模和实现。改进后的RIMER能够对包含模糊不确定性和区间不确定性的专家知识进行建模；能够针对复杂、不确定的创伤信息进行合理的伤情推理；系统知识库能够根据累积的伤员数据进行以性能指标AUC为优化目标的自动学习，提供循证决策支持。该系统将能有效辅助院前救护人员进行全面可靠的创伤评估，辅助把伤员送往合适的急救中心或医院，加强院前、院内创伤信息对接，提升院内创伤救治组配备速度，从而提高创伤的救治成功率、降低致死致残率。

影响院前创伤伤情的因素众多，且院前创伤环境复杂、不确定因素很多，如何对带有不确定性的创伤专家知识进行计算机建模，并在创伤信息不确定的环境下，对院前创伤进行全面可靠的综合评估是创伤领域的研究热点。本项目基于置信规则库推理法（RIMER）对院前创伤评估进行决策系统建模和实现。该创伤决策支持系统能够对带有不确定性的专家知识应用置信规则库进行建模；能够针对复杂、不确定的创伤信息采用证据推理方法进行合理的伤情推理；系统知识库能够根据累积的伤员数据进行以性能指标ROC曲线下面积AUC为优化目标的自动学习。本项目利用专家知识和经验进行规则库建模，并充分利用实地采集的多年创伤病人数据进行置信规则库模型的训练和验证研究。本项目成果主要包括四篇SCI英文论文、三篇中文科技核心期刊论文、一篇英文国际会议论文，以及两篇在审英文论文。本项目研究的专家知识库和推理机制已经成功嵌入到“全国严重创伤救治信息交互联动系统”中，为国内首创，填补了国内创伤救治领域智能化辅助伤情评估的空白。