面向不完备电子病历的属性协同演化约简机理与模型研究

The incomplete electronic medical records contain a great deal of potential applying values，which attribute reduction issue appears widely in the clinical feature selection and decision support system.Therefore,it has become a kernel technology in many applying fields.There is a huge challenge for attribute reduction because a great lot of incomplete values exist.So it is essential to solve the problems about the attribute reduction of imcomplete electrionic medical records by adopting some effective theories and methods.That attribute reduction is being researched from the perspectives of the co-evolutionary theory is new development status in its domain.At present,it is still emerging throughout the world, and there are many unknown fields needed to be explored. Based on the attribute features of incomplete electronic medical records,we will carry on the research on the theory,algorithm and key technology of attribute co-evolutionary reduction,explore the quantum co-evolutionary mechanism based on the evolutionary game theory, and refine an uniform evolutionary framework of the Nash equilibrium for agents' co-evolutionary game reduction under the bounded retionality. Meanwhile we will propose a new complete set of attribute co-evolutionary reduction algorithm and model,which will make up for the lacking of traditional methods and provide the better theory foundation for the application of the clinical decision support system based on electronic medical records.It will be very meaningful for the attribute reduction of the incomplete electronic medical records, and also be valuable for the development of its practical application. It will present the important scientific evidence for the solution of minimum attribute reduction, which is an N-P hard problem.

不完备电子病历蕴藏着巨大的潜在应用价值，其属性约简问题广泛存在于临床特征选择和决策支持系统中，是众多应用的核心支撑技术。由于实际电子病历中大量不完备属性值的存在，给其属性约简带来了巨大的挑战，因此迫切需要考虑给出有效的理论和方法来解决不完备电子病历属性约简问题。本项目拟根据不完备电子病历属性特征系统地开展电子病历属性协同演化约简理论、算法和关键技术研究，探索属性约简中基于演化博弈论的量子协同演化机理，精炼出有限理性下进化主体协同博弈约简的纳什均衡统一演化框架，提出一套完整的面向不完备电子病历属性协同演化约简新算法和新模型, 以弥补以往解决不完备电子病历属性约简方法不足，为实现基于电子病历临床智能决策系统的应用奠定较好理论基础。本项目对探索不完备电子病历属性约简及拓展其相关实际应用具有重要的意义与价值，同时也为最小属性约简NP-Hard问题的求解提供重要的科学依据。

不完备电子病历属性约简问题广泛存在于临床特征选择和决策支持系统中，是医学病历应用的核心支撑技术。本项目目标是开展不完备电子病历属性协同演化约简理论、算法和关键技术研究，为实现不完备病历特征选择、规则提取和临床决策支持系统等提供较好的理论模型和求解算法支持。为达成上述目标，我们在以下四个方面开展了深入研究：1）基于量子计算和协同进化理论的属性约简算法和模型研究。重点解决了属性协同演化约简模型优化、演化自适应性以及大规模属性约简等关键问题，进一步提高了属性协同演化约简算法效率，为不完备电子病历属性约简算法设计奠定了较好基础；2) 基于演化博弈论的属性协同演化约简动态机理研究。探索了进化主体协同博弈约简的纳什均衡策略，实现参与属性协同演化约简的各子种群精英均衡优势解的平衡态，提高了电子病历属性协同演化约简算法的稳定性和鲁棒性；3）适合不完备电子病历属性特征的属性演化约简算法和模型研究。提出了用于复杂噪音环境下多模态MRI病历分割的基于量子云模型粒化的属性协同演化约简方法，设计了一种基于分布式协同云模型的属性均衡优势约简加速器和一种基于多层MapReduce协同集成Pareto均衡属性约简鲁棒算法用于孕龄胎儿脑病历曲面属性约简与特征分割，为实现基于电子病历临床智能决策系统的应用提供了较好的计算模型；4) 不完备电子病历属性约简与知识挖掘应用系统研究。设计了电子病历属性约简挖掘系统，有效提取出电子病历系统中隐含的病人临床病情相关诊断规律和知识。本项目针对不完备电子病历属性约简关键问题开展了系统研究，在不完备电子病历属性约简算法和模型方面取得了一定的进展和突破，较好地完成了本项目各项计划目标。