基于CBR/RBR融合模式的医疗决策代价敏感性研究

Fusion mode of CBR and RBR(FMC/R) has been using in researches on medical decision-making (MD)，and great achievements has been obtained in theory and practice field. While there are also some problems on cost sensitivity in FMC/R, such as high false negative rate (FNR) and false positive rate (FPR). Aiming at characteristics of knowledge field in fusion reasoning space of MD, in this research, time window technique and transfer entropy are introduced to modeling state equation of transfer cost based on homomorphism and conditional transferred probability. Incentives, key factors and its influence of cost sensitivity are to be revealed in fusion reasoning mechanism. Combining both clinical and simulated data, the decision data of cost losses, such as feature sets of different categories from medical cases and their diagnostic results, is to be acquired and analyzed. FMC/R will be established to improve reasoning efficiency of incomplete data and to reduce reasoning losses with determined cost matrix. Based on singular value decomposition theorem, both sensitivity criterions of decision cost and convergence properties of solution spaces in MD will be explored. On circumstances of uncertainty and imbalance, Bayesian network learning model is to be built to explore the correlation between steady-state and random-state. The effects of MD sensitivity are going to be analyzed to improve efficiency and quality of fusion reasoning. Multiple measures will be provided for reducing FNR/FPR in different conditions. Validated by empirical, achievements of this research can be practiced into MD.

基于规则/案例(CBR/RBR)融合推理模式应用于医疗决策的研究已取得初步理论与实践成果,但仍存在误判率、漏判率高的决策代价敏感性问题。本项目针对医疗决策中融合推理空间知识场特征,运用时间窗技术、传递熵原理构建基于同态性及其条件转移概率的转移代价状态方程,揭示融合推理机制中决策代价敏感性诱因、关键因素及作用。将临床和智能模拟病房结合,获取与分析不同类别病例特征集及其诊断结论等决策代价损失数据,在CBR/RBR框架下建立融合推理模型,使其既可提高不完整性数据集推理效率,又可减小代价矩阵一定时的推理损失；运用奇异值分解定理,研究决策代价敏感性判别准则及决策解空间的收敛性。在不确定性与非平衡性条件下,构建考虑稳态与随机态关联性的贝叶斯网络学习模型；分析决策代价敏感性对推理效率与品质的影响,针对不同条件提出降低误判率、漏判的措施。通过实证研究对模型进行验证,将理论与仿真结果应用于医疗决策分析中。

医疗决策系统主要是运用智能算法与推理机制模拟医生临床诊断，基于规则/案例(CBR/RBR)融合推理模式应用于医疗决策的研究已取得初步理论与实践成果，但仍存在误判率、漏判率高的决策代价敏感性问题，推理准确性和效率一直是影响医疗决策系统成败的要因，阻碍医疗决策系统普遍推广与临床应用的瓶颈，也是研究人员普遍关注的关键性科学研究课题。本项目针对医疗决策中融合推理空间知识场特征，运用时间窗技术、传递熵原理构建基于同态性及其条件转移概率的转移代价状态方程，揭示融合推理机制中决策代价敏感性诱因、关键因素及作用。将临床和智能模拟病房结合，获取与分析不同类别病例特征集及其诊断结论等决策代价损失数据，在CBR/RBR框架下建立融合推理模型，使其既可提高不完整性数据集推理效率，又可减小代价矩阵一定时的推理损失；运用奇异值分解定理,研究决策代价敏感性判别准则及决策解空间的收敛性。在不确定性与非平衡性条件下，构建考虑稳态与随机态关联性的贝叶斯网络学习模型；分析决策代价敏感性对推理效率与品质的影响，针对不同条件提出降低误判率、漏判的措施。通过实证研究对模型进行验证，将理论与仿真结果应用于医疗决策分析中。本项目研究期间取得如下成果：已发表论文10篇；其中期刊论文8篇，会议论文2篇，国际刊物论文5篇、国内核心刊物3篇，SCI收录3篇、EI收录6篇；在投论文2篇，Temporal case matching with information value maximization for predicting physiological states投稿Information Sciences期刊，目前处于二审阶段，“随机干扰下基于证据链推理的多属性诊断决策”投稿“系统工程学报”期刊，目前处于一审阶段；获得实用新型专利授权1项、软件著作权2项；待出版中文学术专著1部，书名为《医疗与健康决策支持》，已确定电子工业出版社出版；协助培养博士毕业生1人。本项目对已有研究成果中的科学问题的进一步提炼与完善，并将理论成果与解决措施应用于申请人正在参与的提高医疗决策系统品质与效率的理论研究与临床实证研究中，为医疗决策系统推理机制准确性相关研究提供新思路、新方法，丰富该领域研究的理论体系。