基于异构数据融合的智能医疗临床决策证据推理研究

Uncertainty reasoning has been playing key role in clinical decision-making of intelligent medical system. While, under the environments of internet，the number of input multi-source heterogeneous data has been proliferating, which influences the reasoning efficiency and accuracy of clinical decision-making. Aiming at optimization of D-S reasoning on base of multi-source heterogeneous data, in clinical decision-making, the singular value decomposition (SVD) in matrix theory is implemented to construct the threshold vector of fusion algorithm. The vector is employed to be the constraint of fusion reasoning, by which the boundary of the fusion reasoning problem is determined. Based on heterogeneous data，the joint correlations and their dissimilarity between the input variables and the disturbances will be induced, and the random interference reasoning model is to be built to analyze ergonomic parameters for clinical decision-making. Collaborative filtering algorithm is adopted to higher matching degree, so that multi-source heterogeneous data can be fused via multilevel iteration, and under the constraints of rules on clinical decision-making, feature supplements are achieved. The mechanism of association among heterogeneity evidences is studied. Parallel mechanism of reasoning with good interpretability is provided, in case of multi-criteria features of clinical decision-making. According to clinical path (CP) , the model set is built, which includes clinical decision-making inference based on the minimum risk rules, fusion reasoning mechanism on mis-labeled evidential chains, and mechanism on D-S reasoning based on temporal data-driven decision-making. In the lab that has been built with intelligent patient robots, empirical research on CP and model validation will be completed. Theoretical and simulated results will be applied to clinical decision-making practice.

不确定性推理始终是智能医疗系统临床决策的关键问题。而在互联网环境下，输入的证据数据量剧增，且更加呈现出多源、异构性特征，影响智能医疗系统临床决策的推理效率及准确度。为解决临床决策中，面向多源异构数据的证据推理优化问题，采用问题矩阵空间奇异值法，构造融合推理阈值向量，以确定融合推理解的边界；研究异构数据环境下，临床决策稳态与随机干扰的证据推理模型，分析致扰因子，进行决策系统抗干扰能力检验；采用协同过滤方法，解决匹配度问题，完成多源异构数据多级迭代融合，并在临床诊断规则的约束条件下实现特征补充；研究异构性证据关联机理，在临床决策多准则特征下提供可解释性强的并行推理机制；基于临床路径，构建包含最小风险准则的临床决策推理、类别误标下的证据链融合推理和时态数据证据推理的模型集。通过在已建立的智能病人机器人实验室完成临床路径实证及模型集验证，将理论与仿真结果应用于临床决策分析中。

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，不确定性推理始终是智能医疗系统临床决策的关键问题。而在互联网环境下，输入的证据数据量剧增，且更加呈现出多源、异构性特征，影响智能医疗系统临床决策的推理效率及准确度。本项目为解决临床决策中，面向多源异构数据的证据推理优化问题，采用问题矩阵空间奇异值法，构造融合推理阈值向量，以确定融合推理解的边界；研究异构数据环境下，临床决策稳态与随机干扰的证据推理模型，分析致扰因子，进行决策系统抗干扰能力检验；采用协同过滤方法，解决匹配度问题，完成多源异构数据多级迭代融合，并在临床诊断规则的约束条件下实现特征补充；研究异构性证据关联机理，在临床决策多准则特征下提供可解释性强的并行推理机制；基于临床路径，构建包含最小风险准则的临床决策推理、类别误标下的证据链融合推理和时态数据证据推理的模型集。通过在已建立的智能病人机器人实验室完成临床路径实证及模型集验证，将理论与仿真结果应用于临床决策分析中。本项目研究期间取得如下成果：已发表论文14篇：期刊论文13篇（国际期刊论文10篇、国内核心刊物3篇），国际会议论文1篇；国际期刊论文10篇：SCI/SCIE收录10篇、EI收录4篇、SSCI收录3篇；国内核心刊物3篇：CSSCI收录3篇、基金委A类1篇，基金B类2篇；国际会议论文1篇：EI检索会议论文1篇。已录用国际期刊论文2篇（SCI收录2篇）。出版中文学术专著1部，书名为《医疗大数据》，已由机械工业出版社出版。协助培养博士毕业生3人，硕士毕业生2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，为医疗决策系统推理机制准确性相关研究提供了新思路、新方法，丰富了该领域研究的理论体系。