面向大规模医疗数据的动态模糊最佳模式挖掘研究

With the construction of modern health information system, healthcare organizations are experiencing an explosive growth in medical data. Large-scale medical data often contain a huge of hidden but potentially valuable information, researchers face considerable challenges in deriving valuable information in the overwhelming growth rates of medical data. Advanced pattern mining is an effective way to extract and analyze hidden valuable information. ..However, due to the special characteristics of medical data that including high dynamics of change, high-dimensional, and fuzziness, which makes advanced pattern mining faces three major research issues in terms of combinatorial explosion in pattern mining, huge time and space complexity, and imbalanced mining accuracy and efficiency. The objective of this project is to build dynamic fuzzy optimal pattern mining model to extract useful information and discovery valuable structures and relationships from large but low-quality data sets. ..Firstly, In order to extract core information and ensure data integrity to maintain original information, this project will propose a large-scale dynamic feature extraction model for fuzzy attribute extraction and fuzzy dimensionality reduction. In addition, with the aim of improving the accuracy of fuzzy weights for research objects, we will reduce fuzzy error and inconsistencies of fuzzy transmission error by establishing fuzzy membership function and probability density function. Meanwhile, novel advanced measurement framework for dynamic fuzzy pattern mining will be developed to meet the requirements of stability and scalability, which is capable to circumvent the misleading and deceptive problems in traditional measurement framework. Furthermore, on the basis of above study, we will create a new dynamic fuzzy optimal pattern mining model that has the advantages of high accuracy, ability to dynamically adjusting based on time-varying, and suitable for dynamic fuzzy optimal pattern mining structure and new advanced pattern mining properties. Finally, this project will implement the related algorithms and then model analysis and assessment of the performance of proposed algorithms. Overall, based on above analysis, this project can provide theoretical bases for mining hidden but valuable information in large-scale medical data.

现代医疗卫生体系的信息化建设导致医疗数据呈爆发式增长，大规模医疗数据中往往蕴藏着巨大的潜在有价值信息，高级模式挖掘是提取和分析隐藏的有价值信息的有效途径。但是，由于医疗数据存在高动态变化性、高维性和模糊性，使挖掘面临模式挖掘的组合爆炸、巨大的时空开销及平衡挖掘的精度和效率三大问题。本项目以获取隐藏的有价值信息为目标，在模糊属性提取及模糊属性规约的基础上，通过建立模糊表达-模糊加权-模糊推理模型减少误差及误差不一致性传递，以提高研究对象的模糊加权准确度；同时，设计新的适合高级模糊模式挖掘的度量框架，对挖掘的误导性和欺骗性问题进行规避，以满足挖掘稳定性和可扩展性要求。最后，在此基础上，设计一种新的准确度高、并可随时间变化动态调节的，适合动态模糊最佳模式挖掘结构和新的高级模式挖掘性质的动态模糊最佳模式挖掘模型，进行算法实现和对算法性能建模分析与评估，为大规模医疗数据有价值信息的挖掘提供理论依据。

现代医疗卫生体系的信息化建设导致医疗数据呈爆发式增长，大规模医疗基础数据往往具有多源海量、异构动态、异质庞杂、多维分散的特征。高效地挖掘潜在有意义的医疗信息、智慧医疗数据的认知和发现、多维度数据的协同预测以及缺失数据的信息提取和挖掘等研究对医疗大数据的智能化分析、有效的监管和预测具有重要的作用和意义。..本项目研究的主要内容包含三个方面：①面向大规模医疗数据集的基础数据提取、清洗与整合。项目通过数据抽取、转换、融合来对复杂动态、众多病型、以及多维特征的数据进行处理分析。②对纳入动态模糊模式相关的疾病因素、项目（集）、模糊权重进行方案与相互关系分析。通过特征提取模型和模糊处理模型分析数据的关键核心信息。③构建动态模糊最佳模式挖掘模型，个性化的挖掘提取影响疾病治疗、防控和资源使用的优化组合模式，从而探索疾病服务效能的优化合理配置。..项目取得的重要研究成果以及关键数据主要包含五个方面：①项目结合犹豫模糊集和模糊粗糙集，提出了新的能够适应事务传递变化的犹豫模糊粗糙集，并提出了应用于医疗诊断的犹豫模糊粗糙近邻算法。②在充分研究了处理不完备数据集的方法后，增加不完备数据的缺失性分析以便提高大规模数据集的模糊有效表达。③提出了一种基于模糊选项关系的关键属性提取方法，该方法可以从模糊数据中提取关键属性，并且迅速定位拥有异常关键属性的患者并对其优先处理。④提出了模糊模式结构（核心项和相应的牵引项），并且提出了模糊支持度以及基于模糊支持度的剪枝策略来分析和挖掘隐藏在项目集当中的有用信息。⑤项目采集分析了成都市、南京市、以及法国里昂的涉及到急诊、内外科等多个部门的医疗数据，数据的样本量大且来源多角度化，从而使得提出的算法模型更加地贴近真实值和准确值。..综上，项目在研究内容、研究成果、人才培养以及国际合作交流方面均完成了任务书中的内容。项目取得了良好的研究成果并对大规模医疗数据的有价值信息的挖掘、不确定性信息的有效表达、以及医疗资源的合理配置有着一定的实践应用意义。